Assessing On-Farm Water-Use Efficiency: A New ...

Assessing On-Farm Water-Use Efficiency: A New Approach Methodology and Six Case Studies

A report on collaborative research undertaken by: The International Center for Agricultural Research in the Dry Areas (ICARDA) and The United Nations Economic and Social Commission for Western Asia (ESCWA)

Kamil Shideed Theib Y. Oweis Mohammed Gabr Mohammad Osman

ICARDA International Center for Agricultural Research in the Dry Areas

©2005 International Center for Agricultural Research in the Dry Areas (ICARDA) All rights reserved. ICARDA encourages fair use of this material. Proper citation is requested

The Authors Kamil Shideed and Theib Oweis are Senior Agricultural and Natural Resource Economist and Senior Water Management Scientist, respectively, at ICARDA, Aleppo, Syria; Mohammed Gabr is a fbrmer Senior Economist at ESCWA; Mohanrmad Osman is a Senior Economist at ESCWA. Recommended Citation Shideed, Kamel; Theib Oweis, Mohammed Gabr, and Mohammad Osman. 2005. Assessing On-Farm Water-Use Efficiency: A New Approach. ICARDA, Aleppo, Syria, 86 pp.

ISBN: 92-9127-163-X Key Words: Water-use efficiency, water productivity, on-farm water allocation, variable input model, fixed-allocatable input model, supplemental irrigation

Cover: Relationship between water-use efficiency and yield for wheat grain in a Mediterranean environment. Source: Zhang and Oweis (1999).

ICARDA P.O. Box 5466, Aleppo, Syria Tel.: (+963) (21) 2213433,2213477,225112,2225012 Fax.: (+963) (21) 2213490,2225105,5744622 E-mail: icarda(ujcgiar.org Web Site: http:i/www.icarda.org, http:l/www.icarda.org/arabic

Foreword The increasing scarcity of water for agricultural production around the world is a major cause for concern. The situation is particularly worrying in West Asia and North Africa (WANA) where the per capita share of water in many countries has fallen below the water poverty level. With the rapid growth of the population and the consequent rise in demand for water, water shortages will be an even greater concern in coming years. Currently, agriculture consumes over 75% of the water in the dry areas but this share is projected to drop drastically and may reach 50% in some countries of WANA over the coming 25 years. The food security of the poorer countries that depend largely on agriculture will be particularly threatened. This calls for immediate actions to mitigate the effects of water scarcity. In spite of its evident scarcity, studies have revealed that the available water is not used efticiently for agricultural production and that large volumes of it are lost every year. ICARDA is, therefore, working with national and international research systems to develop methods of improving water-use efficiency and productivity in agriculture. The first step in these efforts is to assess the existing situation at the farm level and identify the areas where water is not used efticiently. In collaboration with the United Nations Economic and Social Commission for West Asia (ESCWA) and partners in the national research programs of Egypt, Iraq, Jordan and Syria, ICARDA conducted research to develop a practical methodology to evaluate how eff~cientlyfarmers in the dry areas use water in various agricultural systems, and tested the methodology in various agroecologies in the dry areas. This report summarizes the research results and includes six case studies used to test the developed methodology. We hope that it will contribute to improving water-use efficiency at the farm level, avoiding the loss of water in agriculture, and, therefore, mitigating the effects of water scarcity in the dry areas.

Prof. Dr Adel El-Beltagy Director General

Acknowledgments The authors would like to thank the NARS of Egypt, Iraq, Jordan and Syria for their active participation in the study. Special thanks are extended to the IPA Agricultural Research Center of Iraq (previous affiliation of the senior author) for developing the methodology and conducting the analyses for the six case studies. Thanks are also due to ICARDA staff, Drs Ahmed Mazied and Pierre Hayek, for collecting data in Syria. Gratitude is also expressed to the lraqi national team, particularly Mr Mahdi Sahar Khaidan (a PhD candidate at the College of Agriculture, University of Baghdad), for collecting field data for the more recent case study of Iraq. We would like to thank the staff of ICARDA's Communication, Documentation and Information Services Unit for their support with editing, typesetting, design and printing this publication. This work was co-funded by the United Nations Economic and Social Commission for Western Asia.

Contents Foreword Acknowledgements Contents Executive Summary Introduction Water-Use Efficiency - Water-use efficiency and productivity - Supplemental imgation and water productivity - On-farm water-use efficiency Methodology Development - The models of water use - Variable input model - Fixed-allocatable input model - Satisficing model - Model validation - Previous research on water allocation On-Farm Water-Use Efficiency in Radwania, Syria - Characteristics of Radwania farms - Model estimation and validation - Empirical results - Water-use efficiency On-Farm Water-Use Efficiency in Rabea, Iraq - Characteristics of Rabea farms - Empirical results - Water-use efficiency On-Farm Water-Use Efficiency in Al Ghor, Jordan - Characteristics of Al Ghor farms - Model estimation and empirical results - Water allocation On-Farm Water-Use Efiiciency in Nubaria, Egypt - Characteristics of Nubaria farms - Empirical results

iii iv v vii

-

Results for winter cropping

- Results for summer cropping -

Water-use efficiency

On-Farm Water-Use Efficiency in Beni Sweif, Egypt - Characteristics of the Beni Sweif farms - Empirical results - Winter cropping - Summer cropping - Water-use efficiency On-Farm Water-Use Efficiency in Wheat Production, Iraq - Data and wheat yield of the sample farms - Empirical results - Impact of supplemental irrigation on wheat yield and water productivity - Model estimation and empirical results - On-farm water-use efficiency Conclusions and Recommendations References Appendices

Executive Summary The dry areas o f West Asia and North Africa ( W A N A ) Pace severe water scarcity due to the limited opportunities for the exploitation o f new sources of water coupled with the rapidly growing demand for water resources.As a result, watcr resource management i s becoming one of the most important economic and social issues o f this century, especially in WANA. Policy makers at all levels and research institutions are increasingly focusing on issues ofwatcr quality and allocation, the increasing demand, changing technologies. water-use efficiency and economic feasibility. Available information indicates that the scarce watcr resources of the region arc inef'ficicntly used, especially for irrigation. Given that irrigation accounts for 80-90 per cent o f all water consumed in the WANA region. improving on-farm water-use efficiency can contribute directly to increased supply of water for other end users. L o w irrigation efficiency is associated with poor timing and lack of uniformity o f water applications, leaving parts o f the field over- or under-irrigated relative to crop needs. Moreover, operators o f irrigation systems do not have an incentive to supply farmers with a timely and reliable delivery of water that would be optimal for on-firm watcr efficiency Farmers, on their part, generally tend to over-irrigate as a result of their perceptions of water requirements and their expectations o f rainfall and market prices. Most studies i n the region on water-use efficiency are based on experimental trials for mono-cropping systems which do not precisely reflect the c o m p l e production x decisions at the farm level under different environmental, technological, and economic conditions. More recently, empirical studies on economic assessment of on-farm water-use efficiency in agriculture were jointly conducted by the International Center for Agricultural Research in the Dry Areas (ICARDA) and the United Nations Economic and Social Commission for Western Asia (ESCWA). These studies clearly demonstrate the low ratios of water-use efficiency in crop production implying the tendency of farmers to over-irrigate their crops. The methodology adopted hy these studies in assessing the status o f on-farm water-use efficiency has proved to he a valid approach for conducting further empirical studies. Engineering and agronomy have contributed the majority o f literature on water-use efficiency. Efficiency i s associated with a transformation o f an input into an output. The engineering perspective focuses on the concept o f irrigation efficiency, defined as the amount o f water from the main water source which can be effectively supplied to the root zone, while the agronomic perspective focuses on the concept

vii

of crop water-use efficiency, defined as the fraction of water transpired by the crop to that stored in the root zone. This concept includes both crop water-use efficiency and crop-water productivity. A combination of the engineering and agronomic perspectives resulted into the concepts of water-use efficiency, defined as the ratio of transpiration (mm) to total water supply (mm); and water productivity, defined as the ratio of crop production (kg) to the unit of water used These technical measures of efficiency are, however, not sufficient to assess the economic use of water which depends on the relative prices of water and other inputs, the marginal products and prices of the inputs, and the amounts of other inputs, including rainfall. To address this complex situation at farm level, the concept of on-farm water-use efficiency (FWUE) was developed and is defined for the purpose of this analysis as the ratio of the required amount of water for a target production level to the actual amount of water used. The resulting indicators of FWUE are very useful in guiding policies toward improving irrigation efficiency. On-farm water-use efficiency was assessed using six empirical studies jointly conducted by ICARDA and ESCWA. Five case studies were conducted in the Syrian Arab Republic and Iraq in 1999, and Jordan and Egypt in 2000. To further support the methodology another case study was conducted in Iraq using farm survey data from a supplemental irrigation project, collected from a cross-sectional survey of 284 farms in Ninavah province in Northern Iraq during the 2001-2002 season. The overall objective of this research was to assess the FWUE of crop production under specific farm conditions in some WANA countries. The methodology adopted by these studies has proved to he a valid approach for conducting further empirical studies and the estimated indicators of FWUE will provide useful options for more efficient use of water. To assess the efficiency of on-farm water use, three specified models, namely, the fixed-allocatable input model, variable input model and the behavioral model, were estimated using the ordinary least squares procedure. To validate the estimated models, three sets of out-of-sample forecasts were made. For an out-of-sample prediction, the observations were randomly divided into two subsets, one with 80% and another with 20% of the observations. The 80% subset was used to estimate each model's parameters, which were applied to the 20% to make out-of-sample predictions and to apply the prediction performance measures. To judge the performance of alternative models and thus provide evidence on model choice, three measures of prediction accuracy were applied: the measures of mean absolute error (MAE), root mean square error (RMSE), and mean absolute percentage error (MAPE). Based on the results of the estimated coefficients and

two of the comparisons of the prediction performance measures, it was concluded that the fixed-allocatable input model is better at estimating and explaining short-run water use. The selected models were then used to determine the amount of water required by each crop in the six study areas by calculating the amount of water required for each crop at the mean levels of independent variables appearing in each crop's equation. Implications of the results of the estimated coefficients and calculated indices of FWUE are summarized below. When the amount of water required for the crops was compared with actual amount used, it was found that there was over-irrigation for all crops and in all the study areas. FWUE for wheat, for example, was found to be 0.61 in Radwania (Syria): 0.37 in Rabea (Iraq), 0.65 in Nubaria and Beni Sweif (Egypt), 0.30 in A1 Ghor (Jordan), and 0.77 in Ninavah (Iraq). These estimates indicate that farmers over-imgated wheat by 20-60%. It is, therefore, possible to save an enormous amount of water which can be used to expand the wheat growing area, and thus increase total production, or to produce other crops. Alternatively. farmers can increase the wheat yield considerably under current levels of water use, and with improved water and crop management practices. Either option can contribute greatly to food security in WANA.

Figure I . Wheat on-farm water-use efficiency (%) in selected areas in WANA.

Estimates of FWUE under full irrigation provide important information on the efficiency of water use in producing competing crops. Cotton FWUE was estimated at 0.75 in Radwania and Beni Sweif', rellecting relatively high water-use efficiency compared to other crops produced in these two areas. However, cotton producers exceeded crop requirements by nearly 25%, an amount that could be saved if farmers were provided with extension recommendations to rationalize the use of scarce water. Likewise, FWUE of two forage crops, bersem and corn, produced under full irrigation in Egypt was estimated at 0.72-0.76. These estimates are higher than those of competing crops (0.55 for faba bean and 0.64 for sunflower) produced under similar conditions in Egypt.

Figure 2. On-farm water-use effciency of competing crops in Beni Sweif and Nubaria, Egypt. FWUE of vegetable crops varied with crop and area of production. Tomato FWUE was estimated at 0.68 in Rabea, 0.53 in Al Ghor, 0.56 in Beni Sweif, and 0.69 in Nubaria. The FWUE of watermelon was estimated at 0.76 and 0.44 in Nubaria and A1 Ghor, respectively. Similarly, the estimates of pepper FWUF were 0.74 and 0.53 in the same two areas, respectively. FWUE for cucumber and eggplant was 0.56 and 0.66, respectively, in Al Ghor. The estimates of FWUE for cereal, industrial, and vegetable crops mentioned above indicate a wide technological gap between the required practices and actual water application. Therefore, improving water-use efficiency for these crops can contribute greatly to the overall water-use efficiency in the study areas, and offers a high potential for saving water. These results are consistent with the findings of a recent FAO study which concluded that water productivity seems lo be lowest in waler-scarce regions of agriculture-based economies (Bazza and Ahmed, FAO, 2002).

Figure 3. Tomato on-farm water-use efficiency in selected areas in WANA. Producers perceive water as a fixed input in the short run, but allocatable among competing crops on the fann. This conclusion is supported by the fact that the coefficient of the water constraint variable is positive in the water-use equations of the crops. The estimates of the individual coefficients of the water constraint suggest that an increase in water availability is allocated most heavily to crops with relatively higher requirements, like cotton, tomato, potato, sugar beet, and bersem, rather than to crops with relatively low water requirements, such as wheat and barley. Output prices and planted areas appear to be strong determinants of water allocation in the short run among competing crops. Moreover, the water price variable is not negative in the water-demand equations for most of the crops in the variable input model. This implies that after planting crops, producers do not respond to water prices in making subsequent short-run decisions. Since water prices in the study areas were highly subsidized, they did not have a major quantitative impact on water allocation. Land allocation, crop choice, irrigation technology and output prices are the main determinants of multi-crop water-use decisions. Previous studies show that water demand is inelastic at low price changes (Fraiture and Perry, 2002). Because water prices are very low in WANA, only high increases in water charges can reduce the amount of water used for irrigation, which in turn will greatly reduce farmer income. Farm yield data from a survey of 284 farms in Iraq provided the following information:

Supplemental irrigation increased land and water productivity in wheat systems. Yields in bread wheat increased by 100%, and between 58 and 81% in durum wheat varieties. Water productivity increased by 32 and 15% in bread wheat and durum wheat, respectively, with an average of 31% in all wheat varieties. Irrigation technology has a significant impact on the amount of water used in crop production. Center-pivot sprinkler technology reduced the amount of water used for wheat production in Iraq by 7.2% compared to the use of solidset sprinkler technology. The efficiency of water use varies among different segments of the farmers. Of all the wheat farmers studied in Iraq, 80% had FWUE of less than one, indicating that these farmers over-irrigated their wheat crop, and it was greater than one for 20% of the farmers, implying that these farmers under-irrigated their wheat crop by 10%. The overall water-use efficiency for the whole sample was 0.77, indicating that wheat producers over-irrigated their crop by 23%. Among over-irrigating farmers, the FWUE of 4% was estimated at 0.34, indicating that the actual amount of water used exceeded the required by about 66%. The water-use efficiency of 20% of the farmers was 0.64, implying that this group over-irrigated wheat by 36%. Furthermore, 56% of the farmers had water-use efficiency of 0.87 and over-irrigated wheat by 13%. These results indicate the need for the extension system to develop targeted recommendations for the different farmer groups. Farm size is an important factor in explaining the variation in FWUE among wheat producers. More than 50% of the farmers grew wheat on farms of less than 10 ha, with an average holding of 6.7 ha. The water-use efficiency of such small farms was 0.77, implying that small farmers over-irrigated wheat by 23%. The water-use efficiency of medium farms (10.1 - 20 ha) was 0.81. However, with the increase in farm size above 20 ha, the water-use efficiency decreased. The water-use efficiency of large farms was 0.72, indicating that these farms exceeded water requirements by 28%. Thus, small and medium farms were more efficient than large farms in using supplemental irrigation water in wheat production. This has important implications on the economies of size under supplemental irrigation. The results obtained for the three models used in this study have important policy implications. The overall water-use efficiency for the sample farms was 77%, indicating a potential to improve water-use efficiency by 23%. The results by farm size, on the other hand, show that small, medium, and large farms had

different potentials for improving their water-use efficiency by 23, 19 and 28%, respectively. However, each individual farm had a different potential to improve water-use efficiency ranging from a low of 13% to a high of 66%. If policy makers encourage the design of appropriate technical as well as incentive packages, water-use efficiency can be improved. By doing this, ample water will be available for productive use leading to increasing water productivity and consequently agricultural productivity. Improvements in water management, imgation, and technologies have the potential to optimize water use at the farm levels. Sound extension strategies and the provision of pertinent advice to farmers will be instrumental (a) in optimizing water use at farm levels, and (b) in reducing the adverse effects of salinization and water logging on the productivity of land which are caused by over-irrigation. Thus, by obtaining optimal water use it is possible to increase wheat productivity in the study area while ensuring the sustainable use of resources, both water and land. The methods of analysis used in this study were valid for assessing FWUE within the framework of multi-crop production systems. Results obtained compare favorably with expert and technical recommendations. However, these results are only applicable to the study areas and should not be generalized at national and regional levels. The main difficulty encountered in such studies is the calculation of actual water use, given the different sources of water and irrigation technologies used on the farms. It is highly recommended to conduct more empirical studies using similar methods in different agroecological zones of WANA to assess the current status and potential of FWUE for different crops to improve water and land productivity in the region and eventually produce more crops per drop.

Introduction Many parts of the world are facing increasing water scarcity. New sources of water are expensive to exploit, limiting any potential for substantial new water supplies. Water for agriculture is increasingly diverted to meet the needs of urban areas and industrial development, while water logging, saliniration, groundwater mining, and water pollution are putting pressure on land and water quality. The majority of the population of people in dry areas of the world are in West Asia and North Africa (WANA). The region is characterized by low rainfall and limited renewable water resources-about 1250 m' per capita. compared to the world an average of 15,000 m' for Europe, 20,000 for North average of over 7000 America and 23,000 m' for Latin America (World Resources Institute, 1999). In some WANA countries, such as Jordan and Tunisia, available water will barely meet basic human needs in the near future. The WANA region receives lower rainfall than seasonal crop water requirements; moreover, its distribution is rarely in a pattern that satisfies crop needs. Periods of severe moisture stress are very common and in most of the locations they coincide with the stages of growth that are most sensitive for crop growth. Soil moisture shortages at some stages cause very low yields. For instance, average rainfed wheat grain yields in WANA range between 0.6 and 1.5 ton/ha, depending on the amount and distribution of seasonal precipitation, much lower than potential yields of over 5 ton/ha. Rapid population growth and improving standards of living are increasing the demand for water in WANA. Presently, over 75% of the renewable watcr resources in the dry areas of the region are used for agriculture. Increasingly, competition for water among various sectors deprives agriculture of substantial amounts every year. It is projected that water for agriculture in WANA will drop from the current level of over 75% to about 50% by 2050. Furthermore, most of the hydrological systems are already stretched to the limit, yet more food production is required to feed the increasing population. Greater efforts will have to be made to improve on the efficiency with which water is used if current levels of agricultural production and environmental protection are to be maintained. Thus, water productivity is an urgent issue in the dry areas. High water productivity can be achieved through promoting water-use efficiency techniques, adopting efficient on-farm water management, selecting proper cropping patterns and cultural practices, and developing suitable crop varieties.

2

Assessing

On-Farm Water-Use Efficiency

Yields and water productivity can be substantially improved with the application of supplemental irrigation in the rainfed areas, the adoption of water harvesting in the steppe areas, and the use of improved irrigation systems and schedules in irrigated areas. Water policies in WANA have mainly focused on investment in irrigation, expansion of the irrigated area and construction of drainage networks (ESCWNFAO, 1994), without considering the associated rise in the water table and salinity. These policies have contributed to the depletion of land and water resources in many WANA countries. National water policies could encourage water savings in water-scarce areas by providing incentives and effectively enforcing penalties. When upstream managers cannot ensure conveyance efficiency, there may be no incentives for water users to make efficiency gains. Though water supply for irrigation has expanded the irrigated areas under cultivation and consequently increased agricultural production, lack of demand management practices has contributed to a low efficiency of water use and consequent waste. In addition, improvement in the availability of water due to the introduction of improved technology diverted attention from demand management and reduced emphasis on low-cost alternatives such as improving efficiency, conservation, and reduction of waste through maintenance of irrigation infrastructure. In conventional irrigation, water is applied to maximize crop yield (maximizing production per unit of land). Operators of irrigation systems do not have an incentive to supply farmers with a timely and reliable delivery of water that would be optimal for on-farm water-use efficiency and use of other inputs (Serageldin, 1998). Therefore, farmers generally tend to over-irrigate as a result of their perceptions of water requirements, their expectations of rainfall and market conditions. ICARDA's research on water-use efficiency has shown that yields and water productivity are greatly enhanced by the conjunctive use of rainfall and limited irrigation water. Substantial increases in crop yield were observed in response to the application of relatively small amounts of supplemental irrigation. For wheat growers, supplemental irrigation also stabilized wheat production from one year to another. The coefficient of variation was reduced from 100 to 20% in rainfed fields that adopted supplemental irrigation (Oweis, 2001). ICARDA's long-term research in Syria has shown that applying only 50% of full supplemental irrigation requirements (over that of rainfall) would cause a reduction in yield of 10-15%. This finding, in light of the increasing water scarcity in Syria,

Water-Use Efficency

3

encouraged ICARDA and the extension system to test a deficit supplemental irrigation strategy at farmers' fields. Results show that with deficit irrigation the yields increase by over 50% compared with full irrigation, which demonstrates the possibility of producing more food with less water. Information on on-farm water-use efficiency in WANA is limited (Oweis and Hacum, 2003). Most of the evidence available in the region is mainly based on experimental trials for mono-cropping systems. Thus, it does not precisely reflect the complex production decisions at the farm level under different environmental, technological, and economic conditions. Recently, six empirical studies on economic assessment of on-farm water-use efficiency in agriculture were conducted. These studies clearly demonstrate the low ratios of water-use efficiency in crop production, implying the tendency of farmers to over-irrigate their crops (ESCWAIICARDA, 2000,2001,2003). The main objective of this work was to empirically quantify the status of water-use efficiency under farmers' conditions using a tested methodology for the assessment of on-farm water-use efficiency in the WANA countries. Specifically, the studies were aimed at developing a methodology for the assessment of on-farm water-use efficiency within the framework of multi-crop production systems. The approach was tested in selected areas o f Egypt, Iraq, Jordan and Syria using farm survey data. Some of the concepts used here and the case studies were reported earlier in ESCWA-ICARDA joint documents (ESCWA/ICARDA, 2000,2001, and 2003).

