Field Evaluation of Irrigation Scheduling Strategies ...

IRRIGATION AND DRAINAGE

Irrig. and Drain. 65: 214–223 (2016) Published online 5 February 2016 in Wiley Online Library (wileyonlinelibrary.com) DOI: 10.1002/ird.1942

FIELD EVALUATION OF IRRIGATION SCHEDULING STRATEGIES USING A MECHANISTIC CROP GROWTH MODEL† SABINE J. SEIDEL1*, STEFAN WERISCH1, KLEMENS BARFUS1, MICHAEL WAGNER1, NIELS SCHÜTZE1 AND HERMANN LABER2 1

Technische Universität Dresden, Institute of Hydrology and Meteorology, Dresden, Germany 2 Sächsisches Landesamt für Umwelt, Landwirtschaft und Geologie, Dresden, Germany

ABSTRACT In a field experiment with white cabbage