4

Assessing On-Form Water-Use Efficiency

Water-Use Efficiency Water-Use Efficiency and Productivity The concept of water-use efficiency (WUE), commonly used to evaluate the performance of an irrigation system, is defined as the ratio of the amount of water used for an intended purpose to the total amount of water input within a spatial domain of interest. In this context, the amount of water applied to a domain of interest but not used for the intended purpose is a loss. To increase the efficiency of a domain of interest, i t is important to identify and minimize losses. Depending on the intended purpose and the domain of interest, there are many efficiency concepts, such as crop water-use efficiency and water-application efficiency, among others (Guerra et al., 1998). Improving irrigation efficiency is a slow but difficult process that depends essentially on the local water scarcity situation. It may be expensive and requires willingness, know-how and action at various levels. Efficient use of irrigation has been studied for three types of systems: the trickle, solid-set sprinkler, and furrow irrigation systems. It was found that imgation efficiency of the sprinkler system was on the average about 22% more than that of the furrow system and about 21% less than that of the trickle system. Overall efficiency of the trickle system, however, was on average about 28 and 45% more than that of the sprinkler and furrow systems, respectively (Dawood and Hamad, 1985). The WUE concept is not directly related to the amount of food that can be produced with an amount of available water. The optimum level of applied water for a particular situation is that which produces the maximum profit or crop yield per unit of land or per unit of water, depending on the underlying objective function and the limiting factor. In this respect, water productivity, defined as the amount of food produced per unit volume of water used, is more relevant. Because the water used may have various components-evaporation, transpiration, gross inflow. net inflow, and others-it is essential to specify which components are included when calculating water productivity. The concept of water productivity, like WUE, requires clear specification of the domain of interest. Water productivity can be improved by increasing yield per unit of the land area using a better crop variety, improved agronomic practices, or by growing the crop during the most suitable period. Thus, water productivity can be achieved by factors other than water management. Higher productivity does not necessarily mean that the crop effectively uses a higher proportion of the water input. For this reason, water productivity alone would not be particularly useful in identifying water-saving opportunities of the system under consideration.

Water-Use Efficency

5

The terms WUE and water productivity should be used complementarily to assess the impact of water management strategies and practices used to produce more crops with less water. However, both terms are scale-sensitive; therefore, failure to clearly define the boundaries of the spatial domain of interest can lead to erroneous conclusions. It is also vital to specify the water-use components that are taken into account when deriving water-use efficiency and productivity. Furthermore, the two concepts of efticiency and productivity are related but different. Due to the scarcity of water, there is a need to evaluate the efficiency with which it i s utilized to arrive at an appropriate irrigation system. WUE and productivity would differ according to different systems of irrigation, crop mix and environment, and are comprised of different dimensions: crop consumptive use (water requirement), an efficient crop mix (maximum irrigable area for given water resources) and maximum output and value per unit of water. Measurements of efficiency or loss are site-specific not only because of variation in physical environment, but also because of variation in the physical infrastructure and management capacity reflected at each location. For measuring water productivity, while the denominator remains the quantity of water diverted or depleted for a particular use, such as crop production, the numerator is measured as the crop output. The numerator and the denominator can be expressed in either physical or monetary terms. Given this measure, there are several different ways of expressing water productivity: Pure physical productivity, defined as the quantity of the product divided by the quantity of the water diverted or depleted; Combined physical and economic productivity, defined in terms of the economic value expressed as gross or net value, or net present value divided by the amount of water diverted or depleted; Economic productivity, defined as the net present value of the product divided by the net present value of the amount of water diverted or depleted (defined in terms of its value or opportunity cost, in the highest alternative use). In this context, many researchers frequently use the term water productivity as the ratio of the physical yield of a crop and the amount of water consumed, including both rainfall and supplemental irrigation. Yield is expressed as a mass (kg or ton), and the amount of water as a volume Water productivity has been evaluated in terms of crop output and value per unit of water. A protective irrigation system was found to perform better in terms of social efficiency, and a perennial system was better in terms of situational efficiency. In India. the yield rates of rice, for example, in the irrigation system were higher than those under the perennial system, although the water requirement was lower. The

6

Assessing On-Farm Wafer-Use Efficiency

average water productivity of rice in physical and monetary terms was 0.25 kg/m3 and 0.265 Rs/m3 under the protective system of irrigation. All other crops had higher water productivity. In the perennial system, the water productivity of rice was lower, but the crop was widely grown during the rainy season because of the agro-climate (Giriappa, 1984).

Supplemental Irrigation and Water Productivity Supplemental irrigation (SI) can be defined as the addition of small amounts of water to rainfed crops when rainfall fails to provide sufficient moisture for normal plant growth, in order to improve and stabilize yields. The cost of water is an important factor in the economics of SI. In most WANA countries, water from public (surface) irrigation schemes is provided almost free to users; and groundwater costs do not reflect their real value, because the energy required for pumping is highly subsidized. As a result, most farmers tend to over-irrigate. ICARDA studies have shown that the SI amounts for wheat reported by farmers is up to three times the optimal rate from research trials. It is common to see sprinklers operating on wheat in December, January and February, when the probability of rain is high, even though the crop water requirement in these months is low and the crop is not very sensitive to water stress. Enhanced exploitation of groundwater for SI on vast areas, which traditionally used to be rainfed, has helped bridge the gap, for example, in the Syrian Arab Republic's basic food production, recovering in particular the wheat balance. However, it has led to over-pumping and excessive water use. Results show that improving the wheat price encourages the use of more water, unless the rate of increase in the cost of water exceeds that of wheat. Optimal applications of SI are determined by both the input/output price ratio and weather conditions. Research by ICARDA scientists also shows that in the Syrian Arab Republic, supplementing only 50% of the rainfed crop irrigation requirements reduces the grain yield by only 10.20% relative to full irrigation (FI). Using the saved 50% to irrigate an equal area gives a much greater return in the total production. In some areas, groundwater resources are being over-exploited for FI and their quality is deteriorating. With such pressure on the existing water resources, sustainable use can be obtained only by producing more crops from less water-improving water productivity. Comparing the water productivity of SI in wheat with that of FI, a real opportunity for water-use improvement was found. According to ICARDA's research trials on farmers' demonstration fields in the Syrian Arab Republic, a cubic meter of water used in SI produced, on average, an extra 3kg of wheat over rainfed yield, whereas

a cubic meter used in FI produced about 0.5kg/m3. This large difference in the water productivity is attributed to the conjunctive use of rainfall and S1 water. In Jordan, water productivity in rainfed wheat in Mushagar (300mm annual rainfall) was 0.33kg/m3. When a cubic meter of rainfall was combined with SI of 0.5m3, the overall water productivity was increased to 3.5kg/m3. In view of this potential for more efficient use of water decision makers at the national levels should consider the feasibility of diverting some irrigation water from FI to SI or use both for optima1 crop-water allocation (Oweis and Salkini, 1992). The average water productivity of rain in producing wheat in the dry areas of West Asia and North Africa is about 0.35kg of although with good management and favorable rainfall (in amount and distribution) it can be increased However, water used in SI can be much more productive. to lkg of ICARDA research showed that a cubic meter of water applied at the right time (when the crop is suffering from moisture stress), combined with good management, could produce more than 2.5kg of grain over the rainfed production. This extremely high water productivity is mainly attributed to the effectiveness of a small amount of water in alleviating severe moisture stress during the most sensitive stage of crop growth and seed-filling. When SI water is applied before such conditions occur, the plant may reach its yielding potential (Oweis, 1997). In comparison to the productivity of water in fully irrigated areas (where the effect of rainfall is negligible), the productivity is higher with SI. In fully imgated areas with good management, wheat grain yield is about 6 tons/hectare using 800 mm of one-third of that under SI water. Thus, the water productivity is about with similar management. This suggests that water resources may be better allocated to SI when other physical and economic conditions are favorable.

On-Farm Water-Use Efficiency The term "efficiency" reflects the ratio of output to input. However, input and output components of efficiency differ from one field to another. Even in the same field. these components vary depending on the level and/or scope at which the term "efficiency" is being used. In irrigated agriculture, there are many efficiency terms, each of which has a specific use. These terms include: Water-Conveyance Efficiency (WCE), which reflects losses of water from the conveyance system; Water-Application Efficiency (WAE), which reflects losses of water below the crop root zone; Water-Distribution Efficiency (WDE), which reflects how uniformly water is applied to the field; Water-Storage Efficiency (WSE), which reflects the adequacy of water stored in the crop root zone;

8

Assessing On-Farm Water-Use Efficiency

Irrigation Efficiency (IE), which reflects the overall losses in irrigation; and Water-Use Efficiency (WUE), which reflects how well water is used in producing crops. The common factor or resource that ties these terms is water (Oweis and Hachum, 2002). WUE has been defined in various ways by hydrologists, physiologists and agronomists, depending on aspects that one wishes to emphasize. For instance, Viets (I 962) used WUE to characterize the ratio of crop yield to crop water use. WUE, in general, refers to the amount of plant material produced per unit of water used. WUE is not a physical engineering efficiency, which is usually dimensionless and has a maximum value of 100%. The efficiency term for this biological ratio has caused some concern and confusion with irrigation and water resources efficiency terms. Although efficiency is a measure of output per unit input, we rarely consider all the outputs or all the inputs involved in crop production; therefore, the term should be used cautiously. Plant growth and yield are more strongly related to transpiration than to evapotranspiration. However, WUE is sometimes evaluated per unit of water transpired (WUET), or per unit of irrigation water applied (WUEI). Often, WUE is based on evapotranspiration (WUEET). The difference is important since suppression of soil evaporation and prevention of weed transpiration can improve the WUEET; however, it need not improve the WUET, which is a measure of crop performance. I-WUEET is a measure of the increase in the crop production (biomass or marketable component) relative to the increase in water consumed when irrigated, over the consumption under non-irrigated conditions as follows, (Burman et al., 1981):

Where:

Y, = mass of marketable crop produced with irrigation Y,, = mass of marketable crop produced without irrigation = mass of water used in ET by the irrigated crop Et, = mass of water used in ET by the non-irrigated crop To further explain WUE terms, Figure 4 depicts, under normal field conditions, the relation between crop yield (Y) and each of transpiration (T) evapotranspiration (ET) and irrigation water (1). From the diagram, the transpiration WUE (WUET), the evapotranspiration WUE (WUEET) and the irrigation WUE (WUEI) are defined as: WUET = Y/T (2)

WURET = Y/(T+E:) (3) WUEI Y/(T+E+(I-r1 r2)(R+D)) (4) Where: R= surface runoff losses D= deep percolation (drainage) losses r l = Sraction of (R+D) that is recoverable with water quality the same as that of the irrigation water, I r2 = reduction in the water productivity potential due to deterioration in water quality; depends on water quality of 1 and rl(R+D), (rl I) It is evident that: WUEI < WUEET< WUET. However, if E=0, then WUEET = WUET. Also if the runoff(R) and drainage (D) losses are equal to zero, then WUEI = WUEET, and so on. The concept of WUE has many facets, but the most relevant are physiological, agronomic, hydrological and socioeconomic. The current definition of WUE is not necessarily accepted by scholars in all of these fields, this is why Hook and Goscho (1988) have proposed the concept of resource-use efficiency (RUE). RUE considers all the components of production inputs including land, water, climate, nitrogen, and cropping systems. When one resource is limiting yield, other non-limiting resources are usually used less efficiently. When that which limits yield is alleviated, all resources will be used more efficiently until another resource becomes limiting. Considering that water in the dry areas is often the most limiting resource, any improvement in water RUE will necessarily lead to enhanced use of other resources. The engineering and agronomy perspectives have dominated the literature on WUE. In the engineering perspective the concept of irrigation efficiency is detined as the amount of water from the main water source which can be effectively supplied to the root zone. This perspective distinguishes between three types of efliciency, namely; conveyance efficiency, farm efficiency and field efficiency (Schmidt, 2001). The agronomic perspective, however, illustrates the concept of crop WUE, detined as the fraction of water stored in the root zone that is transpired by the crop. This concept includes both crop WUE and crop water productivity. A combination of the agronomic and engineering perspectives results in WUE, defined as the ratio of transpiration (mm) to total water supply (mm), and water productivity, defined as the ratio of yield (kg) to total water supply (mm). These concepts reflect the technical measures of efficiency and thus are not sufficient to assess the economic level of water-use efficiency, as water is used in combination with a whole set of other inputs, such as land, fertilizers, labor,

10


machinery, and management, to produce crops. Therefore, singling out any one input such as water to determine efficiency may be misleading (ACIL Tasman, 2003). The economically efficient amount of water use depends on the relative prices of water and other inputs, the marginal products of the inputs, the prices of inputs and the amounts of other inputs, including rainfall. The concept of on-farm water-use efficiency (FWUE) was developed to address this complex situation at the farm level (ESCWA/ICARDA, 2000 and 2001). FWUE is defined as the ratio of the required amount of irrigation water to produce a specific output level to the actual amount of water applied by farmers. With this definition FWUE may take the value of less than, greater than or equal to one. If it is less than one, it implies that farmers over-irrigate their crops. While a value greater than one implies that farmers under-irrigate the crops. However, if the value of the calculated FWUE is equal to one, it means that farmers are fully efficient in using irrigation water because the required and applied amounts of water use are equal. X 100% FWUE = Where: is the amount of water required (m3) by the crop to produce certain level of crop production is the amount of water actually applied (m3) by the farmer to produce that level of crop production This definition combines several aspects from others mentioned earlier; however, its advantage is that the nominator and denominator are of the same units (volume or depth), so the efficiency can be expressed as a percentage. The efficiency definition also reflects the performance of the farmer in using water relative to the resources potential and farm conditions that exist in the area surveyed and not to another area with different farming conditions. The water required in the above equation is determined by the model from the data collected among the complete sample at the community. the canal or the district level. FWUE was assessed using six case studies in Egypt, Iraq, Jordan and Syria (ESCWA/ICARDA 2000,2001, and 2003). These case studies support the usefulness of this concept to assess the efficiency of water use under farmers' conditions and provide vital information for water savings. FWUE in Radwania (Syria), for instance, was found to be 0.61 for wheat, 0.45 for barley and 0.75 for cotton. The estimates indicate that farmers over-irrigate wheat, barley and cotton by 39, 55 and 25%, respectively. Other case studies provide similar information regarding the excessive use of irrigation water by crop producers.

Methodology Development

11

Methodology Development Predicting crop-level input allocation (use) is a major problem in a multi-crop production system largely due to limited availability of data on crop-level input use. The challenge, therefore, is to develop modeling approaches that permit prediction of input allocation from data on farm-input use and crop-level land use. These modeling approaches are needed to develop crop budgets and estimates of costs of production. Furthermore, to evaluate the effects of alternative policies on input use requires an understanding of how producers make decisions on crop-level input use (Moore et al., 1994). Previous research on multi-output input allocation was mainly based on two assumptions about producer behavior: profit maximization and satisficing. Satisficing behavior means that farmers operate on the basis of rules of thumb-basing entirely on their practical experience and not so much on scientific evidence. A simple form of this is that producers follow either a distributor's recommendation or other routine practices concerning a crop's input-application rate per hectare. Crop acreage would, therefore, effectively determine the allocation of an input among crops on a multi-crop farm.

The Models of Water Use Three alternative models of multi-crop input allocation are proposed for this study. These are the fixed-allocatable input model, the variable input model, and the satisficing model. An input considered to be variable in the long run may actually be fixed and allocatable in the short run. Irrigation with ground water is, for example, modeled as a variable input in the long run based on the assumption that it is subject to market forces with groundwater-pumping costs as water 'price'. Yet constraints on the number of wells, pump capacity, and water distribution infrastructure may make groundwater a fixed-allocatable input in the short run (Moore et al., 1994b). Irrigation with surface water may pose similar short- and long-run institutional constraints. Hired labor and farm machinery may also be variable in the long run, but fixed and allocatable in the short run. Crop-level input-use data are required to estimate the fixed-allocatable input model. Farm-level water use serves as an exogenous variable in this model, with crop-level water use serving as the endogenous variable. Unlike the variable input and satisficing models, a procedure does not appear to be available for predicting the results of the fixed-allocatable input model using deficient data because of the essential role of farm-level water as an exogenous variable. In contrast, farm-level water serves as the endogenous variable in the variable input and satisficing models estimated with deficient data. A data set that contains both crop-level irrigation water and acreage data from multi-crop farms is applied.

The three alternative models of short-run input use can bc directly estimated econometrically with crop-level water data. The availability of crop-level micro data on W U E makes the data 'non-deficient' in terms of information on water a llocation in a multi-crop system. In this study, the variable input and the fixedallocatable input models are derived based on the profit maximization assumption. The satisficing model is a simple model of bounded rationality. These three models of multi-crop water allocation can be compared using two techniques of model selection-model specification tests and prediction accuracy measures. The empirical application analyzes multi-crop water irrigation in Syria, Iraq, Jordan and Egypt using data from farm surveys. The basic unit for the analysis is the individual farm and the study targeted the whole cropping system. WUE was assessed for all crops planted in a given season and for all seasons over the year. Different farms were selected to cover the following major conditions in the region: Water sources including surface and ground water and rainfall. Cropping systems including field crops, orchards, vegetables and mixed systems. Water managemenl systems including rainfed, SI, and FI. Farm size and type including small, medium and large. Farmers involved in irrigated agriculture make a variety of decisions concerning crop-choice, land use, and irrigation water application. As an irrigator, the farmer also makes crop-level water decisions conditional on land allocations, which determine water use within an irrigation season (Moore et al., 1994b).

In this analysis, the farmer makes intermediate production decisions including the acreage and combination of crops. Subsequent short-run decisions include quantity of irrigation water to apply to each crop over the irrigation season. Thus, crop-specific acreages are exogenous to the water-use decisions. The common thread across the three alternative models, according to Moore et al. (1994a) is that crop-level land use serves as one of the determinants of crop-level water use in each model. To mathematically present the proposed models, the following notation is used: P = crop prices which are given to producers = price of crop i, (i= I ,.. . ,m) = water price r = variable input prices other than water (v= 1, ...,z) = water allocated to crop i W = farm-level quantity of water n, = land allocated to crop i


13

X = variables taken as given in the short run (e.g., crop-level irrigation technology and weather; s=1,. . .,t) (.) = the short-run restricted profit function of crop i (.) = the multi-output restricted profit function of the farm Input non-jointness is assumed so that the multi-crop profit function decomposes into the sum of distinct crop-specific profit functions. The profit functions are assumed to fit well in the conventional assumptions. Various functional forms can be used. However, flexible functional fonns are more appropriate for multi-output production decisions. This study applied the normalized quadratic profit function. which is a flexible functional form of the profit function and has been widely used in previous multi-output agricultural production research. The full specification of the quadratic profit function includes linear, squared, and cross product terms for all exogenous variables. Prices are expressed in relative terms, with one price serving as a numeraire; this maintains linear homogeneity of the function.

Variable input model The variable input model has commonly been used to analyze short-run irrigation water use (Moore et al., 1994a, 1994b; Chambers and Just, 1989; Just et al., 1983). Following the dual approach, application of Hoteling's lemma by taking the firstorder partial derivative of the restricted profit function with respect to the water price variable gives crop-level water-demand functions for the variable input model. These demand functions are as follows: m (I) The forms of these derived crop-level demand functions to be estimated are linear functions of the independent variables.

Fixed-allocatable input model The fixed-allocatable input model of water use represents a second approach based on a profit maximization assumption. This model is based on a short-run water constraint in the sense that the available amount of water is fixed at a given time and this amount should be allocated among competing crops at the farm level. For example, in groundwater irrigation, the groundwater wells, pump capacity, and irrigation capital during the growing season are fixed. This constraint does not reflect a long-run, institutionally-defined water quota. Thus, the fixed-allocatable input model offers a more reflective model of multi-crop decisions on the farm level than the variable input model.

14


In this model, producers operate with a short-run constraint on farm-level water use because of fixed groundwater pumping capacity. m

=W)

=

To obtain optimal short-run water allocation functions using the duality theory, the following profit-maximization problem needs to be solved. Applying the first-order condition for profit maximization by taking the first derivatives of the profit function gives the input demand functions. The necessary (first-order) conditions for solving the problem are: =

L

for i = 1, ...,m

Where, L is the shadow price on water constraint. Optimal water allocation functions can be obtained by solving this equation system. These water demand functions are:

i=

rn

(3)

The fixed-allocatable input model has two distinct features. First, water allocations to one crop depend on the output prices and acreage levels of all other crops. Thus, in contrast to the variable input model of equation (1), intercrop price and acreage variables supplement own-crop price and own-crop acreage as determinants of water use. Second, the farm-level water quantity constraint in equation 3 replaces water price as a determinant of short-run crop-level water use. Equation 3 is linear in the exogenous variables and it is the water demand function to be estimated for the fixed-allocatable input model. The optimal allocation in equation 3 illustrates the apparent jointness created by a fixed-allocatable input. Despite the assumption of input nonjointness, the fixed water input creates interdependence across crops. For example, consider a multi-crop farm that grows wheat, potato and lentil. The water use for wheat depends on acreage in potato and lentil in addition to acreage in wheat.

Satisficing model Under the satisficing model of short-run water use, crop-level land use virtually determines crop-level water use, with all price variables and the water constraint removed from the specification. Other variables (irrigation technology and weather) explain any additional variation in water use. The general form of this model is (Moore et al., 1994b):


To be consistent with previous research (e.g., Moore et al., 1994b) on the variable input and the fixed-allocatable input models, a linear specification is used to estimate equation 4. In intuitive terms, the satisficing model stems from the idea that longer-run decisions have a larger quantitative impact on profit than short-run decisions. Thus, producer behavior might conform more closely to the profit maximization assumption in the intermediate or long run. However, satisficing in the short run by following a rule of thumb or a distributor's recommendation may reduce information requirements with little sacrifice in profit. An alternative model may explain the producer decisions on water use in the short run. A 'behavioral' model relating water use primarily to planted area in the crop is an example (Just et al., 1990). According to this model, producers apply a fined water-land ratio in the short run. It describes variable input allocation in a region with a group of (i) producers (i = 1,2,.. ., I ) producing K crops ( k= 1, 2, . . .,K ) using water input W. The statistical analysis estimates the allocation of variable water input among crops. The two items of information used for these estimates are which is the area allocated by individual i to the production of crop k ; and which is the aggregate quantity of water input used by individual i. Thus,

is the unobserved quantity of water input allocated by individual i to Where, production of crop k. Lnformation on Wi is relatively easier to obtain at farm level than at crop level, since land-allocation data is more likely to exist than data on allocation of water among competing crops. With this model, producers are assumed to act as though their production functions have constant returns to scale. Hence, their decisions consist of the water/land ratios and land allocations (Just et al., 1990). This is based on the assumption that producers exchange information in assessing technologies and markets, and imitate one another. This allows water/land ratio decisions to be characterized by an overall average level and a systematic farmer deviation reflecting land quality, human ability, and perceptions. To develop the estimated form of this model, consider the following: = be the quantity of water per unit of land used by producer in Let can be disaggregated as producing crop k. The systematic element of follows: a, (6)

15

16


Where a, is an average regional use of water per unit of land in the production of crop k; Bi denotes deviations by farmer i from the regional average for use of water. Substitution of ( 6 ) into (5) gives:

(7) Where, ei is a random error term assumed to be normally distributed. Estimation of equation 7 requires regressing total use of water on the area allocated to each of the crops crossed with dummy variables corresponding to the crop effect and farmer effect. The sum of estimated parameters B,) is an estimate for the per unit area allocation of water to crop k by farmer i. Multiplication of this estimate by the land allocated to the crop results in the behavioral estimate of the allocation of water to crop k,

(a,

(8)

Equation 4 can be estimated using the ordinary least squares (OLS) procedure. In case of one-period cross-sectional data, this model can he estimated with no fanner differences.

Model Validation To validate the proposed models, model specification tests and prediction accuracy measures are used. Comparison of specification tests of the three models of short-run water use is done in pairs. Similarly, prediction performance measures are used to estimate the prediction accuracy of the estimated models and, thus, their validity. Potential prediction accuracy measures include mean absolute error (MAE), root mean square error (RMSE), and mean absolute percentage error (MAPE). Like the model specification tests, the measures of prediction accuracy are calculated using a farm-level approach. The calculated measures, thus, represent the accuracy of a model in predicting short-run water use for the set of m crops under consideration. Both in-sample and out-of-sample predictions are made to evaluate the alternative models. Mathematically, these measures can be presented as follows:

RMSE

Development

17

Where Y, is the observed value of the dependent variable for observation t, Y', is the predicted value of dependent variable for observation t, and T is the number of observations. As in the case of the model specification tests, the prediction measures are calculated using farm-level data and conducted crop by crop. The measures represent the accuracy of a model in predicting short-run water use for each of m crops under consideration. Both in-sample and out-of-sample predications need to be calculated for evaluating the alternative models of water use. Another method for model validation includes the plausibility of the estimated model. An example is the comparison of water-use recommendations by farm advisors in the region with the amount of water use calculated by a model (Just et al.,1990). The reconiniendations represent a range of water application rates that the extension agents consider to reflect sound agricultural practices in the region. Other measures used to evaluate the reliability of the estimated models include the log-likelihood function and the percentage of correct predictions (Caswell and Zilberman, 1985). To collect required data for methodology testing and validation, a questionnaire was developed and pre-tested. It covered various farm-level and crop-specific information including socioeconomic characteristics of producers, size of holdings, sources of household income, soil characteristics, cropping pattern, water availability and cost. In addition, detailed information on input use and allocation, type of land tenure, output levels, input and output prices. amount of water applied, irrigation technology, and annual water budget were included for each crop. Similarly, groundwater quality and wells characterization as well as water management practices used in the farm were included in the questionnaire.

Previous Research on Water Allocation Studies conducted in the past have provided empirical data on water allocation at farm level using various modeling approaches. Caswell and Zilberman (1985) introduced an econometric technique to analyze the factors affecting land shares of alternative irrigation technologies in agriculture. It estimates the likelihood of the use of drip, sprinkler, and surface irrigation by fruit growers in the Central Valley of California. Higher water costs, the use of groundwater. the production of nuts and location were found to increase the likelihood of using drip and sprinkler irrigation The results are used to demonstrate the effectiveness of water price increases in inducing water conservation.

18


Applying a model of the multi-output firm, econometric results are reported for irrigated production in four multistage regions of the American West (Moore et al., 1994a). Cross-sectional micro-data and limited-dependent variable methods are used to estimate crop-choice, supply, land allocation and water demand functions for field crops. Fam-level water demand is disaggregated into the crop-level water demands, which are further separated into an extensive margin (land allocations) and an intensive margin (short-run water use). Response to water price (measured as groundwater pumping cost) occurs primarily at the extensive margin. Moore et al. (1994b) compared three models of input allocation in multi-crop systems. In addition to the variable input and satisficing models analyzed in previous research, a fixed-allocatable input model of short-run input use was derived. The empirical application studied irrigation water use in the Central Plains region of the United States. Based on results from model specification tests and prediction accuracy measures, the fixed-allocatable input model performed better than other models in explaining multi-crop water allocation. In addition, the study presents an alternative approach to the study of deficient data on multi-crop production. By transferring econometric results from analysis of "non-deficient" crop-level data, input allocation in deficient data sets can be predicted. Chambers and Just (1989) solved the problem of determining fixed but allocatable input allocations by dual methods. A flexible, profit function approach for estimating input-nonjoiut technologies with allocatable fixed factors was developed. Variable input allocations can be calculated from the estimated technology. A correct test for input nonjointness that discriminates between true and apparent jointness is derived in a framework that permits fully linear estimation of a second-order flexible technology. Using data only on aggregate variable input use and land allocation, Just et al. (1990) suggest a methodology for allocating variable input use among crops and improvement of regional crop budget information. Two approaches for estimation of variable input allocations among production activities were examined. One relies on behavioral rules, whereby input allocations follow accepted rules of thumb. The alternative approach is derived from profit maximization where input use responds instantaneously to changes in input and output prices. The behavioral rules dominate instantaneous response to prices in explaining the data analyzed in this paper and suggest the validity of a sample behavioral approach for developing enterprise budgets and cost of production estimates. The main problem in estimating non-experimental agricultural production functions is that input data typically are not available by crop. A producer normally grows several crops, but the allocation of inputs among crops is not recorded.


19

What is more commonly available is data on the total use of variable inputs, such as water. On the other hand, allocations of the major fixed factor, land, are observed. Input and output prices and production are generally observable. Thus, a full information-estimation approach must utilize the observed land allocations and compensate for the lack of information on allocations of other inputs. Just et al. (1 983) addressed this issue of multi-crop production function estimation with allocated inputs. The approach uses all available information from both technological and behavioral assumptions in producing estimates of multi-output production functions where data on allocations of variable inputs among crops in not available. Krulce et al. (1997) modeled groundwater as a renewable resource and as replaceable at a fixed cost by a backstop resource (desalination). A steady state is reached when groundwater is depleted to the point where the efficiency price is equal to the unit cost of the backstop resource. The efficiency price (i. e.. the marginal opportunity cost of water) has three components: extraction cost, scarcity rent and residual user cost (called "drawdown cost"). The drawdown cost, always equal to zero for a non-renewable resource, increases with depletion and in the steady state may be larger relative to the extraction cost. For cost recovery of water services in agriculture, three charging mechanisms were evaluated (Perry, 1997): Flat rate, independent of crop type or cropping intensity, Crop-based charge, broadly relating the service charge to water consumption, and Volumetric charge. The results showed that full recovery of allocated costs to agriculture would reduce farm incomes by about 4.5%. Imposition of flat rate charges has no impact on crop selection. More interestingly, a crude crop-based charge (water charges set at levels proportional to typical farm demand by crop) is almost exactly as efficient as full volumetric pricing in inducing beneficial shifts in cropping patterns toward more water-efficient crops. It is concluded that charges for water services will not induce significant changes in cropping patterns, or improvements in system performance, because the cost of system operation is low in relation to the benefits of irrigation. Under present conditions of supply. volumetric charges for water are only marginally more successful in encouraging efficient water use than crop-based charges, which in turn are somewhat better than a flat land tax. Volumetric charges are an unrealistic means of encouraging significant reductions in demand because very high charges are required to have a significant impact.

20


On-Farm Water-Use Efficiency in Radwania, Syria Characteristics of Radwania Farms A total of 80 farmers were sampled from 24 villages in Aleppo province, northwest Syria. They were from Al-Safira (56.3%), Azaz (28.8%) and Jabal Samaan (15%) districts. Majority were from the villages of Radwania (10%), Aum-Hosh (11.3%), and Kanater (10%). Most sample farms (63.8%) were located in rainfall zone 2 while others (36.3%) were in zone 4. On average, each farmer had irrigation experience of 16 years, with some individual farmers having up to 49 years of experience. Majority of the farmers (91.1%) were full-time operators and only 8.9% were part-time. About 89% of them farmers depended completely on farm income. Sixty-two farmers reported that crop production accounted for 100% of their farm income. Soil type was reported to be clay, sandy and loamy by 41.3,26.3 and 32.5% of the farmers, respectively. Also, 79 farmers reported that soil salinity was low. Though most farms (57.5%) had deep soil, 13.8 and 28.8% had medium and shallow soil, respectively. Nearly 94% of the farmers indicated that their cropping pattern was mainly determined by market conditions, while 6.3% were influenced by agricultural policies. Though majority of the farmers (61%) believed that water supply was not limited, 39% thought that it was. Location of water source relative to the farm was not a major concern to the sampled fanners because 78.8% of the farms used groundwater for irrigation. No major restrictions in the form of quantity and quality regulations were imposed on water availability. More than 87% of the farmers indicated that the main reason to irrigate crops was that the amount of rainfall was not sufficient for economic rainfed yield. For the sampled farmers, rule-of-thumb factors determined the amount of water they applied to each crop. Private ownership was the predominant land tenure in the sample farms as reported by 82.3%. The rest of the sample had rented or shared type of land ownership. Other characteristics of the sample farms are presented in Table 1. Among the 80 sample farms, 78, 27 and 72 farmers irrigated wheat, barley and cotton, respectively. Other crops were not of significant importance to the farmers of this study area. For example, corn, potato, sunflower, watermelon, tomato, and green pepper were grown by 2, 13, 1, 3, 21, and 3 farmers, respectively. Therefore, the rest of the analysis will concentrate on wheat, barley and cotton which

On-Furm water-use Efficieny in Radwania,Syria

21

accounted for about 87% of total land on the sample farms. Total farm size averaged 14.78 ha, while average area for wheat, barley, and cotton was 5.21, 3.27, and 4.35 ha, respectively. Table 1. Basic crop and water data for Radwania sample farms. Item

Parameter

Total 80

Wheat 78

Crop Barley 27

Cotton 72

Number of farms Area (hectare) Mean SD Water applied (m3) Mean SD

14.79 19.92

5.21 8.85

3.27 2.78

4.35 6.60

19,831 12,338

4833 3287

3770 2281

15,385 9239

Water applied (m3/ha) Irrigation Rainfall

1341 2840

928 2840

1153 2840

3537

.843

,798

1.108

Crop yield (ton/ha) Mean SD

The average amount of water applied to a farm was 19.831 m3, and average water application by crop was 4833, 3770, and 15,385 for wheat, barley and cotton, respectively. Annual rainfall in the study area during the 199711998 season was 284.4 mm with a standard deviation of 52 mm. Crop yield was 3.391, 2.245, and 3.636 ton/ha for wheat, barley and cotton, respectively. Water productivity, defined in technical terms as kg of output per m of water, was highest for wheat (0.90 kg1m3), followed by cotton (0.57 then barley (0.56 indicating that water yielded more output in wheat compared to barley and cotton. However, this result is mainly based on technical efficiency. To better represent actual farm conditions, economic criteria need to be taken into account by analyzing water allocation among competing crops in a multi-crop system, which is the focus of the rest of this study.

Model Estimation and Validation All data were collected from farm surveys conducted during the summer of 1999. The collected data cover information on crop production, input use, output and input price variables and water management practices for the 199711998 season. The survey also included questions on crop-level acreage, irrigation technology, soil information water sources, rainfall information, annual water budget and

22


on-farm irrigation practices. Irrigation water use by each crop was the dependent variable for the analysis. Several quantitative-and qualitative- independent variables were formed from the survey data. The quantitative variables include irrigated area for each crop (ha), output price of each crop (SL/kg), amount of total water available to the farm (m3), farmer's experience in irrigation (year), water price (SL/ha/year) and prices of variable inputs, such as fertilizers (SL/kg). The qualitative variables include dummy variables on water location to the farm, soil type, soil salinity, soil depth, crop irrigation technology, water application and water management practices. The multi-crop production system is an econometric issue in that equations might be connected not because they interact, but because their error terms are contemporaneously related. For example, a shock affecting demand for water in one crop may spill over and affect water demand for other crops. In this case, estimating these equations as a set using the seemingly unrelated regression estimation (SURE) procedure should improve efficiency (Kennedy, 1985). Other studies applied a limited-dependent variable model to obtain unbiased estimates for multi-crop systems (Moore et al., 1994). The multi-crop system for on-farm water allocation allows contemporaneous correlation between the error terms across equations. This implies that the error terms in the equation were correlated with the error terms in the equation. and thus, the variancecovariance matrix of the system's error term would not be diagonal. Estimating these error correlations and the diagonal elements using the residuals from each equation estimated separately by the Ordinary Least Squares (OLS) procedure, should allow efficient estimates of the parameters. This can be done by estimating the multi-crop system using the SURE procedure. The gain in efficiency given by the SURE estimator over OLS increases directly with the correlation between disturbances from the different equations; and decreases as the correlation between the different sets of explanatory variables increases. Indeed, the SURE estimator reduces to OLS if either the correlations are all zero or the explanatory variables are identical in all equations (Johnston, 1984). For this study the SURE procedure is not expected to provide more efficient estimates compared to the OLS procedure. This is because the explanatory variables are identical in all equations and the correlation between equation disturbances is very small. Under these circumstances, the result will be estimates with greater standard errors than those of the OLS coefficients. Even if the true correlation between disturbances of different equations is zero, the sample OLS residuals may give non-negligible covariance, and one might mistakenly compute

On-Farm, Water-Use Efficieny in Radwania, Syria

23

SURE estimates. But as the correlation increases, the efficiency of the SURE over the OLS estimates increases substantially.

Empirical results Following methodology development for FWUE, the fixed-allocatable input model, the variable input model and the satisficing model were estimated using on-farm data of 80 producers. The estimated models are presented in Appendix A, Tables 1 to 3. Having estimated these models, the second step involves the comparison of alternative models using prediction accuracy measures as a means of model validation. Both in-sample and out-of-sample forecasts for crop-level water use were made and compared with the actual on-farm water applications. The three measures of prediction accuracy were used to judge the performance of alternative models and thus provide evidence on model choice. The measures of mean absolute error (MAE), root mean square error (RMSE). and mean absolute percentage error (MAPE) were calculated to compare the models of FWUE for each crop. Summarized calculations are presented in Table 2.

Table 2. Performance of estimated models in predicting on-farm water use in Radwania, north Syria. Crop

Prediction Accuracy Measure

Fixed-allocatable input model Within Out of sample sample

MAE RMSE MAPE

1260.9

Variable input model Within sample

Out of sample

Satisficing model Within Out of sample sample

1672.3 772.0 0.628

2163.8 2053.0 0.750

3159.2 1263.3 1.106

2076.2 2096.2 0.708

2434.9 1077.9 1.106

Wheat 0.391

Barley

Cotton

MAE RMSE MAPE

958.8 947.2

1479.9 641.6 0.813

924.9 990.3 0.894

1241.5 587.6 0.746

1045.8 1103.6 0.811

1306.2 624.5 0.911

MAE RMSE M APE

1823.8 1593.9 0.168

2641.9 1090.4 0.262

6007.1 5522.3 0.485

9052.9 3865.8 1.156

7577.5 6687.6 0.768

7469.6 3780.3 0.851

For an out-of-sample prediction, the observations were randomly divided into two subsets. one with of the observations and another with the remaining 20% of the observations. The 80% subset was used to estimate each model's parameters. These parameters were applied to the 20% subset to make out-of-sample predictions and to apply the predictionaccuracy measures. This procedure was repeated three times. The calculated measures of prediction accuracy were then averaged for the three draws as presented in this table. Indicates the model that most accurately predicts short-term water use for a given accuracy measure and experiment. "

80%

24


Four sets of forecasts were made, including one in-sample prediction and three out-of-sample predictions (Moore et al., 1994). Applying the three measures to each of the four predictions generated twelve cases for evaluating the alternative models for each crop. It also provided a basis for model choice. This process of comparison among alternative models was repeated for each crop. With the in-sample prediction, the fixed-allocatable input model outperformed the two alternative models (variable input model and satisficing model) according to each of the three measures (MAE, RMSE, and MAPE) for wheat and cotton. For barley, the fixed-allocatable input model outperformed the other two models according to RMSE and MAPE, while the variable input model was superior to the others based on the MAE measure. Results of the out-of-sample predictions demonstrate a similar pattern of performance. For both wheat and cotton, the fixed-allocatable input model performed better than the two alternative models according to the three measures of prediction. The numerical values of MAE, RMSE, and MAPE of the fixedallocatable input model were considerably lower than those of the variable input model and satisficing model for both wheat and cotton. For barley, the variable input model outperformed the other two alternative models according to MAE, RMSE, and MAPE. Results obtained from the application of the prediction accuracy measures support the conclusion that the fixed-allocatable input model represents a better model for explaining short-run water allocation for both wheat and cotton in multi-crop systems than the variable input and satisficing model. For barley, however, the results of out-of-sample predictions provide evidence to support the choice of the variable input model to explain on-farm water use for this crop, whereas in-sample predictions favor the selection of the fixed-allocatable input model. Estimated models can provide additional information on the model selection process for each crop. A close look at the estimated models supports the choice of the fixed-allocatable input model. A key factor is the multi-crop jointness evident in the crop acreage variables as shown in the estimates of this model. For each of the wheat and cotton equations, water use depends strongly on own- and cross-acreages. The relative performance of the water constraint variable (represented by the variable of total water available to the farm) provides additional support for the choice of the fixed-allocatable input model. The water constraint variable is positive and highly significant, especially in wheat and cotton equations of the model. This result suggests that producers perceive irrigation water as a fixed input in the short run. In contrast, water price is not

On-Farm Water-Use Efficieny in Radwania. Syria

25

negative for wheat and corn equations. It is only negative, but not significant, for the barley equation under the estimations of both the fixed-allocatable input and the variable input models. This implies that after planting crops, producers do not respond to water prices in subsequent short-run decisions. Moore et al. (1 994) argue that this result indicates that in terms of the effect of water price on multi-crop profits, the major impact depends on crop-choice, irrigation technology, and land allocation decisions. Once cropland is allocated, the level of water price appears not to have a major quantitative impact on profit; otherwise, water price would be a significant determinant of short-run water use. This conclusion is further supported by the fact that water prices in Syria are fixed and determined by official agricultural authorities. The prices, therefore, have no major influence on the amount of water allocated to each crop. The tixed-allocatable input model may explain producer decisions on short-run water use better than the variable input model (Moore et al., 1994). It also offers a more complex explanation of multi-crop decisions than the behavioral model. In the fixed-allocatable input model, producers operate with a short-run constraint on farm-level water use because of the fixed amount of surface water available to the farm according to the rationing system of water allocation among farms. This leads to competition among crops for water because groundwater is subject to fixed pumping capacity in the short run. It can be concluded that the fixed-allocatable input model is better at studying on-farm water use in a multi-crop system in Radwania area, since it performs best in explaining short-run water use. The estimates reveal the following points: The values of the adjusted coefficient of determination indicate that the model performs well in explaining crop-level water use in the multi-crop system for cross-sectional data. The estimated is 0.93 for cotton, 0.65 for wheat, and 0.53 for barley. Even the lowest value of 0.53 for barley indicates relatively good performance for a crop which is mainly dependent on rainfall. Only a small portion of the sample farms (27 farmers) supplemented irrigation water for barley during spring. The estimated water demand equations for wheat, barley and cotton are significant at the 1% level according to the F-test. Output prices appear to be a strong determinant of short-run decisions on water allocation among competing crops. Own-price variables for wheat, barley, and cotton are positive and signiticant in explaining water use. An increase in the wheat price by 1 SL/kg, holding other variables constant, will increase water use for wheat by 198 m'. Similarly, a 1 SL/kg increase in the

26


cotton price will increase water demand for cotton by 1240 m3. Cross-price variables reflect intercrop interdependence, i.e., a change in the price of one crop induces water reallocation among other crops given that water is a fixed allocatable input. An increase in the cotton price by 1 SL/kg will reduce the amount of water allocated to wheat by 102 m3, whereas a 1 SL/kg increase in the wheat price will induce a 216 m3 reduction in water use for cotton. The performance of the cross-price variables demonstrates the competition among wheat, barley, and cotton in a multi-crop system for a fixed quantity of water. Estimates of the water constraint variable (total water) indicate the allocation among crops of a marginal increase in farm-level water availability for producers growing competing crops. The individual estimated coefficients of the water constraint suggest that a 1 m3 increase in water available to the farm will be used for cotton in the first place (0.77 m3), for wheat in the second place (0.20 m3), and for barley with a negligible amount (0.03 m3). These figures indicate that an increase in water availability is allocated most heavily to crops with relatively high water requirements like cotton rather than to crops with relatively low water requirements such as wheat and barley. In fact, barley is considered a residual crop in terms of on-farm water use, thus, only minimal amount of marginal increment in water is allocated to this crop.

Water-Use Efficiency The target production level for wheat, barley and cotton are the average yield levels of the sample farms as reported in Table 3. To obtain the required amount of water to produce these average yield levels the estimated crop equations with the fixed-allocatable input model were used. This was done by calculating the amount of water required for each crop at the mean levels of the independent variables appearing in that equation. The calculated levels of required water are presented in Table 3 and compared with the actual amount of water use. It is clear that FWUE in wheat and barley was 0.61 and 0.45, indicating that actual water use exceeded the water requirement by about 39 and 55%, respectively. Cotton producers, on the other hand, exceeded the water requirement by 24%. FWUE for cotton of 0.76 is relatively high given that cotton is a very water-demanding crop and its growing season is about nine months. Either below-average yields or inefficient use of irrigation can explain these estimates of relatively low ratios of FWUE for barley, wheat and cotton in the study area.

On-Farm Water-Use Efficieny in Radwania, Syria

27

Table 3. FWUE of Radwania farmers in the production of wheat, barley and cotton. Crop Wheat Barley Cotton

Irrigated area (ha) 5.21 3.27 4.35

Yield Actual (ton/ha) water used (m3) 7678 3.391 6615 2.245 18230 3.636

Required water (m3) 4721 2972 13837

(%)

0.90 0.56 0.57

61 45 76

aW P and FWUE are total water productivity and water-use efficiency. respectively, including rainwater of 2844

Farmers in this study area over-irrigated barley and cotton crops as a result of their perceptions of water requirements, their expectations of rainfall and market conditions. More cases of over-irrigation occurred in barley production, suggesting that recommendations for SI could benefit some barley fanners. The low ratio of FWUE in wheat production suggests that a wide technology gap exists between the recommended SI practices for wheat and the actual water application in the study area. This has important policy implications since the Syrian agricultural plan assigns quotas of land to be planted with wheat, which accounts for 35% of total irrigated land in the sample farms. Therefore, improved FWUE for wheat can contribute to the overall FWUE for the agricultural sector. In this study the overall (combined) FWUE for wheat, barley, and cotton was 62%.

28


On-Farm Water-Use Efficiency in Rabea, Iraq Characteristics of Rabea Farms In Rabea district, northwest Iraq, 100 farmers were sampled for the study from the Al-Jazera Supplemental Irrigation project located in the moderate rainfall zone (350-450 mm). Their irrigation experience ranged from 1 to 20 years, with an average of 11 years. Surface water is the main source for irrigation for all farmers in the project, and sprinkler imgation is the predominant supplemental irrigation technology for winter crops (e.g., wheat, sugar beet, and potato). Flood irrigation is used for summer cropping (e.g., tomato). The soil type was reported to be predominantly loamy to clayey. Also, 68% of the farmers reported that soil salinity was low. Most farmers (87%) reported that the soils on their farm were deep, while 13% reported that theirs were shallow. Majority of the farmers (69%) indicated that water supply was limited. Only 31% of the producers reported that the amount of water was not limited. Location of water source relative to the farm was a key concern to the sample farms since only 27% of them were at head location, while 73% were at medium or tail locations. This is an important factor as surface water is the main source for irrigation. The cropping pattern was mainly determined by market conditions and agricultural policies. Rented and sharecropped land were the predominant types of land tenure as reported by 76% of the sample farms. The rest of the samples (24%) were on privately-owned land. Other characteristics of the sample farms are presented in Table 4. Among the 100 sample farms, 47, 45, 22 and 71 were of irrigated wheat, potato, sugar beet, and tomato, respectively. While supplemental irrigation (SI) was applied to wheat, the other crops were subjected to full irrigation. Since there were no other irrigated crops produced in the study area, the rest of the analysis will concentrate on wheat, potato, sugar beet, and tomato. Total farm size averaged 37.02 ha. Average irrigated area was 25.35, 8.92, 3.48, and 4.83 ha for wheat, potato, sugar beet and tomato, respectively. Among winter crops, wheat accounted for 68% of total irrigated land. The rest of the land for winter cropping (32%) was allocated for potato and sugar beet.

On-Farm Water-Use Efficieny in Rabea, Iraq

29

Table 4. Basic crop and water data of Rabea sample farms. Item Parameter/Units Total farm Wheat Potato Sugar beet Tomato Number of farms 100 47 45 22 71 Area (ha) 4.8 8.9 3.5 25.4 Mean 37.02 6.5 11.4 5.3 43.8 SD 74.19 Water applied (m3)

Mean

Irrigation Rainwater Crop yield (ton/ha)

448,006

5424

75,216 40,289 70,763

12,102 2930

214 2930

8432 2930

11,577 14,651

Mean

The average amount of water available to the whole farm for winter cropping was for 448,006 m3. The water application by crop was 5424 m3 for wheat, 75,216 3 potato, and 40,289 m for sugar beet. For summer cropping, water available to each was used for tomato. Annual farm was 210,970 m3, on average, of which 70,763 rainfall for the study area during the 199711998 season was 292.78 mm, with a standard deviation of 89 mm. Crop yield was 2.184, 16.31 1, 14.091, and 12.761 ton/ha for wheat, potato, sugar beet and tomato, respectively. Water productivity, defined in technical terms as kg of output per m3 of water, was highest for potato (1.44 kg/m3), followed by sugar tomato (0.73 and wheat (0.70 kg/m3). Therefore, water beet (0.97 yielded more output in potato production compared to wheat, sugar beet, and of water gave 1.44 kg of potato tubers. The output of tomato. Each additional other crops was much lower for each additional unit of water. A comparison of water productivity of wheat under rainfed and SI provides important findings. Among the 100 sample farmers, 53 farmers grew wheat under rainfed conditions and 47 used SI. The average grain yield was 2.36 ton/ha for wheat under SI and 1.36 ton/ha for rainfed wheat. This indicates that SI increased wheat grain yield by 62%. In addition, SI improved water productivity from 0.6 under SI, taking the amount of rainfall into kg/m3 for rainfed wheat to 0.70 account.

The other important advantage of using SI in wheat production is yield stabilization. Using the coefficient of variation (C. V) as a crude measure for yield stabilization, it was found that the C. V for wheat yield under rainfed conditions was 71%. but reduced to 35% under SI.

30


This is a very important result for risk-averse producers as SI increases yield stability, and thus, reduces risks associated with rainfed farming. This result compares favorably with other findings in the region. Average water productivity of rain in producing wheat in the dry areas of WANA is about 0.35 kg although with good management and favorable rainfall it can be increased to 1 kg grain/m3. However, water used in SI can be much more efficient. Research at ICARDA showed that a cubic meter of water applied at the right time (when the crop is suffering from moisture stress), combined with good management, could produce more than 2.5 kg of g a i n over the rainfed production (Oweis, 1997). In Jordan, rainfall water productivity in rainfed wheat in Mushagar (300 mm) was 0.33 When one cubic meter of rainfall was combined with 0.5 m3 SI, the overall water productivity was increased to 3.5 kg/m3 (Oweis and Salkini, 1992).

Empirical Results The three models were estimated using on-farm data of 100 farmers. The estimated models are presented in Appendix B, Tables 1 to 3. Both in- and out-of-sample forecasts for crop-level water use were made and compared with the actual on-farm water applications. Application of the three measures of prediction accuracy was used to judge the performance of alternative models and to provide evidence on model choice. The measures of MAE, RMSE, and MAPE were calculated to compare the models of on-farm water use for each crop. Summarized calculations are presented in Table 5. Four sets of forecasts were made, including one in-sample prediction and three out-of-sample predictions. Applying the three measures to each of the four predictions generated twelve cases for evaluating the alternative models for each crop. Application of the three measures of prediction accuracy provided evidence on model choice. This process of comparison among alternative models was repeated for each crop. With the in-sample prediction, the variable input model outperformed the other two models for wheat, potato, and sugar beet according to the MAE. The satisficing model performed the best for in-sample predictions of tomato according to the same accuracy measure. The RMSE indicates that the fixed-allocatable input model outperfomed the other two models for the in-sample predictions of the four crops. However, the satisficing model gave the best in-sample predictions for the four crops according to the MAPE.

On-Farm Water-Use Efficieny in Rabea. Iraq

Table 5. Performance of estimated models in predicting on-farm water use in Rabea. Crop Prediction Fixed-allocatable Variable input model Satisficing Accuracy input model model Measure Within Out of Within Out of Within Out of sample sample sample sample sample sample Wheat 3127 3366 3286 4222 MAE 2152 4619 4679 1579 RMSE 1.56 1.36 MAPE Potato 7576 MAE 6170 8752 5939 8596 4975 RMSE 4730 0.410 MAPE 0.409 0.229 0.366" Sugar beet 4703 6192 5716 MAE 4813 5248" 7058 6092 8041 615Xh 3998 RMSE MAPE 0.362 0.382 0.664 0.681 Tomato MAE 13849 17433 13271 14153 RMSE 13656 20106 10586 21477 8937" MAPE 0.326 0.315 0.348 0.367 For an out-of-sample prediction. the observations were randomly divided into two subsets, one with 80% of the observations and another with the remaining 20% of the observations. The 80% subset was used to estimate each model's paramrters. These parameters were applied to the 20% subset to make out-of-sample predictions and to apply the predictionaccuracy measures. This procedure was repeated three times (Moore et al., 1994). The calculated measures of prediction accuracy were then the model that most accurately averaged for the three draws and presented in this table predicts short-term water use for a given accuracy measure and experiment.

For out-of-sample predictions, the results were mixed. The satisficing model performed best for potato, sugar beet. and tomato according to the MAE. Similarly, this prediction model was the best for wheat. potato. and tomato based on the MAPE. The satisficing model also outperformed the other two for sugar beet and tomato according to the RMSE. However, the variable input model was the best for wheat according to the MAE and RMSE. This model was also the best for sugar beet according to the MAPE (Table 5). The overall prediction performance of the estimated models can be judged on both in-and out-of-sample predictions. Based on the accuracy measures, the MAE favors both the variable input model and satisficing model four times each. Meanwhile, the RMSE favors the fixed-allocatable input model five times, while each of the variable input model and satisficing model are favoured once.

31

32


The MAPE supports the choice of satisficing model six times, and the other two models once each. Regardless of the type of accuracy measure, the satisficing model is the best eight times for out-of-sample prediction, and four times for in-sample prediction. The variable input model is the best three times, each, for both in- and out-of-samplepredictions. The fixed-allocatable input model is the best five times for on-sample predictions and one time for out-of-sample predictions. The overall performance of the three models regardless of the type of predictions and accuracy measures is that the satisficing model outperforms the other two in 12 out of 24 cases. Meanwhile, both the fixed-allocatable input model and the variable input model are the best for in six cases each. Accordingly, results obtained from the application of the prediction accuracy measures are inconclusive, although they are in favor of the satisficing model. To better explain the farmers' short-run decisions of water allocation among competing crops, the estimated models can provide further insights. These estimates reveal the following points: Own-crop area and price appear to be the most important two variables explaining the farmers' water use decision in irrigating potato, sugar beet, and tomato. The estimated coefficients of these two variables are positive and highly significant in each water-use equation of the three crops. This is the case for the estimates of both the fixed-allocatable input model and the variable input model. With the fixed-allocatable input model, total water available is the most important factor explaining the water use for wheat. The crop-irrigated area is also positive but not significant in the water demand equation for wheat. Total water available to the farm is also a key variable in the water-use equation for tomato, with the fixed-allocatable input model. Cross-acreage variables in the fixed-allocatable input model do not fully support the multi-crop jointness in the study area. This can be attributed to the fact that wheat is mainly considered a rainfed crop and only residual amount of water was used to supplement rainfall by only 47% of the farmers. For the other two competing crops, potato and sugar beet, the water was available in sufficient amounts and not a limiting factor for their production. For tomato, as a summer crop, the amount of water available to the farm was less than that available during winter, but there were no competing crops with tomato. The water constraint variable is positive in the water-use equations of the four crops, but it is significant in wheat and tomato equations. This result suggests that producers perceive water as a fixed input in the short run.This is further supported by the fact that the water price is not negative in the water demand equations of the four crops with the variable input model. This implies that after planting crops, producers do not respond to water prices in subsequent short-run decisions. Since the price of water in the country is highly subsidized, it does

I

On-Farm Water-Use Efficieny in Rabea, Iraq

.

33

not have a major quantitative impact on water allocation. Land allocation, crop choice, irrigation technology, and output prices are the main determinants of multi-crop water-use decisions. The values of the adjusted coefficient of determination indicate that the three models perform well in explaining crop-level water use for potato, sugar exceeds 0.90, supporting the high beet, and tomato. The estimated explanatory power of the three models. Only the fixed-allocatable input model gives acceptable explanation for water use in wheat production. Even the low value of 0.33 for wheat indicates relatively fair performance for a crop which is mainly dependent on rainfall, with only residual amount of SI. All estimated water-use equations for potato, sugar beet, and tomato are significant at the 1% level according to the F-test. For wheat, however, only the water-use equation of the fixed-allocatable input model is significant. Estimated coefficients of total water available to the farm provide important implications on water allocation among competing crops in a multi-crop system. will be allocated for potato in the first An increase in water availability by I place (0.79 m'), sugar beet in the second place (012 m3), and wheat and and 0.018 respectively. This is tomato with minimal amounts of 0.01 1 consistent with the fact that potato actually uses more water than the other winter crops (wheat and sugar beet).

Water-Use Efficiency The target production levels of wheat, potato, sugar beet, and tomato are the average yield levels of the sample farms as reported in Table 6 . To obtain the required amount of water to produce these average yield levels, the estimated crop water-use equations with the three models are used. This is done by calculating the amount of water required for each crop at the mean levels of the independent variables appearing in that equation. The calculated levels of required water are presented in Table 6 and compared with the actual amount of water use. Both the variable input and the satisficing models under-estimate the amounts of required water for the four crops compared to the estimates of the fixed-allocatable input model. FWUE was highest for tomato (68%), indicating that actual water used exceeded water requirements by about 32%. The lowest FWUE for sugar beet of 32% suggests that producers over-irrigate this crop considerably. Sugar beet producers exceeded the water requirement of the crop by 68%. Therefore, any improvement in FWUE of this crop will save a large amount of water that can be used to expand its irrigated area or for other crops. Either below-average yields or inefficient use of irrigation can explain the low ratios of FWUE for sugar beet.

34


Table 6. FWUE and water productivity of major crops using three models at Rabea, northern Iraa. Item Irrigated area (ha) yield (ton/ha) Actual water used' (m3) Water productivity (Kg/m3) a Required water level (m3) Fixed-allocatable input model Variable input model Satisficing model (%) Fixed-allocatable input model Variable input model Satisficing model Average

Wheat 25.35 2.18 8352 0.70

Potato 8.92

Sugar beet Tomato 3.48 4.83

3059 2733 2550 0.37 0.33 0.3 1 0.34

* includes rain water of 2928 winter crops only. 'Water productivity = Crop yield (Kg)/ Amount of water used (m3), FWUE is calculated based on the estimates of the fixed-allocatable input model. Farmers in the study area over-irrigated wheat and potato by 63 and 55%, respectively, compared to the required amount of water to produce the achieved yield levels. The FWUE for wheat and potato was 37 and 45%, respectively. These figures suggest that a big technology gap exists between the required irrigation practices for wheat and potato and the actual water application in the study area. This has important policy implications in that most of the land is allocated for wheat and potato production, and because potatoes consume a large amount of water. Therefore, improved FWUE for wheat and potato can contribute to the overall FWUE in the study area. In this study the overall FWUE for wheat, potato, sugar beet, and tomato was 40%, suggesting a high potential for saving water once FWUE is improved.

On-Farm Water-Use Efficieny in Al Ghor: Jordan

35

On-Farm Water-Use Efficiency in Al Ghor, Jordan Characteristics of Al Ghor Farms In Al Ghor, 70 farms were included in the sample. These were located in North Al Ghor (63%) and Deir Alla Ghor (37%) districts, distributed among 23 villages. Most of the sample farms (63%) were located in a rainfall zone of 350-500 mm, with average rainfall of 425 mm, and others (37%) were in a rainfall zone of 150-300 mm, with an average annual rainfall of 225 mm. The annual rainfall for the study area during the 200012001 seasons was 351 mm, with a standard deviation of 97.3 mm. The farmers' experience in irrigation ranged from 1 to 47 years, with an average of 17 years. Majority of those in the sample (89%) were full-time farmers, while only 1 1% were part-time. Farming was the main source of income, with 83% of the producers completely dependent on farm income (100% of the household income). The rest of the farmers (17%) were only partially dependent on farm income (20 to 90% of the household income). Crop production accounted for 100% of the farm income, as reported by 97% of the farmers interviewed. Soil type was reported to be medium, sandy and heavy by 79, 13 and 8% of the farmers, respectively. Most farms (67%) were of medium-deep soil, and deep and shallow soil accounted for 29% and 4% of the total farms, respectively. Meanwhile, 63,29, and 8% of the farmers reported that soil salinity was low, medium and high, respectively. The cropping pattern was mainly determined by market conditions, as indicated by 66% of the sample farmers. The cropping pattern for 27% of the farmers was determined by a combination of market conditions and agricultural policies. Other factors explain the cropping pattern of the remaining 7% farmers. The amount of water available to the farms was very limited, as indicated by 86% of the producers. Only 14% of the farmers reported that water was not limited. Location of water source relative to the farm was not a main concern, since 94% of the farms were at headwater locations. Many restrictions, in the form of quantity, quality and regulations, are imposed on water availability. Of all the producers in the sample, 98 and 77% reponed restrictions on water quantity and quality, respectively. Irrigation date was another type of water restriction imposed on 96% of the farms. The main reason for irrigating crops was that the amount of rainfall mas not

36


sufficient for an economic rainfed yield, as indicated by 97% of the sample farmers. General rules and irrigation experience determine the amount of water the farmers apply to each crop, according to the survey sample. All farms were fully imgated from surface water sources. However, land use was not restricted as indicated by 71% of the producers. Rented land was the predominant land tenure feature in the sample farms, as reported by 52%. The rest of the sample farms were characterized by private ownership (17%) and share-cropped land ownership (3 1%). Other features of the sample farms are depicted in Table 7. The crops grown by the farmers in the sample were: tomato (26), potato (21), squash (12), pepper (12), cucumber (21), cauliflower (10), citrus crops (19), melon (7), wheat (6), eggplant (7), lettuce (8), beans (14), broad beans (9), cabbage (7) and onion (8). The average farm size was 5.90 ha. The average crop area for tomato was 1.13 ha, potato 1.93 ha, squash 1.02 ha, pepper 0.65 ha, cucumber 0.74 ha, cauliflower 0.39 ha, citrus 3.09 ha, melon 0.37 ha. wheat 0.60 ha, eggplant 0.74 ha, lettuce 0.95 ha, beans 1.92 ha, broad bean 1.36 ha, cabbage 0.84 ha, and onion 0.53 ha. The average amount of water applied to a farm was 12,284 m3. The average amount of water applied for the various vegetable crops was highest (7109 m3) for eggplant and lowest (2041 m3) for beans. Crop yield was highest (60.3 ton/ha) for tomato, followed by cucumber (52.2 ton/ha). Water productivity, defined in technical terms as kg of output per m3 of water, was highest for tomato and lettuce (8.48 kg/m3 and 7.22 kg/m3. respectively). In the calculation of water prodnctivity, excluding the amount of rainfall changes the crop water productivity considerably. Thus, with only inigation the highest water productivity was 17.84 kg/m3 for potato and bean. The water productivity of and lettuce (16.97 was third and fourth, respectively tomato (16.89 (Table 8). These results indicate that water yields more output in the production of tomato, potato, lettuce and beans. This, however, is mainly based on technical efficiency. To better represent farm economic conditions, output prices need to be taken into account as well. Thus, water productivity was redefined in monetary terms as of water. On the basis of this definition Jordanian Dinars (JD) of output per water productivity was highest for lettuce (1.877 JD/m3), followed by bean (1.806 and broad bean (1.01 JD!m3). Therefore, changing the definition of water productivity from technical to monetary terms has important implications on the ranking of crops with respect to water productivity.


37

38

Assessing On-Far-m Water-Use Efficiency

Table 8. Water productivity of the major crops in Al Ghor, Jordan, sample farms. Crop Price (JD/kg) Water productivity Water productivity (JD/m3) (kg/m3) Total Irrigation Total Irrigation Tomato 0.101 8.48 16.89 0.856 1.706 Potato 0.160 4.50 17.84 0.720 2.854 0.173 3.89 10.05 Squash 0.673 1.739 0.138 5.42 8.29 Cucumber 0.748 1.144 Citrus 0.256 5.73 0.732 1.467 2.86 Wheat 0.200 0.45 0.86 0.090 0.172 Eggplant 0.175 2.70 3.73 0.653 0.48 1 0.260 7.22 16.97 Lettuce 1.877 4.412 0.424 4.26 17.84 Beans 1.806 7.564 Broad bean 0.500 2.01 4.96 1.005 2.480 0.157 5.38 9.07 Cabbage 0.845 1.424 Onion 0.131 3.05 4.81 0.400 0.630 Note: JD is Jordan Dinar and was equivalent to 1.42 US$ during the surveyperiod.

Model Estimation and Empirical Results To better assess FWUE, the study analyzed water allocation among competing crops in a multi-crop system. Farm survey data collected during the summer of 2001 included information on crop production, input use, output and input price variables and water management practices for the 200012001 season. The survey also included questions on crop-level acreage, irrigation technology, soil information, water sources, rainfall, and annual water budget and on-farm imgati on practices. Irrigation water use by each crop was the dependent variable for the analysis. Several quantitative and qualitative independent variables were formed from the survey data. The quantitative variables included irrigated area (ha) for each crop, output price of each crop (JD/kg), amount of total water available to the farm (m3), farmer experience in irrigation (year), amount of rainfall (mm), water price (JD/ha/year) and prices of variable inputs, such as fertilizers (JD/kg). The set of qualitative variables included dummy variables on water location relative to the farm, soil type, soil salinity, soil depth, and crop irrigation technology and water management practices. The three specified models: the fixed-allocatable input model, the variable input model and the satisficing (behavioral) model were estimated using the ordinary least squares procedure. Comparing the estimated coefficients of the three models was not plausible based on economic theory in terms of the signs of price and


39

acreage variables. For example, it is expected that the effect of crop price on water use would be positive. However, estimated coefficients demonstrated negative relationships between crop prices and amount of water use. Similarly, the estimates did support the negative relationship between water price and its use. Therefore, estimates of the variable input and the fixed-allocatable input models were not used for the rest of the analysis. Instead, simplified estimates of the behavioral model were used. This is consistent with irrigation water use in Al Ghor. Results of the survey clearly demonstrate that the cultivated area for each crop mainly determines water allocation among competing crops. Economic conditions were not seen to affect water allocation and application among crops. Furthermore. the amount of water applied to each crop was mainly determined by general rules and farmers' experience. Under these circumstances the main problem facing farmers in the Al Ghor area was allocation of water among competing crops. Survey data indicate that the amount of irrigation water applied for squash, broad bean and cabbage was fixed for all farmers, consequently, no models were estimated for these three crops. The estimates of the behavioral model for the remaining crops are presented in the Appendix C and Table 1. The behavioral estimates provide a simple procedure for estimating water allocation among crops.

Water Allocation Farmers rely on extension information and their irrigation experience to allocate water among competing crops. Since farmers are independent decision makers, there is a wide variation in the share of land allocated to the different crops. It was also noted that land allocation had a direct relationship with the amount of water allocated to a particular crop. The study shows that farmers behave as if their production functions follow constant returns to scale. Therefore, farmers adapt recommended input-output ratios (norms) developed by the extension system. However. individual farmers deviate from these norms according to their specific personal and locational characteristics (Just et al., 1990). Table 9 presents average estimated and actual water use on the farms, derived from the behavioral model. Estimated (required) quantity of water represents the amount of water required to produce the output levels actually produced by the sample farms. The target production levels of the crops are the average yield levels of the sample farms. To obtain the required amount of water to produce these average yield levels, the estimated crop equations of the behavioral model were used. This was done by calculating the amount of water required for each crop at the mean levels of the independent variables appearing in that equation. The calculated levels

40


of required water are also presented in Table 9 and compared with the actual amount of water used. If the amount of rainfall is taken into consideration in the calculation of FWUE, the efficiency of irrigation water is less than one, implying that producers over-irrigate their crops. The percentage of over-irrigation ranged from a minimum of 23% in the production of citms crops to a maximum of 70% in the production of wheat. Citrus and eggplant production is relatively more efficient with a FWUE of 77 and 66%, respectively. Farmers of potato, cauliflower, melon, wheat, lettuce, bean and onion were less efficient as they exceeded water requirements by more than 50%. Producers of tomato, pepper and cucumber achieved medium level of FWUE as they exceeded water requirements by less than 50%. Table 9. FWUE in the Jordan Valley sample farms.

I

Crop

Tomato Potato Squash Pepper Cucumber Cauliflower Citrus Melon Wheat Eggplant Lettuce Beans Broad bean Cabbage Onion

Irrigated area (ha) 1.13 1.93 1.02 0.65 0.74 0.39 3.09 0.37 0.60 0.74 0.95 1.92 1.36 0.84 0.53

Yield (ton/ha) 60.30 20.45 21.70 Na 52.23 NA 22.50 NA 30.83 35.86 40.8 1 18.97 11.36 33.57 25.00

Actual water used (m3) 4038 2212 2203 3780 4660 2877 12126 3221 2160 7109 2284 2041 3110 3110 2770

Required water (m3) 4014 2222 NA 3824 4572 2914 12046 2938 1684 6999 2305 2023 NA NA 2809

WUET %

53 40 NA 53 56 46 77 44 30 66 40 37 NA NA 45

WUET is the estimated water-use efficiency including both irrigation and ramwater 3500 (NA) The amount of water applied is fixed for all farmers, thus required water was not estimated

The study has proved that farmers over-irrigate crops because of their perceptions of water requirements and their expectations of rainfall and market conditions. The low levels of FWUE in potato, cauliflower, melon, wheat, lettuce, bean and onion production suggest that a wide technology gap exists between the recommended irrigation practices and the actual water application in the study area. This result has important policy implications, since Jordan is classified as a water-scarce country. Therefore, improving FWUE for these crops can contribute to the overall FWUE for the agricultural sector.

On-Farm Water-Use Efficieny in Nubaria, Egypt

41

On-Farm Water-Use Efficiency in Nubaria, Egypt Characteristics of Nubaria Farms Nubaria area is located in the Nile Delta of Egypt between Cairo and Alexandria. The sample farms comprised 50 producers distributed among three villages (village 13, village 15 and village 17). Their irrigation experience ranged from 1 to 41 years, with an average of 17.4 years. Surface water was the main source of irrigation for all farmers in the survey. The major crops produced on the sample farms were wheat, faba bean and bersem (winter cropping) and watermelon, tomato, green pepper, squash and corn (summer cropping). On most farms (66%) the soils were deep, and the remaining 37% were medium or shallow soils. According to the data, 56% of the farmers reported that soil salinity was low, while high and medium salinity was reported by 30 and 14% of the farmers, respectively. The cropping pattern at the time of the study was determined by market conditions in the first place and agricultural rotation in the second place as indicated by 98% of the sample farms. Selection of crops was based on market conditions as indicated by 72% of the farms. Crop labor requirements were the second important factor in crop selection as reported by 24% of the sample farms. Location of water source relative to the farm was a major concern to the sample farmers, since only 36% were at the head location, while 64% were at medium or tail locations. This is an important factor, since surface water is the main source of irrigation. Majority in the sample (80%) were full-time farmers, while 20% were part-time. Farm income constituted 84% of the total income, while off-farm income contributed 16%. Crop production was the main source of farm income (81%) and livestock production accounted for 19%. The number of farmers and crops produced on sample farms can be summarized as follows: wheat (43), faba bean (34), bersem (28), water melon (30), tomato (26), green pepper (7), squash (22) and corn (21). Other crops were produced by a negligible number of farmers. The average farm size was 5.7 feddan. Average farm size per crop was: 2.16 feddan for wheat, 1.78 feddan for faba bean. 1.23 feddan for bersem, 2.30 feddan for watermelon, 0.8 1 feddan for tomato, 0.57 feddan for green pepper, 2.98 feddan for squash and 0.64 feddan for corn. Among winter crops, wheat accounted for 36% of total irrigated land, while 30 and 21% of the land was allocated for faba bean and bersem, respectively. For summer cropping, more land was devoted to watermelon and squash compared to other crops.

Assessing On-Farm Water-Use Effciency

42

The average amount of water available to the entire farm was 10,149.5 m3 for the sample farms. The water application per crop for winter cropping was 2328 m31feddan for wheat, 1764 m3/feddan for faba bean and 3501 m3/feddan for bersem. For summer cropping, water application, as an average for the sample farms, was 1700 m3/feddan for watermelon, 2333 m3/feddan for tomato, 3030 for green pepper, 2139 m31feddan for squash and 2979 m3/feddan for corn. According to most of the sampled farmers (90%), the amount of water applied to each crop was mainly determined by the crop water requirements. The other 10% indicated extension recommendations as the determinants for the amount of water application to each crop. Other characteristics of the sample farms are presented in Table 10.

Table 10. Basic crop and irrigation water data for Nubaria, Egypt. Item

Total

Number of farms

50

Winter cropping Wheat Faba Bersem Water bean melon 43 34 28 30

I

Area cropped (fed) a Mean 5.97 SD' 0.16

2.16 1.32

1.78 1.23 1.50 1.56

2.30 1.92

0.81 0.98

0.57 0.53

2.98 2.77

0.64 0.76

Crop yield (ton/fed) Mean SD'

2.9 4.2

0.68 32.9 0.52 49.1

9.7 25.3

6.75 9.38

4.00 3.83

4.35 5.00

2.57 5.49

1764 3501 830 1146

1700 511

2333

3030 1165

2139 1425

2979 736

Water applied (m3) 10049 2328 4443 663

-

hectare = 2.38 Feddans

Water productivity, defined in technical terms as kg of output per m3 of water, was highest for bersem (7.24 kg/m3) in winter cropping and watermelon (3.5 in summer cropping. Therefore, water yielded more output in bersem production, compared to wheat and faba bean in winter cropping. Each additional m of water yielded 7.24 kg of bersem, whereas the output of other crops was much lower for each additional unit of water. For summer cropping, water gave more output in watermelon, compared to tomato, green pepper, squash and corn.

On-Farm Warer-Use Efficieny in Nubaria, Egypt

43

Empirical Results The three specified models were estimated using on-farm data of 50 producers. The estimated models are presented in Appendix D, Tables I to 3 for winter cropping and Tables 4 to 6 for summer cropping. Both in-sample and out-of-sample forecasts for crop-level water use were made and compared with the actual on-farm water applications. The three measures of prediction accuracy (MAE, RMSE and MAPE) were used to judge the performance of alternative models, and thus provide evidence on model choice. The measures were calculated to compare the models of on-farm water use for each crop. Summarized calculations are presented in Tables 11 and 12 for winter and summer cropping, respectively. Two sets of forecasts were made-one in-sample and another out-of-sample prediction. Applying the three measures to each of the two predictions generated six cases for evaluating the alternative models for each crop and provided evidence on model choice. Results for winter cropping With the in-sample prediction, the tixed-allocatable input model outperformed the other two models for winter crops (wheat, faba beans and bersem) according to the three prediction accuracy measures (Table 11). Table 11. Performance of estimated models in predicting on-farm water-use in Nubaria, for winter cropping. Crop

Prediction Accuracy Measure

Fixed-allocatable input model Within Out of sample sample'

Variable input model Within Out of sample sample'

Satisficine model Within Out of sample

642.0 758.1 0.288

904.5 1218.4 0.339

578.3 738.8 0.244

691.7 925.0 0.234

606.43 857.0 0.355

63 896.1 0.884

498.7 783.4 0.306

635.01

1037.7 1243.6 0.356

947.2 1067.3 0.278

805.0 1054 0.927

Wheat

MAE RMSE MAPE Faba bean MAE RMSE MAPE

662.9 887.5

0.464

Benem MAE RMSE MAPE

491.7, 0.127"

415.4 547.7

For an out-of-sample prediction. the observations were randomly divided into two subsets, one with 80% of the observations and another with the remaining 20%. The 80% subset was used to estimate each model's parameters. Theseparameters were applied to the 20% subset to make out-of-sample predictions and to apply the prediction accuracy measures. Indicates the model that most accurately

predicts short-tenn water use for a given accuracy and experiment.

44

Assessing On-Farm Water- Use Efficiency

For out-of-sample predictions, the fixed-allocatable input model performed best for wheat according to the three measures of prediction accuracy. This prediction model also performed best for bersem according to tbe RMSE and MAE, and for faba bean according to the MAE. The behavioral model outperformed the other two models for faba bean and bersem, according to the RMSE and MAPE, respectively. Only in one case did the variable input model perform best for faba bean based on the MAE. The overall predictions performance of the estimated models can be judged on both in-sample and out-of-sample predictions. Regardless of the type of accuracy measures, the fixed-allocatable input model was the best nine times for in-sample predictions, and six times for out-of-sample predictions, whereas the behavioral model was the best twice for out-of-sample predictions. The variable input model was the best only once for out-of-sample predictions. Overall, regardless of the type of predictions and accuracy measures, the fixed-allocatable input model outperformed the other two in 15 of 18 cases. Meanwhile, the behavioral model and the variable input model were the best in only twice and once, respectively. Accordingly, the results obtained from the application of the prediction accuracy measures clearly support the dominance of the fixed-allocatable input model. The estimated parameters of this model as well as those of the other two models can provide further insights in explaining the farmers' short-run decisions on water allocation among competing crops. These estimates reveal the following points: Own- and cross-crop area and price appear to be the most important variables in explaining the farmers'water-use decisions in irrigating wheat, faba bean and bersem. The estimated coefficients of area and price variables have the correct signs and are highly significant in the fixed-allocatable input model. Cross-acreage variables in this model support the multi-crop jointness in the study area, implying a high degree of competition among wheat, faba bean and bersem for the available amount of water.

.

Total water available to the farm is the most important factor explaining the water use for wheat, faba bean and bersem. The estimated coefficients of this variable in the fixed-allocatable input model provide valuable information on water allocation among competing crops in a multi-crop system. An increase in is allocated equally for the three crops by an amount water availability of 1 3 of 0.33 m , each. This result implies that farmers give equal weight to the three winter crops in water allocation.

On-Farm Water-Use Efficienyin Nubaria,Egypt

45

Results for summer cropping Alternative models were compared using the three prediction performance measures. With the in-sample prediction, the behavioral model outperformed the other two models for watermelon, tomato, green pepper and corn. The fixed-allocatable input model performed best for in-sample predictions of squash, according to the same accuracy measures. The MAPE indicates that the variable input model outperformed the other two models for the in-sample predictions of watermelon and green pepper. For out-of-sample predictions, the behavioral model performed best for watermelon, tomato, and corn, based on the three measures of prediction performance.The behavioral model also outperformed the other two models for squash, according to the RMSE. However, the fixed-allocatable input model was best for squash according to the MAE and MAPE. The variable input model outperformed the other two models for green pepper based on the three prediction performance measures. The overall prediction performance of the estimated models was finally judged based on both in-sample and out-of-sample predictions. Based on the accuracy measures, the behavioral model dominated the other two models 15 times based on the MAE, RMSE and the MAPE. The RMSE supported the choice of this model twice, while the MAE supported the behavioral model once. Regardless of the type of accuracy measures, the behavioral model was the best 10 times for in-sample predictions, and 10 times for out-of-sample predictions. The variable input model was the best twice and thrice for in-sample and out-of-sample predictions, respectively. The fixed-allocatable input model was the best thrice for in-sample and twice for out-of-sample predictions. The behavioral model outperformed the other two in 20 out of 24 cases. Meanwhile, both the variable input model and the fixed-allocatable input model are the best in five cases each. Accordingly, the results obtained from the application of the prediction accuracy measures support the choice of the behavioral model for water melon, tomato, and corn. The fixed-allocatable input model is the best for explaining the on-farm water application for squash, while the variable input model is recommended for green pepper. The estimated models can explain the farmers' short-run decisions on water allocation among competing crops. The following conclusions can be made from the estimates: The values of the estimated coefficient of determination (R') indicate that the behavioral model performs well in explaining crop-level water use in the multi-crop system for cross-sectional data. The estimated is 0.84 for watermelon, 0.88 for tomato and green pepper. 0.62 for squash and 0.93 for corn.

46


Output and water prices do not have a noticeable effect on short-run decisions on water allocation among competing crops. Crop water requirements and extension recommendations are the main determinants of water-allocation decisions. This conclusion is also supported by the fact that the coefficient of the crop-effect variable in the behavioral model is positive and highly significant. In fact, the estimated coefficients of the water price variable in the variable input model are positive and highly significant, implying a positive rather than negative response to water prices. Estimates of the water constraint variable (total water) in the fixed-allocatable input model indicate a marginal increase in farm-level water allocation for competing crops. Farmers give high priority to tomato, squash and green pepper in allocating extra water. The estimated coefficients of the water constraint increase in water available to the farm will be used for suggests that a 1 tomato first (0.365 mi), for squash second (0.301 m3) and for green pepper in the third place (0.251 m3). The least amount will be allocated for corn (0.176

Water-Use Efficiency The target production levels of wheat, faba bean, bersem, water melon, tomato, green pepper, squash and corn are the average yield levels of the sample farms as reported in Table 13. To obtain the required amount of water to produce these average yield levels, the estimated crop water-use equations were used. This was done by calculating the amount of water required for each crop at the mean levels of the independent variables appearing in that equation. For winter cropping, the fixed-allocatable input model was used in estimating the amount of water required for wheat, faba bean and bersem. For summer cropping, the behavioral model was used in calculating the water required for watermelon, tomato and corn. The fixed-allocatable input model was used for calculating the water required for squash and the variable input model was used in estimating water required for green pepper. The calculated levels of required water are presented in Table 13 and compared with the actual amount of water used. FWUE was the highest for bersem (76%), green pepper (74%) and corn (74%), indicating that actual water use exceeded water requirements by about 24.26%. The lowest FWUE of 47% for squash suggests that producers over-irrigated this crop by 53% compared to its requirements. Therefore, any improvement in the FWUE of this crop will save a large amount of scarce water that can be used to expand the farm's irrigated area or for other crops. Either below-average yields or inefficient use of irrigation water can explain the low-ratio estimates of FWUE for squash and faba bean.

On-Farm Water-Use Efficieny in Nubaria, Egypt

47

Table 12. Performance of estimated models in predicting on-farm water use in Nubaria, for summer cropping. Crop

Prediction Fixed-allocatable Accuracy input model Measure Within Out of sample sample Water melon MAE 594.4 3405.5 733 5492.8 RMSE MAPE 0.343 2.14 0 Tomato MAE 640.1 869.7 816.4 1112.1 RMSE MAPE 0.278 0.368 Pepper MAE 1691.4 2063.9 RMSE 1919.9 2444.0 MAPE 0.556 0.654 Squash MAE 945.6' RMSE 1320.9 MAPE Corn MAE 1148.1 2232.4 RMSE 1361.3 2449.3 MAPE 0.375 0.753

Variable input model Within Out of sample sample 558.6 682.7 700.8 892.8 0.299

513.7 589.4 0.319 750.5 1020.0 0.336


0.924

0.272'

507.1 0.230'

818.0 1147.7 1387.2' 0.252 1099.1 1261.5 1434.1 1428.9 0.688 1.46

920.8 1248.2 0.643

921.2 1568.1 1048.6 1685.6 0.326 0.604

503.9'

1080 1526.5 0.290 1191.9 1.39

For an out-of-sample prediction, the observations were randomly divided into two subsets. one with 80% of the observations and another with the remaining 20% of the observations. The 80% subset was used to estimate each model's parameters. These parameters were applied to the 20% subset to make out-of-sample predictions and to apply the prediction accuracy measures. Indicates the model that most accurately predicts short-term water use for a given accuracy and experiment,

The study shows that farmers in Nubaria allocate a large amount of water in excess of the crop requirements for all winter and summer crops. They over-irrigate their crops by 24 to 53%, depending on the crop, compared to the required amount of water to produce the achieved yield levels (see Table 13). These figures suggest that a big technology gap exists between the required irrigation practices for wheat, faba bean, bersem, water melon, tomato, green pepper, squash and corn, and the actual water application in the study area.

48

Assessing On-Farm Water- Use Effciency

Table 13. FWUE for major crops in Nubaria, Egypt. Crop

Irrigated area (Fed) Winter croppiog Wheat 2.16 Faba 1.73 bean Bersem 1.23 Summer cropping Water 2.30 melon Tomato 0.81 Pepper 0.57 Squash 1.54 Corn 064

Av. yield (ton/Fed)

Actual water Req. water use

WP( Total Irrig.

%

2 90 0.663

3368 2807

2181 1554

0.86 1.24 0.24 0 39

65 55

32.68

4541

3452

7.24

9.39

76

9 66

2740

3452

3 50

5.68

76

6.75 4.00 4.35 2.57

3373 4070 3179 4019

2334 3031 1489 2983

2.0 0.98 1.37 0 64

2 89 132 203 0.86

69 74 47 74

Including ramwater of 1040

This has important policy implications in that improving FWUE for these crops can contribute to the overall FWUE in the study area. In this study, the overall FWUE for winter cropping is 0.65 and for summer cropping is 0.61, suggesting a high potential for saving water once FWUE is improved.

On-Farm Water-Use Efficienyin Nubaria, Egypt

49

On-Farm Water-Use Efficiency in Beni Sweif, Egypt Characteristics of the Beni Sweif farms In the Beni Sweif area of Egypt, 50 farms were sampled for the study. Farmers' irrigation experience ranged from 1 to 53 years, with an average of 33 years. The major crops grown were wheat and bersem for winter cropping, and cotton, sunflower, tomato and corn for summer cropping. The soil type was medium and heavy for 66 and 30% of the sample farms, respectively. Most farms (78%) had deep soil, while the others (22%) had medium soils. Furthermore, 90% of the farmers reported that soil salinity was low, and 10% reported that it was medium. The cropping pattern, as indicated by 84% of the farmers, was mainly determined by market conditions and family consumption and on-farm use. For the others in the sample, crop rotation explained the cropping patterns. However, for 80% of the farmers the choice of crops was based on market conditions, while labor requirements of crops and other factors explained the crop choice for the remaining 20%. Location of water source relative to the farm was a major concern to the farmers because surface water was the major source of irrigation water and only 32% of the farms were at head locations, while 36% and 32% were at medium and tail locations, respectively. Most farmers (80%) were full-time operators. Farming was the main source of household income, accounting for 88% the total income. While much of this income (74%) was from crop production, livestock contributed an additional 26%. Other characteristics of the sample farms are presented in Table 14. The farmers in the sample were growing the following crops: bersem (42), cotton (25), wheat (38): corn (37), sunflower (9) and tomato (10). Other crops were produced by a negligible number of farmers. The total farm size averaged 4.85 feddan, while the average crop area was 1.54 feddan for bersem, 3.62 feddan for cotton, 1.2 feddan for sunflower. 2.31 feddan for tomato, 2.26 feddan for wheat and 1.89 feddan for corn. Among winter crops, wheat accounted for 59% of total irrigated land, while 41% was allocated for bersem. For summer cropping, more land was devoted to cotton (40%) and tomato (25%) compared to other crops.

50


The average amount of water available for each of the sample farms was 14,001 m3/feddan. Water application by crop for winter cropping was 4545 m3/feddan for bersem and 2348 m3/feddan for wheat. For summer cropping, water application, as an average for the sample farmers, was 4531 m3/feddan for cotton, 1972 feddan for sunflower, 2555 m3/feddan for tomato and 2987 m3/feddan for corn. The amount of water applied to each crop was determined by the crop's water requirements (in the case of 80% of the farmers) and the price of crops and extension recommendations (in the case of 20% of farmers).

Table 14. Basic water and crop data of sample farms in Beni Sweif, Egypt. Item

Total

Farm Number of farms 50 Experience (year) 33 Area (feddan) Mean 4.85 SD 7.80 Crop yield (ton/feddan) Mean SD Water applied (m3) 14,001 Mean SD 42,726

Crops Bersem Cotton Sunflower Tomato Wheat Corn 42 25 9 10 38 37 33 37 31 30 35 33

1.54 1.56

3.62 7.93

1.20 1.05

2.31 1.74

2.26 4.15

1.89 1.44

54.70 53.16

5.90 1.68

1.37 0.64

10.26 4.89

2.32 0.57

2.14 0.73

4546 858

4531 717

1972 282

2555 413

2348 275

2988 646

Water productivity, defined in technical terms as kg of output per m3 of water, was highest for bersem (9.79 in winter cropping and tomato (2.85 kg/m3) in summer cropping. Therefore, water yielded more output in bersem production, compared to wheat in winter cropping. Each additional m3 of water yielded 9.79 kg of bersem output, whereas the output of wheat was much lower for an additional For summer cropping, water gave more output in unit of water (0.68 tomato production, compared to cotton, sunflower and corn. When the amount of rain water was not included in the calculation of the amount of water applied, water productivity of irrigation water increased considerably.

Empirical Results Following the methodology for determining on-farm water use (explained earlier), the fixed-allocatable input model, variable input model and behavioral model were estimated using on-farm data from the 50 sample farms. The models were then compared using prediction accuracy measures for validation. Both in-sample and out-of-sample forecasts for crop-level water use were made and compared with the

On-Farm Water-Use Efficieny in Beni Sweif; Egypt

51

actual on-farm water applications. Three measures of prediction accuracy (MAE, RMSE and MAPE) were calculated to compare the performance of the altemative models in predicting on-farm water use for each crop. Applying the three measures to each of the two predictions (in and out of sample) generated six cases for evaluating the altemative models for each crop. Summarized calculations are presented in Table 15. Table 15. Performance of estimated models in predicting on-farm water-use in Beni Sweif. -

Crop

I

Prediction Fixed-allocatable variable-input Accuracy input model model Measure Within Out of Within Out of sample sample sample sample


Wheat MAE RMSE MAPE

I Cotton

MAE RMSE MAPE

702 1218 0.329 941 1117 0.216

MAE RMSE MAPE

I

MAE RMSE MAPE Sunflower MAE RMSE MAPE Tomato MAE RMSE

MAPE

1845 2698 0.495

329 508 0.138

462'

279 316 0.150 1239'

301 328 0.160

498 646 0.208

732 0.335

887 1155 0.242

919

0.377

4423 9597 1.10

1900 2069 0.411

5140 11258 1.28

600 954 0.197

605 669 0.212

800 1006 0.252

1228 1527 0.41

0.

0.1

1012 1085 0.509

1136 1028 0.526

1139 1273 0.441

1171 1371 0.447

1251 1379 0.489

1302 1466 0.505

0.216 1765

0.697"

689" 1123

277b

I

0.211'

For an out-of-sample prediction. the observations were randomly divided into two subsets, one with 80% of the observations and another with the remaining 20% of the observations The 80% subset was used to estimate each model's parameters. These parameters were applied to the 20% subset to make out-of-sample predictions and to apply the prediction accuracy measures (Moore et al., 1994a and 1994b). Indicates the model that most accurately predicts short-tenn water use for a given accuracy and experiment.

I

52

On-Farm Water-Use Efficiency

Winter Cropping With the in-sample prediction, the fixed-allocatable input model outperformed the other two models for wheat according for the three prediction accuracy measures (Table 15). while both the variable input and the behavioral models performed best for bersem. For out-of-sample predictions, the variable input model performed best for wheat according to the MAE and MAPE measures of prediction accuracy. The behavioral model outperformed the other two models for bersem according to the three measures of prediction accuracy. The overall prediction performance of the estimated models can be judged on both in-sample and out-of-sample predictions. Regardless of the type of accuracy measures, the fried-allocatable input model was best in three cases for in-sample predictions. This model was, however, dominated by the other two models for out-of-sample predictions. The variable input model was best twice for out-of-sample predictions, and twice for in-sample predictions. The behavioral model was best twice for in-sample predictions and four times for out-of-sample predictions. The overall performance of the three models, regardless of the type of predictions and accuracy measures, showed that the fixed-allocatable input, the behavioral and variable input models outperformed 3, 6, and 4 of the 12 times, respectively. Thus, the results obtained from the application of the prediction accuracy measures are mixed. However, it can be concluded that both the fixed-allocatable input model and the variable input model should be used for calculating the required amount of water application for wheat, and both the behavioral model and the variable input model should be used to calculate the required amount of water application for bersem. The estimated parameters of the three models can help to explain the farmers' short-run decisions on water allocation among competing crops. These estimates reveal the following points: Own- and cross-crop area and prices appear to be the most important variables in explaining the farmers' water-use decisions in irrigating wheat and bersem. The estimated coefficients of area and/or price variables have the correct signs and highly significant in the three estimated models. Cross-acreage variables in the fixed-allocatable input model support the multi-crop jointness in the study area, implying a high degree of competition between wheat and bersem for inputs.

On-Farm Water-Use Efficieny in Beni Sweif; Egypt

53

Total water available to the farm was not a limiting factor in explaining the water use for wheat and bersem. The estimated coefficients of this variable in the fixed-allocatable input model were negative and not significant, implying that enough water was available for winter cropping. Thus, farmers did not perceive water as a fixed input in the short run. Water price, represented by operational costs of water charged by the public authority, was not negative in the water demand equations for wheat and bersem with the variable input model. This shows that after planting, farmers did not respond to water prices in subsequent short-run decisions, implying that the water price does not have a noticeable quantitative impact on water allocation. This result is further supported by the fact that the amount of water applied to each crop was solely determined by crop-water requirements and extension recommendations as indicated by the farmers.

Summer cropping Alternative water-use models were compared using the three prediction performance measures. With the in-sample predictions, the fixed-allocatable input model outperformed the other models, according to the three prediction performance measures, for cotton, corn and tomato. The variable input model performed best for in-sample predictions of sunflower, according to the same accuracy measures. The same performance of alternative models applies to the out-of-sample predictions, where the fixed-allocatable input model was the best for cotton, corn and tomato, and the variable input model was the best for sunflower (Table 15). The results obtained from the application of the prediction accuracy measures are very conclusive in the choice of the fixed-allocatable input model to calculate the required amount of water for cotton, corn and tomato. Similarly, the results support the choice of the variable input model to estimate the required amount of water for sunflower. To better explain the farmers' short-run decisions on water allocation among competing crops, the estimated models can provide further insights. These estimates support the following points: Own-and cross-crop area and prices appear to be the most important variables in explaining the farmers' water-use decisions in irrigating cotton, corn, sunflo er and tomato. The estimated coefficients of area and crop price variables had the correct signs and were highly significant in the three models. Cross-acreage and cross-price variables in the fixed-allocatable input model clearly support the multi-crop interdependence in the study area, implying a high degree of competition among summer crops for the available amount of water.

54

Assessing On-Farm Water-Use Efficiency Total water available. to the farm was one of the most important factors explaining the water use for cotton, corn and tomato. The estimated coefficients of this variable in the fixed-allocatable input model provided important information on water allocation among competing crops in a multi-crop system. was allocated to cotton in the first An increase in water availability by 1 place (0.62 m3), a water-consuming crop, and corn in the second place (0.257 m3). The least amount of water was allocated for tomato (0.083 m3). Sunflower was a residual crop in terms of additional amount of water. The water constraint variable is positive and highly significant in the water-use equations of cotton, corn and tomato. This result suggests that farmers perceive water as a fixed input in the short run.This result is further supported by the fact that the water price is not significant in the water demand equations of the three crops with the variable input model. This implies that after planting crops, producers do not respond to water prices in subsequent short-run decisions, implying that the water price does not have an impact on water allocation. This result is further supported by the fact that the amount of water applied to each crop is solely determined by crop water requirements and extension recommendations, as indicated by the sample farmers.

Water-Use Efficiency The target production levels of wheat, bersem, cotton, corn, sunflower and tomato are the average yield levels of the sample f a m s as reported in Table 16. To obtain the required amount of water to produce these average yield levels, the estimated crop water-use equations were used. This was done by calculating the amount of water required for each crop at the mean levels of the independent variables appearing in that equation. For winter cropping, both the fixed-allocatable input model and the variable input model were used in estimating the amount of water required for wheat, while the behavioral and the variable input models were used to calculate the amount of water required for bersem. For summer cropping, the fixed-allocatable input model was used to calculate the amount of water required for cotton, corn and tomato. The variable input model was used to calculate the amount of water required for sunflower. The calculated levels of required water are presented in Table 16 and compared with the actual amount of water used. FWUE is highest for cotton (75%), bersem and corn (72% each), indicating that actual water use exceeded water requirements by about 25 to 28%. The lowest FWUE of 56% for tomato suggests that farmers over-irrigated this crop by 44% compared to its requirements. Therefore, any improvement in the FWUE of this crop could save a large amount of water that can be used to expand the farm's irrigated area or for other crops. Likewise, farmers of

On-Farm Water-Use Efficieny in Beni Sweif, Egypt

55

wheat and sunflower exceeded crop water requirements by 35% Either below average yields or inefficient use of irrigation water can explain these low ratios of FWUE for tomato, wheat and sunflower. Table 16. WUE for crops in Beni Sweif, Egypt.

I

Crop

Irrigated area (Feddan)

Winter croppi Wheat Bersem 1.54 Summer cropping Cotton 3.62 1.89 Sunflower 1.20 Tomato 2.31

Yield Actual (ton/fed) use (m3)

Req. water (m3)

WP

FWUE % Total Irrigation

2.32 54.7

3388 5586

2214 3998

9.79 1.06

12.03 1.30

65 72

5.90 2.14 1.37 10.26

5571 4027 3012 3595

4190 2898 1940 2021

0.45 2.85 0.68 0.53

0.69 4.02 0.99 0.72

75 72 64 56

Includes rainfall water quantity of 104 m m annual

Farmers in the Beni Sweif area of Egypt over-irrigated all winter and summer crops by 25 to 44%, depending on the crop under consideration, compared to the required amount of water to produce the achieved yield levels (Table16). These figures suggest that a big technology gap exists between the required irrigation practices for wheat, bersem, cotton, corn, sunflower and tomato, and the actual water application in the study area. This has important policy implications in that improving FWUE for these crops can contribute to the overall FWUE in the study area. In the Beni-Sweif area the overall FWUE for winter and summer cropping is 68%, indicating a high potential for water savings once FWUE is improved.

56

On-FarmWater-Use Efficiency

On-Farm Water-Use Efficiency in Wheat Production, Iraq Wheat is the most important cereal crop in Iraq, grown in both rainfed and irrigation conditions. During the 1970s the total wheat area peaked to 1.51 million hectares, with rainfed area and production reaching 1.284 million hectares and 795.5 thousand tons, respectively. Total wheat production was highest in the 1990s, reaching 1.2 million tons. During this period, wheat production under irrigated conditions reached its highest level of 742.4 thousand tons. The highest average national wheat yield was 0.794 ton/ha during the 1990s. In the 1970s, the highest national yield was 1.576 ton/ha for irrigated farming, and for rainfed farming it peaked during the 1980s, reaching 0 614 ton/ha. During the period 1970-2000, national average wheat yield did not exceed 0.747 ton/ha although the average yield under irrigated firming was 1.258 ton/ha. This is mainly attributed to the low yield level (0.597 ton/ha) of rainfed farming. The contribution of rainfed farming to total wheat production was highest during the 1970s and 1980s, accounting for 85% of the total wheat area and 70-75% of the total production during the two decades. However, the contribution of irrigated farming has increased since then, accounting for 47% of the total wheat area and 63% of the total production during the 1990s. On average, the contribution of rainfed farming to total wheat production during 1970-2000 was 58%, also accounting for 74% the total area planted with wheat. Wheat production in Iraq experienced sharp seasonal fluctuations during the last three decades caused by volatile weather conditions that affect yield in rainfed farming systems. To increase wheat production and stabilize its yield, supplemental irrigation (SI) has been introduced to rainfed farming in the country. In 1991, an S1 project, the North Al-Jazeera Irrigation Project, was launched in order to serve approximately 60,000 hectares of arable land, using a linear-move sprinkler irrigation system. The project is fed by the river Tigris via a large submerged pumping station located inside the reservoir of a large dam. Winter crops under SI occupy about 80% of the project's cultivable land. In the summer, however, the percentage cropped area is limited to only 30%. Wheat is the major winter crop, covering 73% of the project area. Other winter crops are barley, lentil, and potato, which cover 3, 1.5, and 2.5% of the area, respectively. During the summer, tomato is the major crop, occupying 22% of the project area. Other summer crops include spring potato (5.5%), sugar beet (I %), and vegetables (1.5%).

A development project on "Dissemination of Improved Irrigation Technologies" was introduced into the farming systems of Iraq, particularly in rainfed areas, aimed at planting up to 0.5 million hectares of wheat under SI by 2007. Currently, there are about 3,500 new farms in the Mosul province under SI with an average size of 25 hectares.

Data and Wheat Yield of The Sample Farms Data was collected from a cross-sectional survey of 284 farms in 20 districts in the Ninavah province in northern Iraq during the 2001-2002 season. A stratified random sampling technique was used to ensure that there was no overlap between farmers in one district with those in other districts. A proportional allocation procedure was applied to determine the sample size for each district (Scheafer et al., 1979). In selecting the sample, there was consideration for the type (center-pivot sprinkler system and solid-set sprinkler system) and origin (country where the irrigation technology was manufactured and/or imported) of irrigation technologies used by farmers. All the farmers in the sample used supplemental irrigation in wheat production. Since some farmers grow more than one variety in their fields, each variety was taken as a separate observation, thus a farmer who planted hvo varieties was counted to be two observations. Accordingly, the total number of observations was 318. Among the sample farms, 148 farmers (46.5% of the sample) were using center-pivot sprinkler systems, which are mainly imported from Saudi Arabia, and 170 farmers (53.5% of the sample) were using the solid-set sprinkler system. The farmers using center-pivot sprinkler systems produced an average yield of 3.08 ton/ha, while who used solid-set sprinklers produced an average yield of 2.91 ton/ha. Variety is an important source of yield variation among wheat producers. Data in Table 17 shows that improved varieties ('Waha,' 'Um Rabia,' 'IPA 99', 'Sham 3') gave higher yield levels than the local variety, 'Abu-Ghraib'. This has an important implication for increasing the average yield level under SI as majority of farmers grew high-yielding improved wheat varieties. The other important consideration is that durum wheat varieties gave a higher yield level of 3.1 toniha compared to bread wheat varieties which gave 2.8 ton/ha. The improved variety, 'Um Rabia,' gave the highest yield level of 3.4 ton/ha among all improved varieties. This variety was planted by nearly 13% of the sampled farmers.

58


Table 17. Yield of sample farms according to varieties. -

Variety

Type

'Waha' 'Abu-Ghraib' 'Um Rabia' IPA 99 'Sham 3' Other varieties All bread varieties All durum varieties All varieties

Durum wheat Bread wheat Bread wheat Bread wheat Durum wheat Durum/bread

Average yield

Farmers

3.01 2.42 3.40 3.05 2.84 2.94 2.77 3.08 2.92

24.2 21.4 12.6 7.9 5.4 28.6 52.0 48.0 100

-----ton/ha-----

---- %----

Source: Calculated from farm survey data of the study

Empirical Results Impact of supplemenutal irrigation on wheat yield and water productivity SI, like any other new technology, may change the optimal levels of inputs used, and thus the productivity. The profitability of adopting new irrigation technologies depends on the level of productivity improvement (Lin, 1994). Thus, information on the effect of new irrigation technologies on productivity is crucial for potential diffusion because a new technology will not be accepted by farmers unless it raises productivity. Table 18 shows the extent of productivity increase due to the introduction of improved irrigation technology. The results indicate that wheat yield under SI was 2.93 ton/ha, higher than that under rainfed conditions which was 1.28 ton/ha. In other words, the total impact of SI was an increase in wheat yield by 128% as compared to rainfed farming. It is important to note that SI had more impact on increasing the yield of bread than durum wheat varieties. SI increased the yield of bread wheat varieties by more than 100%, whereas the contribution to the yield of durum wheat varieties ranged between 58 and 81%. This may be attributed to the biological characteristics of the varieties or the scheduling of irrigation water.

017-Farm Water-Use Efficieny in Wheat Production. Iraq

59

Table 18. Impact of SI on wheat productivity. Variety

Yield under SI Yield under rainfed ton/ha -------------...---------'Waha' 3.01 1.66 'Abu-Ghraib' 2.42 1.20 'Urn Rabia' 3.38 1.64 'IPA 99' 3.05 1.27 1.80 'Sham 3' 2.84 Other varieties 2.94 1.60 All varieties 2.93 1.28

Increase Per cent 81 102 106 140 58 84 128

I

I

Source: Calculated from farm survey data of the study.

Policy makers are interested in knowing the impact of SI on water productivity. The data in Table 19 compares water productivity under SI and rainfed conditions. The results show that SI increased water productivity for all varieties except 'Sham 3'. Water productivity increased the most (63%) for 'IPA 99.' Water productivity increased by 32 and 15% as a result of using SI for bread wheat and dumm wheat varieties, respectively. Thus, the impact of S1 on water productivity for bread wheat was higher than for dumm wheat varieties. On average, the impact of SI was an increase in water productivity by 31% for all wheat varieties. SI greatly increased land and water productivity; however, in the production of wheat under SI, more inputs (such as fertilizers) were used as compared to rainfed conditions. Therefore, it is difficult to decide whether the land- and water-yield advantages are totally attributable to SI and associated technologies. Table 19. Impact of S1 on water productivity. Variety

Water productivity Rainfed SI

Increase

-per cent. 'Waha' 'Abu-Ghraib' 'Urn Rabia' IPA 99 'Sham 3' All dumm varieties All bread varieties All varieties Source: Calculated from farm survey data of the study.

60


Productivity (output per unit of input) is usually measured by index numbers. However, the index numbers impose restrictions on the form of the underlying production function. The Laspeyres index, for example, assumes that the production function is linear, implying perfect substitution among all inputs in the production process. To avoid this restrictive assumption, Ball (1985) derived revised indices for productivity from a flexible multi-output multi-input representation of the production function of the form of the translog transformation function. However, the translog function is constrained by constant returns to scale. This is a restrictive assumption which lacks empirical evidence to be imposed as a priori on wheat production in northern Iraq. Moreover, the multi-colinearity problem among variable inputs is always evident in the translog function. As a result, it gives lower values for the estimates of t-test for many variables in the function. These two difficulties may discourage the use of the index number approach for studying the impact of SI technology on total factor productivity. Instead a frequently used method for estimation is the Cobb-Douglas function approach due to its ease of estimation and interpretation (Lin, 1994). An appropriate technique for assessing the impact of SI on productivity is the regression analysis. The impact of SI on total factor productivity is estimated by adding a dummy variable to the function. The production function estimated here has the following form:

Where Q is total wheat production for each farm (tons) are the parameters to be estimated X, is total planted area per farm (ha) is total amount of seed per farm (tons) X, is total amount of pesticides per farm (tons) is total amount of fertilizers per farm (tons) is total Machinery per farm (hours) is total labor per farm (hours) is a dummy variable for 'Waha' variety, taking a value of 1 for 'Waha' and zero otherwise is a dummy variable for 'IPA 99' variety, taking a value of 1 for 'IPA 99' and zero otherwise is a dummy variable for 'Um Rabia' variety, taking a value of 1 for 'Um Rabia' and zero otherwise is a dummy variable for 'Sham 3' variety, taking a value of for 'Sham 3' and zero otherwise

On-Farm Water-Use Efficieny

in Wheat Production Iraq

61

is a dummy variable for other varieties, taking a value of 1 for other varieties and zero otherwise is a dummy variable for SI, taking a value of 1 if the farmers applied water and zero otherwise V is a residual term to capture the effect of other variables not directly included in the model The 'Abu-Ghraib' variety takes a zero value as a base local variety to measure the impact of improved varieties on wheat productivity under SI. The production function was estimated by using the OLS procedure. The estimated coefficients are presented in Table 20. Table 20. Estimated coefficients of the wheat production function. Independent variables

Function (I): Function (2): before correction after correction Intercept -1.625 (-1.89) -0.867 (-2.66) ** Ln Land (ha) 0.247 (1.28) 0.422 (2.53) ** Ln Seed (kg) 0.167 (0.94) 0.251 (1.96) * Ln Pesticides (kg) 0.025 (1.62) 0.038 (2.77) ** Ln Fertilizers (kg) 0.035 (3.09) ** 0.002 (0.26) 0.537 (4.54) ** 0.299 (2.40) ** Ln Machinery (hi) Ln Labor (hr) -0.028 (-0.53) 0.029 (0.56) 0.189 (2.94) ** 0.165 (2.86) ** 'Waha' dummy variable (0,1) 'IPA 99' dummy variable (0,l) 0.128(1.45) 0.189 (2.56) ** 'Um Rabia' dummy variable (0,l) 0.361 (4.85) ** 0.241 (3.56) ** 'Sham' dummy variable (0,1) 0.232 (2.43) ** 0.143 (1.72) 0.185 (3.21) ** (I) the varieties dummy (0,1) 0.150 (3.04) ** SI dummy variable (0,l) 0.653 (11.64) ** 0.646 (8.84) ** R-Square 0.83 0.85 Adjusted R-Square 0.83 0.85 D-W test 1.03 2.03 F-test 195.07** 233.05** Note: Numbers in parentheses refer to the calculated t-test. *, ** Significant at 5 and 1% levels of significance, respectively SI= Supplemental irrigation

The value of the coefficient of multiple determination (R-Square) indicates that the explanatory variables included in the model explain 85% of total variation in wheat production. The SI variable has more effect on wheat production, and its estimated coefficient is significant at 1% level of significance. Other variables of significant influence on wheat production are planted area: seed, machinery and improved varieties.

62


Among the potential econometric problems of multi-collinearity, heteroscedasticity, and autocorrelation, only the latter was found to be of harmful nature. Thus, the original data was corrected for the problem of autocorrelation by using the "generalized difference" approach and then by using the corrected data the production function was re-estimated. The approach involves the regressing of the dependent variable on the independent variables in the difference form, which is obtained by subtracting a coefficient of autocorrelation of the value of a variable in the previous time period from its value in the current time period (Gujarati, 1988). The estimated coefficients of the dummy variables measure the shift in the intercept of the production function due to the corresponding improved varieties or SI. The estimated coefficient of the SI dummy variable of 0.646 (function 2) indicates that wheat productivity under SI is 91% higher than that of rainfed conditions. That is, given the same level of inputs, the yield advantage of the use of SI over rainfed farming is about 91%. This is the magnitude of the upward shift in the wheat production function resulting from the use of SI. Similarly, the estimated coefficients of the dummy variables of improved variables measure the net impact on wheat yield as compared to the local variety, 'Abu-Ghraib.' The coefficient of 'Um Rabia' dummy variable of 0.241, for example, implies that the yield advantage of 'Um Rabia' variety over the local variety is 27% under SI. In other words, under the same input levels, the improved variety 'Um Rabia' increases wheat productivity by 27% as compared to the local variety, 'AbuGhraib,' supporting the conclusion that the use of improved varieties results in higher wheat productivity as compared to the local varieties under SI. Similar conclusions can be drawn from the estimated coefficients of the dummy variables of other improved varieties. Yield comparison of wheat producers under SI and rainfed conditions reveals that SI increases wheat productivity by 128% as compared to rainfed farming. This total impact of SI on wheat productivity reflects the combined effect of other variables as well. The production function, however, gives a net impact of increasing wheat yield by 91% as a result of using the S1 technology. The net impact is totally attributed to the use of SI. The difference of 37% between the total impact and net impact is attributed to other variables not used under rainfed conditions.

On-Farm Water-Use Efficieny

in

Wheat Production, Iraq

63

Model Estimation and Empirical Results All data was collected from the farm survey conducted in the Ninavah province in Iraq during the 2001 -2002 season. It included information on wheat production, input use, output and input price variables and water management practices. The survey also included questions on crop-level acreage. irrigation technology, soil information, water sources, rainfall, annual water budget and on-farm irrigation practices. Collected information indicated that all farmers grew wheat as their only or principal crop for their Livelihoods. Of the farmers in the sample, only 3 grew barley, 12 potato, 3 sunflower, and 4 sugar beet. Accordingly, water allocation equations were estimated only for wheat, as there was not enough number of observations to study water allocation decisions for other crops. Several quantitative and qualitative independent variables were formed from the survey data. The quantitative variables included the irrigated wheat area (ha), price of wheat (ID/kg), amount of total water available to the farm (m), amount of rainfall (mm), prices of variable inputs-such as labor wage (ID/hour), and water cost (ID/hectare/year). Water cost was calculated after taking into account all direct and indirect costs related to water application. These costs included fuel cost, repairs and maintenance, labor wages (both family and hired labor), and annual depreciation. The annual depreciation was calculated using the fixed approach (of 10 years) on the costs of irrigation systems, ponds, well-digging, pumping machines, transportation costs and installation. An annual interest rate of 8% was applied. The set of qualitative variables included dummy variables on water limitation (taking a value of 1 if the amount of water available to the farm was limited, and zero if it was noi), irrigation technology (taking a value of 1 if the irrigation technology was center-pivot sprinkler system and zero for solid-set sprinkler system), land tenure (taking a value of 1 if the land was privately owned and zero if it was not), and water source (taking a value of 1 if the water source was groundwater and zero if it was not). Having specified and quantified related variables, the second step involved the estimation of models following methodology for on-farm water-use. The three specified models of the fixed-allocatable input model, variable input model and the behavioral model were estimated using the ordinary least squares procedure (ESCWA and ICARDA, 2000 and 200 1). Three sets of out-of-sample forecasts were made. For an out-of-sample prediction, the observations were randomly divided into two subsets, one with 80% of the observations and another with 20%.

64


The 80% subset was used to estimate each model's parameters, which were applied to the 20% subset to make out-of-sample predictions and to apply the prediction performance measures. This procedure was repeated three times (Moore et al., 1994a and 1994b). The estimated models are presented in Appendix F, Tables 1 to 3. The estimates provide some insights on the selection of an appropriate water-use model. The estimated water-use models are actually the derived factor demand functions for water. This has important implications on the expected signs of estimated coefficients of price variables: the relationship between irrigation water and its price is negative, while that between the quantity of water demanded and the output price (which is the price of wheat in this case) is positive. Accordingly, the consistency of the signs of the estimated coefficients with the economic theory is a prerequisite for model selection. In the fixed-allocatable input model all estimated coefficients had the correct signs, supporting their plausibility and consistency with economic theory and a priori information. Irrigation technology, total quantity of water available to the farm, irrigated wheat area and labor wage rate appeared to be strong and significant determinants of farmers' decisions on the amount of water applied to wheat. Although the wheat price had the correct positive sign, its effect on the amount of water applied was not significant. Similarly, water limitation had the correct negative sign, implying that when the amount of water available to the farm is limited, farmers reduce the amount of water for wheat. The relative performance of the water constraint variable (represented by the variable of total water available to the farm) provides additional support to the choice of the fixed-allocatable input model. With this model, the water constraint variable was positive and highly significant in the water demand equation of wheat. This result suggests that producers perceive irrigation water as a fixed input in the short run. In contrast, for the variable input model, the water cost (as a proxy variable for water price or water valuation) was not negative in the wheat demand equation for water. The water price (water cost) variable actually had the incorrect (positive) sign in the variable input model, indicating that after planting the crop, producers do not respond to water price in subsequent short-run decisions. This result suggests that the major impact on the amount of water applied to wheat under the variable input model is related to irrigation technology and landallocation decisions, as both irrigation technology and wheat irrigated area appeared to he the only two determinants of farmers' decisions on the amount of water applied to wheat. Once cropland is allocated, the level of water price appears

not to have a major quantitative impact on profit; otherwise, the water price variable would be a significant determinant of short-run water-use. This conclusion is further supported by the fact that water in the study area was available to farmers free of charge. thus, water prices had no major influence on the amount of water allocated to wheat. The behavioral model, on the other hand, provided good information on the water-use decisions according to the estimated coefficients. With this model. irrigation technology, acreage allocation, land tenure and rainfall had the correct signs, thus they significantly influenced the producers' decisions on the amount of water applied to wheat. Farmers of privately owned land use more irrigation water for wheat compared to other types of land tenure. In line with common practices, farmers reduce the amount of water applied when the rainfall increases. In comparing the fixed-allocatable input model with the behavioral model, the former provides better goodness of fit (which is measured by the coefficient of determination) than the latter. The estimated coefficient of determination (R-Square) of the fixed-allocatable input model of 0.85 was much superior to that of the behavioral model of 0.60, justifying the selection of the fixed-allocatable input model. In conclusion, the fixed-allocatable input model explains producer decisions on short-run water use better than the variable input model (Moore et a/.. 1994a and 1994b). It also handles more complex multi-crop decisions than the behavioral model. In the fixed-allocatable input model, producers operate with a short-run constraint on farm-level water use because of the fixed amount of surface water available to the farm, according to the rationing system of water allocation among farmers. Meanwhile, groundwater is subject to fixed pumping capacity in the short run. The three altemative models were compared using prediction accuracy measures (MAE, RMSE and MAPE) for model validation. Four sets of forecasts were made, including one in-sample prediction and three out-of-sample predictions. Out-of-sample forecasts for wheat water use were made and compared with the actual on-farm water applications. A detailed presentation of the calculated measures is shown in Appendix F, Table 4. Applying the three measures to each one of the one in-sample prediction and the three out-ofisample predictions generates twelve cases for evaluating the alternative models for wheat and provides evidence on a model choice. With the in-sample prediction, the fixed-allocatable input model outperformed the two alternative models (the variable input model and the behavioral model) according

66

Assessing On-Farm Water- Use Effciency

to each of the three measures of prediction accuracy. The results of out-of-sample predictions demonstrate a similar pattern of performance, whereby the fixedallocatable input model outperformed the other two according to the three measures of prediction performance (MAE, RMSE and MAPE). The numerical values of MAE, RMSE and MAPE of the fixed-allocatable input model are considerably lower than those of the variable input model and those of the behavioral model for both in-sample and out-of-sample predictions. Accordingly, the results obtained from the application of the prediction accuracy measures support the conclusion that the fixed-allocatable input model is better for explaining short-run water allocation by wheat producers than the variable input and the behavioral models. We can further conclude that the fixed-allocatable input model is a better model to study on-farm water use by wheat producers. However, the model needs additional description. The estimates reveal the following points: The value of the coefficient of multiple determination (R-Square) indicates that the model performs well in explaining crop-level water use for cross-sectional data. The estimated R-Square is 0.85, implying that nearly 85% of the variation in the amount of water applied to wheat is explained by the independent variables included in the estimated equation. The estimated water demand equation for wheat under the fixed-allocatable input model is significant at 1% level. Labor wage rate appears to be a strong determinant for short-run decisions on water use for wheat. An increase in the labor wage rate by one Iraqi Dinar, holding other variables constant, will decrease water use for wheat by 25.54 m3. This result has important implications for wheat producers as the labor demand is highly inelastic under rainfed conditions, and thus a small increase in the demand for labor will increase the labor wage rate considerably which in return will reduce the amount of water for wheat. The estimate of the water constraint variable (total amount of water available to the farm) indicates the allocation of a marginal increase in fann-level water availability for wheat producers. The individual estimated coefficient of thewater constraint of 0.67 shows that for a one increase in the availability 3 of water to the farm, 0.67 m of it will be used mainly for wheat production. This implies that almost two-thirds of the incremental water is added to wheat production. Irrigation technology has an important and a significant impact on the amount of water-use in wheat production. The use of center-pivot sprinkler technology reduces the amount of water-use by 4934.53 m3 per farm compared to the use of solid-set sprinkler technology. The estimated coefficient of the variable of the irrigation technology is negative and significant at the 1% level of significance. Similarly, if farmers perceive water available to the farm as limited, they will

On-Farm Water-Use Efficieny in Wheaf Production, Iraq

67

reduce the use of water for wheat production by 824 per farm. Planted area under supplemental irrigation is a strong and highly significant determinant of the amount of water applied for wheat production. The estimated coefficient o f irrigated area implies that expanding the wheat area by one hectare will result in increasing water use by 745.74 m3.

On-Farm Water-Use Efficiency For the purpose of this study, on-farm water-use efficiency (FWUE) is defined as the ratio of the required amount of water to produce a target production level to the actual amount of water used (including the amount of rainfall). The target production level for wheat is the yield of the sample farms. To obtain the required amount of water to produce the average yield levels, the estimated water demand equation with the fixed-allocatable input model was used. This was done by calculating the amount of water required for each individual farmer at the levels of the independent variables appearing in the demand equation. The calculated levels of required water were compared with the actual amount of water used for each individual farmer. The results are summarized in Table 21. Of the total wheat farmers, 80% had FWUE of less than one, indicating that these farmers over-irrigated their wheat crop. The FWUE of 20% of the farmers was greater than one, implying that these farmers under-irrigated wheat by 10%. The overall FWUE for the whole sample was 0.77, which is less than one, indicating that the wheat producers over-irrigated their crop by 23% Table 21. FWUE in wheat production in Ninavah province.

Level of FWUE 1.0 Total

Percentage of farmers (%) 4 20 56 20 100

Average FWUE 0.34 0.64 0.87 1.10 0.77

Source: Farm survey data plus results of the fixed-allocatable input model

Among over-irrigating farmers, the FWUE of 4% of the farmers was estimated at 0.34, indicating that actual water use exceeded required water use by about 66%. The FWUE of 20% of the farmers was 0.64, implying that this group over-irrigated its crop by 36%. Meanwhile, 56% of the farmers having FWUE of 0.87 over-irrigated the wheat crop by 13%.

68


Table 22. Effect of farm size on the efficiency of water-use in wheat production in Ninavah province. Farm size, (ha) < 10 10.1 - 20 > 20 Total

Average farm size (ha) 6.7 15.5 29.1 13.8

Farmers

(%) 51 25 24 100

Average FWUE 0.77 0.81 0.72 0.77

Source: Farm survey data plus results of the fixed-allocatable input model

Farmers over-irrigate wheat because of their own perceptions of water requirements and their expectations of rainfall and market conditions. Farm size is an important factor in explaining the variation in FWUE among wheat producers. To demonstrate the impact of farm size on the efficiency of irrigation water-use, the results were re-tabulated in Table 22. More than 50% of wheat producers were small holders of less than 10 hectares with an average farm size of 6.7 hectares. The FWUE for small holders was 0.77, implying that small farmers over-irrigated wheat production by 23%. The FWUE improved with bigger farm size. The FWUE of medium farms (10.1 - 20 hectares) was 0.81. With farm size above 20 hectares the FWUE decreased. The FWUE of large farms was 0.72, indicating that large farms exceeded water requirements by 28%.

Conclusions and Recommendations

69

Conclusions and Recommendations The methodology that was developed, and widely tested, for assessing on-farm water-use efficiency is sound and can be used under a range of farm conditions to identify how farmers use water for multi-crop and different economic situations. However, this methodology can help to improve water-use efficiency (WUE) only at the farm level. Water lost at the farm due to over-irrigation may not necessarily be lost at the field or at the basin level. Recovering on-farm losses comes only at a cost and usually has negative environmental consequences. Knowing farm WUE is also a key to evaluating the efficiency at all scales of resources use. The methodology was applied in four countries using six case studies. The studies revealed the following: Farmers' performance regarding farm water use vanes from site to site and from one crop to another. Reasons include: farmers' perspective of crop water needs; the type of irrigation system used and level of control, availability and cost of irrigation water; farm size, awareness and extension services; and other less important factors. Though the overall WUE, as defined in this study, is generally low, there is still great potential for improvement to save substantial amounts of water that can be used to expand irrigated areas and/or diversion to other uses. The study shows that policies create most of the conditions that determine levels of water-use efficiency. Examples include farm size, water allocation and costs, cropping patterns, input subsidies and crop pricing. To determine the amount of water allocation among crops, techniques and growing seasons there should be consideration of the level of efficiency which may be attained under each option. Developing and implementing proper policies is probably the most feasible option not only for using water efficiently but also for increasing farmers' income and protecting the environment. The capacity of NARS, in the countries studied, to analyze WUE at the farm and identify the constraints to improved water use is generally not sufficient. They still largely depend on conventional concepts of efficiency related to manage water under normal conditions. These concepts are unsuitable under conditions of water scarcity and should be changed. Emphasis should be put on developing the human and technical capacity of the NARS to a level where they can help farmers improve their performance and also influence policy. The approaches developed and used for this study may not be applicable to all situations. They should, however, be improved and combined with other approaches for evaluating water-use efficiency at field and basin levels to suit conditions in other dry areas. Such a combination is needed in all water-scarce areas to assess the current water use and evaluate the constraints to improved efficiency. Only then can meaningful measures be taken at all levels to cope with increasing water scarcity.

70


References ACIL Tasman. 2003. Scope of water use efficiency savings as a source of water to meet increased environmental flows-independent review. A report to the Murray Darling Basin Commission, March 2003. ACIL Tasman, Economics Policy Strategy, CAN 102 652 148 Sydney Office. ACSAD (Arab Center for the Studies of Arid Zones and Dry Lands). 1993. Use of Saline Water for Irrigation: Regional Symposium on Water Use and Conservation, 28 November - 2 December 1993, Amman, Jordan. ESCWA, CEHA. Ball, V.E. 1985. Output and productivity measurement in U.S. agriculture: 19481979. American Journal of Agricultural Economics, 67 (3). Bazza, M. and M. Ahmad. 2002. A Comprehensive assessment of links between irrigation water pricing and irrigation performance in the Near East. Paper presented at the Conference on Irrigation Water Policies: Micro and Macro Considerations, organized by the World Bank, Agadir Morocco, 15-17 June 2002. Burman, R.D., P.R.Nixon, J.L.Wright, and W.O. Pruitt. 1981. Crop water requirements. In: Design and operation of farm irrigation systems. M.E. Jensen(ed). Amer. Sos. . Agr. Eng. Monograph No 3. ASAE, St Joseph, MI, USA. Caswell, M. and D. Zilberman. 1985. The choice of irrigation technologies in California. American Journal of Agricultural Economics 67: 224-234. Chambers, R. G and R. E. Just. 1989. Estimating multioutput technologies. American Journal of Agricultural Economics 71: 980-995. Dawood, S. A. and S. N. Hamad. 1985. A Comparison of on-farm irrigation systems performance. Pages 540-545 in Proceedings of the Third International Drip/Trickle Irrigation Congress. Nov. 18-2I, 1985. California, USA. English, M. and S. N. Raja. 1996. Perspective on deficit irrigation. Agricultural Water Management 32: 1-14. ESCWA/FAO Joint Agriculture Division. 1994. Land and water policies in the Arab region. A Paper of The Expert Consultation on Sustainable Agricultural and Rural Development, June 1994,45 pp. ESCWA/ICARDA. 2000. Economic assessment of on-farm water use efficiency in agriculture: Methodology and two case studies. United Nations, New York 76 PP.

ESCWA/ICARDA. 2001. Economic and technical assessment of on-farm water use efficiency: three case studies from water scarce countries. United Nations, New York. ESCWAIICARDA. 2003. Enhancing agricultural productivity through on-farm water use efficiency. An empirical case study of wheat production in Iraq. United Nations, New York. Fraiture, C. and C. Peny. 2002. Why is irrigation water demand inelastic at low price ranges? Paper presented at the Conference on Irrigation Water Policies: Micro and Macro Considerations, 15-17 June 2002, Aghadir, Morocco. Giriappa, S. 1984. Water-use efficiency at the farm level in Bhavani Sagar Command Area of Tamil Nadu. Indian J. Agric. Sci. 54 (7): 572-575. Guerra, L. C., S. I. Bhuiyan, T. P. Tuong, and R. Barker.. 1998. Producing more rice with less water from irrigated systems. IRRI, SWIM, IIMI Discussion paper series, No. 29, 1998. Gujarati, D. 1988. Basic econometrics. Mc Graw Hill Book Company, New York. Johnston, J. 1984. Econometric methods. Third edition. McGraw Hill Book Company, New York. Just, R. E., D. Zilberman, E. Hochman, and Z. Bar-Shira. 1990. Input allocation in multicrop systems. American Journal of Agricultural Economics 72: 200-209. Just, R. E., D. Zilberman and E. Hochman. 1983. Estimation of multicrop production functions. American Journal of Agricultural Economics 65: 770780. Kennedy, P. 1985. A guide to econometrics. Second edition. The MlT Press, Massachusetts. Krulce, D., J. A. Roumasset and T. Wilson. 1997. Optimal management of a renewable and replaceable resource: the case of coastal groundwater. American Journal of Agricultural Economics 79: 1218-1228. Lin, J.Y. 1994. Impact of hybrid rice on input demand and productivity. Agricultural Systems, vol 10. Molden, D. J., R. Sakthivadivel, C. J. Peny, and C. de Fraiture. 1999. Indicators for comparing performance of irrigated agricultural systems. Research report 20, IWMI, Sri Lanka. Moore, M. R., N. R. Gollehan and M. B. Carey. 1994a. Multicrop production decisions in western irrigated agriculture: the role of water price. American Journal of Agricultural Economics 76: 859-874.

72


Moore, M. R., N. R. Gollehon and M. B. Carey. 1994b. Alternative models of input allocation in multicrop system: irrigation water in the Central Plains, United States. Agricultural Economics l 1: 143-158. Oweis, T. 1997. Supplemental imgation: A highly efficient water - use practice. ICARDA, Aleppo, Syria, 16 pp. Oweis, T. 2001. Cropping with increased water scarcity in dry areas: increased water productivity. Pages 75-97 in Proceedings of the International Workshop on New Approaches to Water Management in Central Asia. A Joint UNUICARDA International Workshop held in Aleppo, Syria, 6-1 1 November 2000. Oweis, T. and A. B. Salkini. 1992. Socio-economic aspects of supplementary irrigation. Paper Presented at the International Conference on Supplemental Irrigation and Drought Water Management, Sept. 27 to Oct. 2, 1992, Bari. Italy. Oweis, T. and H. Zhang. 1998. Water-use efficiency: index for optimizing supplemental irrigation of wheat in water-scare areas. Journal of Applied lmgation Science 33: 321-336. Oweis, T., A. Hachum and J. Kijne. 1999. Water harvesting and supplemental irrigation for improved water-use efficiency in dry areas. SWIM paper 7. IWMI, Sri Lanka, 1999. Oweis, T., K. Shideed and M. Jabr. 1999. Economic assessment of on-farm water use efficiency in agriculture: methodology and two case studies. United Nations, New York, 00-0051, 200, 76 pp. Oweis T. and A. Hachum. 2003. Improving water productivity in the dry areas of West Asia and North Africa. In Water Productivity in Agriculture: Limits and Opportunities for Improvement. Eds. J.W.Kijne, R.Barker and D. Molden. CABI Publishing. pp179-198. Perry, C . J. 1997. Alternative approaches to cost sharing for water service to agriculture in Egypt. Research Report 2. IIMI, Colombo, Sri Lanka. 15 pp Rosegrant, M. W. 1997. Water resources in the twenty-first century: challenges and implications for action. IFPRI 2020 Vision. Food, Agriculture, and the Environment Discussion Paper 20, March 1997, 27 pp. Rodriguez, A. 1997. Rural poverty and natural resources in the dry areas: the context of ICARDA's research. Working Paper, ICARDA, Aleppo, Syria, 20 PP. Salkini, A. B. and T. Oweis. 1993. Optimizing groundwater use for supplemental imgation of wheat production in Syria. Pages 1-12 in FRMP Annual report, ICARDA, Aleppo, Syria.

References

73

Scheaffer, R., Mendenhall and Ott. 1979. Elementary survey sampling. Second edition. Duxbury press, Massachusetts, USA. Schmidt, E.J. 2001. Water-use efficiency: an overview and economic perspective. Proceedings of the Annual General Meeting of South Africa Sugar Industry Agronomists' Association. Seckler, D. and R. A. Young. 1985. The choices of irrigation technologies in California. American Journal of Agricultural Economics 67 (2). Serageldin, I. 1998. Water in the twenty-first century: a dialogue. Water Policy. vol. 1. The World Bank. 1994. A strategy for managing water in the Middle East and North Africa. Washington, D. C. Tribe, D. 1994. Feeding and greening the world: the role of international agricultural research. CAB International & The Crawford Fund for International Agricultural Research, Wallingford, UK. Viets, F. G 1962. Fertilizers and the efficient use of water. Adv. Agron. 14:223-264. Whittlesey, N. K. and R. G. Huffaker. 1995. Water policy issues for the twenty-first century. American Journal of Agricultural Economics 77: 119-1203. Wolf, A. T. 1996. Middle East water conflicts and directions for conflict resolution. IFPRI, 2020 Vision. Food, Agriculture, and the Environment Discussion Paper 12. March 1996, 28 pp. World Resources Institute. 1999. World resources: a guide to the global environment. Special focus on climate change--Data on 146 Countries. Oxford University Press. New York. Zhang, H. and T. Oweis. 1999. Water-yield relations and optimal irrigation scheduling of wheat in the Mediterranean region. Agricultural Water Management 38: 195-21 1.


74

-

Appendix A Table Al. Estimates of on-farm water use in Radwania: fixed-allocatable input model. Notes: Numbers in parentheses refer to the calculated values o f t -statistics. (* and **) = significant

Independent variables Wheat Intercept 5886.03*,(1.62) Wheat area (ha) -179.29*,(-2.05) Cotton area (ha) 110.90,(1.12) Barley area (ha) 110.90*,(1.12) Wheat price (SWkg) 197.70*,(1.82) Cotton price (SWkg) -101.6**,(-3.42) 194.62*,(2.11) Barley price (SWkg) Total water (m3) 0.203**,(8.46) Experience in irrigation @i-44.02*,(-1.79) ) Water location (0, 1) " -528.04,(-0,301) Soil type (0, 1) 670.29,(0.941) Soil salinity (0, 1) -2965.5,(-1.004) Soil depth (0, 1) * -728.74,(1.36) Irrigation technology (0, 1) ' -2259.0**,(-2.75) Diesel price (SWliter) -16.64,(-0.126) Water application (0, 1) 80.36,(0.094) Water management (0, I) 8 127.72,(0.102) Water price (SLhalyear) 0.024,(0.077) 192.37,(1.14) Price of urea (SLkg) Price of phosphate (SLkg) -366.44**,(-2.12) K2 0.65 R2 0.75 D - W statistics 1.62 9.05** F statistic

'

-

Barley 3366.39,(1.17) -73.41,(-1.07) 1119.67,(1.39) -46.37,(-0.59) 21.36,(0.25) -19.71,(-0.84) 486.19**,(6.73) 0.031*,(1.63) -15.41,(-0.795) 192.51,(0.139) -175.24,(-0.312) -3866.45*,(-1.66) -43.48,(-0.103)

Cotton -9854.55*,(-2.04) 239.95**,(2.07) -377.75**,(-2.61) -66.139,(-0.502) -215.66,(-1.50) 124.38**,(3.15) -714.26**,(-5.87) 0.772**,(24.29) 55.01*,(1.69) 290.25,(0.125) -476.09,(-0.504) 7027.64*,(1.79) 900.39,(1.27)

79.185,(0.76) 167.90,(0.25) 739.91,(0.75) -0.032,(-1.31) -3.744,(-0.029) 15.35,(0.114) 0.53, 0.63 1.61 5.81**

-29.30,(0.167) -190.16,(-0.168) -1001.63,(-0.601) 0.043,(1.04) -261.18,(-1.20) 436.91*,(1.94) 0.93 0.95 1.29 61.34**

at 5% and 1% level of significance, respectively. Dummy variable for water location refers to the location of the farm to water source, taking the value of 1 for tail location and zero otherwise. (b) Dummy variable for soil type, taking a value of 1 for sandy soil and zero otherwise. Dummy variable for soil salinity, taking a value of I for low salinity and zero otherwise. Dummy variable for soil depth, taking a value of 1 for deep soil and zero otherwise. Dummy variable for supplemental irrigation technology, taking a value of I for sprinkler and drip irrigation technology and zero otherwise. Dummy variable if water application according to calendar schedule, taking a value of if water application is according to calendar schedule, and zero otherwise. Dummy variable for water management, taking a value of 1 if farmer rely on advanced management practices and zero otherwise.

Appendices

75

Table A2. Estimates of on-farm water use in Radwania: variable input model. Independent variables Intercept wheat a r e a(ha) Cotton area (ha) Barley area (ha) Wheat price (SL/kg) Cotton price (SLkg) Barley price (SLkg) Experience in irrigation (yr) Water location (0, 1) Soil type (0, 1) Soil salinity (0, 1) Soil depth (0, 1) Irrigation technology (0,l) Diesel price (SL/liter) Water application (0, 1) Water management (0,l) Water price (SL/ha/year) Price of urea (SL/kg) Price of phosphate (SL/kg)

R2 D - W statistics F - statistic

Wheat 2760.45,(0.483) 15.44,(0.28)

Barley 2508.75,(1.24)

Cotton -8238.88,(-0.612)

243.35,(1.38)

-77.03*,(-1.99) -1240.95,(-0.43) 1805.81 *,(1.98) -474.23,(-0. l I) -1055.15,(-1.29) -2043.95,(-1.53) 251.39,(1.19) 707.93,(0.53) 613.31,(0.30) 0.035,(0.69) -192.48,(-0.73) 54.11,(0.196) 0.07 0.23 1.71 1.40

Note: Same footnotes as for Table Al.

Table A3. Estimates of on-farm water use in Radwania: satisficing model. Independent variables Intercept Wheat area (ha) Cotton area (ha) Barley area (ha) Experience in irrigation (yr) Water location (0, 1) Soil type (0, 1) Soil salinity (0, 1) Soil depth (0, 1) Amount of rainfall (mm) Water application (0, 1) Water management (0, 1)

Wheat 10278.3*,(2.03) 11.103.(0.214)

D - W statistics F - statistic

Note: same footnotes as for Table A l .

Barley 10823.8**,(4.83)

Cotton 10496.6,(0.718)

76


Appendix B Table B1. Estimates of on-farm water use in Rabea: fixed-allocatable input model. Independent variables Intercept Wheat area (ha) Potato area (ha) Sugar beet area (ha) Tomato area (ha) Wheat price

Wheat -5231**,(-2.9) 14.17,(0.72) -6.61,(-0.91) -1.35,(-0.04) 9.87*, (1.56)

Potato -758.4,(-0.18) 20.4,(0.76) 53.6*,(2.30) -39.3,(-0.48) -8.5,(-0.61)

Sugar beet -1155,(-0.39) 4.88,(0.25) -16.73,(-1.38) 354.4**,(5.45) -8.05,(-0.77)

. Sugar beet price Tomato price Water limitation . Total water Imgation technology (0,l) Wage rate

R2 D W statistics

Tomato 46.81,(0.066) -30.18,(0.605) -18.04,(-0.61) -186.9,(-1.02) 138.7**,(5.13) 4'.?5.10.431

.

-115.3, 0 048) 0.41 0.33 2.08

0.795;(0.19)

0.119,(0.40)'

0.018*;(2.09)

-10.8 ,(-1.89) 0.98 0.98 2.12

0.92 0.91 2.07

0.94 0.93 1.65

F statistic Notes: Numbers in parentheses refer to the calculated values of t-statistics. (* and **) = significant at 5% and 1% level of significance, respectively. Dummy variable for water limitation, taking a value of 1 if the amount of water available to the farm is limited, and a value of zero otherwise. Dummy variable for supplemental irrigation technology, taking a value of I for sprinkler and drip irrigation technology and a value of zero otherwise. For tomatoes all farmers use flood irrigation.

Table B2. Estimates of on-farm water use in Rabea: variable input model.

I

Independent variables Intercept Wheat area .(ha). Potato area (ha) Sugar beet area (ha) Tomato area (ha) Wheat price (ID/kg) Potato price (ID/kg) Sugar beet price Tomato price (ID/kg) Water price (ID/ha/year) Water limitation (0, 1) lrrigation technology ( 0 Wage rate (I.D/hr)

-

D W statistics F - statistic

Wheat Potato -267.8,(-0.15) 1016,(0.44) 3.267,(0.315) 8106**, (53.0)

Sugar beet 1124,(0.637)

Tomato 3285,(0.551)

8663**,(21.9) 12641**,(26.5) 11.20,(0.48) 44.19*,(2.02) 205.9**,(5.22) 0.043, (0.071) 227.6, (0.173) 124.4, (0.044 (3.02)

0.013,(0.115) 0.039,(0.404) -1605.5,(-0.65) -1976,(-0.96) 5559.1,(1.35) 4089,(1.17)

0.06

0.98 0.98 2.11 667.9**

0.11 1.85 1.96

Note: Same footnotes as for Table B1.

0.89 0.90 2.18 135.3**

16.1 **,(3.75) 0.033,(0.125) -4429,(-0.758) -10.64 0.89 0.90 1.94 139.7**

9)

appendices

77

Table B3. Estimates of on-farm water use in Rabea: satisficing model.

I

Independent variables Intercept Wheat area (ha) Potato area (ha) Sugar beet area (ha) Tomato area (ha) Water limitation (0,l) Irrigation Land tenure 0, 1) Rainfall Water location (0, I)

Wheat 899.9.(0.383) -5.99,(-0.54)

Potato

Sugar beet 5029.(1.23) ,

- 1459.(-0.33)

Tomato 5986.(0.564)

8101**,(56.7) 9113**,(22.9) -301.9,(-0.22)

-881.3,(-0.34)

-3539*,(-1.51)

13055**,(26.2) -4134,(-0.65)

99.77,(0.24) 5.264,(-0.68) 3733*,(2.53) 0.09

-23.8,(-0.031) 6.825,(0.52) -1623,(-0.596) 0.98

69.98,(0.099) -7.03,(-0.582) -687,(-0.279) 0.87

-967.6,(-0,518) 13.72,(0.429) -6791,(-1.02) n 89

D W statistics

Note: Same footnotes as for Table B1.

Appendix C Table Cla. OLS estimates of on-farm water use of some crops in A1 Ghor, Jordan: behavioral model. Independent variables Intercept

Tomato

Potato

Pepper

Cucumber

Cauliflower

Citrus

132.8, (1.08)

-2.19, (-0.03)

372.7**, -154.1, (3.01) (0.69)

80.76*, (2.43)

-103.1, (-0.98)

Crop area (ha)

-14.13, (-0.3 1)

-21.13, (-1.42)

119.12, (1.33)

-109.5, (-0.75)

80.22, (0.87)

41.39*, (2.25)

Crop effect (0,l) "

4027.9**, (52.6)

2243**, (45.2)

3704**, (35.4)

4681**, (29.5)

2834**, (52.2)

12005**, (154)

Rainfall (mm)

-0.28, (-0.89)

0.001, (0.006)

-1.17**, (-3.6)

0.03, (0.04)

-0.29**, (-3.4)

0.17, (0.59)

Irrigation experience -2.31, (-075)

1.61, (0.88)

0.74, (0.25)

-3.39, (-0.67)

1.06, (1.32)

-2.01, (-0.78)

Soil depth (0, 1)

-80.74, (-1.1) 49.53, (0.77)

-25.78, (-0.56) -24.97, (-0.62)

-5.33, (-0.07) 34.56, (0.49)

-105.3, (-0.79) -62.4, (-0.55)

-28.21, (-1.39) 21.63, (1.17)

32.79, (0.52) 91.01, (1.64)

29.75, (0.31) 0.98, 1.91, 698.64**,

-28.18, (-0.46) 0.97, 2.08, 460.12**,

45.04, (0.4) 0.97, 1.62, 327.67**,

-88.7, (-0.51) 0.96, 1.23, 269.58**,

Soil salinity (0,l)

Soil type (0,

D - W statistics F - statistics

-3.08, 37.76, (-0.11) (0.45) 0.99, 0.99, 1.97, 2.10, 2210.77**, 6667.69**,

78


Table Clb. OLS estimates of on-farm water use of some crops in the Al Ghor: behavioral model. Independent variables Melon Intercept 188.6*, Crop area (ha)

(2.29) -408.5, (-1.46)

Crop effect Rainfall

(mm)

Irrigation experience

Soil depth (0, 1) Soil salinity (0, ) Soil type (0, 1)

D - W statistics F statistics

-

(26.4) -0.70**, (-3.3) 0.17, (0.08) 60.29, (1.22) 37.78, (0.86) 128.3*, (1.9) 0.97 1.98 347.00**

Wheat

Eggplant

Lettuce

-84.97, (-1.58) -1140**, (-6.0) 2824**, (22.7) 0.39*, (2.78) -1.48,

-263.2, (-1.64) -773**, (-3.45) 7771**, (41.3) 0.85*, (1.95) -0.32, (-0.08) 29.55, (0.30) -51.47, (0.60) 75.52, (0.57) 0.98 2.53 625.95**

-126.3*, (-1.9) 66.42, (0.72) 2238**, (24.7) 0.41*, (2.32) 1.01, (0.63) 4.37, (0.1) -87.7'*, (-2.4) 117.2*, (2.14) 0.97 1.98 319.94**

-49.85, (-1.52) -40.38, (-1.40) 115.3*, (2.58) 0.96 1.62 317.9**

Onion 150.2**, 81.46*, (2.91) (1.99) 12.64*, 96.85, (2.27) (-3.45) 20104*, 2712, (60.8) (38.7) -0.55**, -0.27**, (-4.13) (-2.61) 1.65, -0.31, (1.31) (-0.32) -2.53, 15.63, (-0.07) (0.63) 13.46, 19.16, (0.50) (0.84) 32.79, 33.69, (0.79) (0.98) 0.98 0.99 1.78 2.09 669.73** II96.16** Beans

Notes: Numbers in parentheses refer to the calculated values of t- statistics. (* and **) = significant at 5 percent and 1 per cent level of significance, respectively. Dummy variable for crop effect. taking a value of 1 if the farmer produce that crop and zero otherwise, Dummy variable for soil depth, taking a value of I for deep soil and zero otherwise. Dummy variable for soil salinity, taking a value of 1 for low salinity and zero otherwise. Dummy variable for soil type, taking a value of 1 for sandy soil and zero otherwise. OLS = Ordinary Least Squares Estimation Procedure.

Appendices

79

Appendix D Table D l . Estimates of on-farm water use in Nubaria, winter crops: fixed-allocatable input model. Independent variables Intercent

Wheat -771.0.

Faba bean 1267.1*.

Barsem -496.1.

Bersem price (LE/ton) Faba bean price (LE/ton) Wheat price (LE/ton) Bersem area (Feddan) Wheat area (Feddan) Faba beans area (Feddan) Price of urea (LE/kg) Total water Soil salinity (0, I) Soil depth (0, Experience in irrigation (yr)

R2 D-W statistics F-statistics

Notes: Numbers in parentheses refer to the calculated values oft- statistics. (* and **): Significant at 5 percent and I percent level of significance, respectively. LE = denotes Egyptian pound. Feddan = denotes area unit hectare = 2.38 Feddan). Dummy variable for soil salinity. taking a value of 1 for low salinity and zero otherwise Dummy variable for soil depth, taking a value of I for deep soil and zero otherwise.

80


Table D2. Estimates of on-farm water use in Nubaria, for winter crops: variable input model.

I

Independent variables Intercevt Wheat (Feddan) Wheat price Faba bean area (Feddan) Faba bean price Bersem area (Feddan) Bersem price Water price Soil salinity 1) Soil depth 1) Experience in irrigation (yr) Price of urea statistics

Wheat 958.04.(1.09)

Faba bean 785.39.(1.11)

Bersem

0.44 2.27

Note: Same footnotes as for Table

Table D3. Estimates of on-farm water use in Nubaria, for winter crops: behavioral model. Independent variables Intercept Wheat area (Feddan) Faba bean area (Feddan) Bersem area (Feddan) Soil salinity (0, 1) Soil depth (0, 1) Experience in irrigation (yr) Price of urea Crop effect (0, 1) D-W statistics F-statistic

Wheat -416.5,(-0.62) -67.33,(-0.90)

Faba bean -489.84,(-0.76)

Bersem 591.29,(0.82)

-30.74,(-0.56) -103.5,(-0.56) 66.22,(0.33) 3.33,(0.31) 2427**,(9.15) 0.66 2.31 14.13**

Note: Same footnotes as for Table Dl

57.51,(0.26) -22.32,(-0.10) -5.89,(-0.46) 334.7,(1.21) 1774**,(7.74) 0.61 2.06 11.14**

-70.28,(-0.83) -41.30,(-0.16) -42.45,(-0.15) -26.73*,(-1.79) 4.42,(0.01) 3516**,(13.86) 0.83 2.04 34.28**

Table D4. Estimates of on-farm water use in Nubaria, for summer crops: fixed-allocatable input model.

I

variables

Green pepper Squash

Water melon Tomato

Corn -0.14)

Intercept

.

.

.

Water

price area (Feddan) Tomato price Green pepper area (Fed) Green pepper price Squash area (Feddan) Squash price Corn area (Feddan) Corn price Soil salinity (0, Soil depth Experience in irrigation

0.1

3

. .O)

-0.1

0.066,

1.20, 0.348, (1.32)

7.77, (0.30)

Price of urea Total

D-W statistics F-statistics

.

.,

,

** 0.28 2.01 0.89

0.35 1.72 1.22

0.59 1.78

0.52 2.61 2.41

0.34 1.60 1.18

Note: Same footnotes as for Table

Table D5. Estimates of on-farm water use in model. variable Independent variables Water melon Intercept Crop area (Feddan) Crop price Water Soil salinity (0, 1)' Soil depth (0, Experience in irrigation of urea D-W Statistics F-Statistics

185.1 0.55 2.49

Tomato

Green pepper

for summer crops:

0.90) 1.70) 0.70 1.78

Corn

Squash

0.24) 0.83 1.98

0.48 1.79

2.26) 0.74 1.53 I

Note: Same footnotes as of Table Dl

Table D6. Estimates of on-farm water use in Nubaria, for summer crops: behavioral model. Independent variables Intercept Water area (Fed) Tomato area (Fed) Green pepper area (Fed) Squash area (Feddan) Corn area Soil salinity Soil depth 1)

Water melons Tomato -199 6,(-0.59) -549 14.97, (0.58)

Price of urea Crop effect (0, 1)' D-W statistics F-statistics

Green peppers Squash 31) 49 14) 06)

Corn 247

(14) 0.84 2.29

0.88 1.90 51.55"

0.88 2.56

I

I

Note Same footnotes as of Table

Appendix E Table E l . Estimates of on-farm water use in Beni Sweif, for winter crops: fixed-allocatable input model. Independent variables Intercept Wheat area (Fed) area (Fed) Wheat price (LEI ton) price Soil salinity (0, 1) Soil depth Experience in Price of manure Total water (m3) D-W statistics F-statistics

Wheat

Bersem

-107.68, 1.67

625.1

0.71 2.40

0.29 1.77 1.86

refer to the calculated values oft- statistics. (* and **): Significant at Notes: Numbers in 5 percent and 1 percent level of significance, respectively. Fed = Feddan, which denotes area unit (I hectare = 2.38 Feddan). LE = denotes Egyptian pound. Dummy variable for soil salinity, taking a value of 1 for low salinity and zero otherwise. Dummy variable for soil depth, taking a value of I for deep soil and zero otherwise.

Appendices

Table E2. Estimates of on-farm water use in Beni Sweif, for winter crops: variable input model. Independent variables Intercept Crop area (Fed) Crop price Water price Soil salinity (0, Soil depth (0, Experience in irrigation (yr) Price of manure R2 D-W statistics F-statistics

Wheat

Bersem

0.72 2.31

0.25 1.68 2.01

Note: Same footnotes as for Table El

Table E3. Estimates of on-farm water use in Beni Sweif, for winter crops: behavioral model. Independent variables Intercept Crop area (Fed) Soil salinity Soil depth (0, Experience in (yr) Price of manure R2 D-W statistics F-statistics


Wheat

2

0.24 2.58

0.25 1.67

83

84


Table E4. Estimates of on-farm water use in Beni Sweif, for summer crops: fixed-allocatable input model. Independent variables Intercept Cotton area (Fed) Tomato area (Fed) area (Fed) price Sunflower price Tomato price price Soil salinity (0, Soil depth (0, Experience in

Cotton -1293,(-1.52)

Corn -375.51,(-0.63)

Sunflower

Tomato

0.82 1.88 12.21

0.97 2.23

0.70 1.83

-2.3

Price of manure Total water statistics F-statistics

0.86 1.52


Table E5. Estimates of on-farm water use in Beni Sweif, for summer crops: variable input model. Independent variables

Intercept

Cotton -639.6,(-0.44)

Corn

Sunflower

Crop area (Fed) Crop price Water price Soil salinity Soil depth (0, I) experience Price of manure D-W statistics F-statistics

Tomato

-0.1 LO,(-0.50)

0.44 1.93

Note: Same footnotes as of Table El

0.60 1.85 8.86

0.97 2.18

0.66 1.92 11.40**

Table E6. Estimates of on-farm water use in Beni Sweif, for summer crops: behavioral model. Independent variables Intercept Crop (Fed)" Soil salinity (0, Soil depth (0, 1) experience (yr) Price of manure R2 D-W statistics F-statistics

Note: Same

Cotton

Corn

Sunflower

Tomato

0.60 1.54

0.59 2.05

,

0.19 2.39 2.13

0.2 1.94 2.41

EI

as

Appendix F Table F1. Estimates of on-farm water-use in Ninavah: fixed-allocatable input model.

Wheat irrigated area Price of Total amount of water available lo the farm (m') technology wage rate Water limitation

Adjusted R-Square F- statistic

663.06, (7.26)

690.50. (6.59) **

756.89.

745.74, (8.24)

0.74, (24.25) **

0.67, (18. 65)

0.65, (20.53) * *

0.67, (22. 78)

*

*

*

55) 0.88 0.88

0.83 0.83

Numbers in Parentheses refer to the calculated t-statistic. levels of significance, respectively.

.

* **

** 0.85 0.84

,

* 0.85 0.85

Significant at 5% and

Table F2. Estimates of on-farm water use in Ninavah: variable input model. I

variables Intercept Wheat irrigated Price of wheat Irrigation technology Water cost Water Source

3572.02 (0.52) 994.44 (7.23)

Adjusted R-Square

0.60

0.027 (2.26) * 1732.97

2

3

** **

*

*

**

0.02 (2.16) * 2340.82

588.73 (0.3 0.65

1567.30 (0.83 0.60 0.60

0.61

Note. Numbers in Parentheses refer to the calculated t-statistic. levels of significance, respectively.

*. **: Significant at 5% and

Tahle F3. Estimates of on-farm water use in Ninavah: behavioral model. Independent variables Simulation I

Wheat irrigated technology

Land tenure Rainfall R-Square Adjusted F-statistic

Simulation 2

Simulation 3

10960.79 (1.95) * 1249.76 (9.76)

13585.41 875.23 (5.90)

(6.03)

16172.03 (6.74)

20764.85 (7.19)

17789.95

4911.96 (2.1 8) * -39.60 (1.85) * 0.61 0.60

2721.48 (1.36) -31.09 (-1.58) * 0.67 0.66

6183.64 (2.62) -38.11 (-1.65) * 0.59 0.58 85.28"

4717.51 (2.37) -28.60 (-1.48) * 0.60 0.60

13576.41 (2.24) (7.83) **

Note. Numbers in parentheses refer to the calculated t-statistic. levels of significance, respectively.

*, **:

*

All sample simulation 10464.51 (1.88) 1047.66 (8.36)

at 5% and 1%

*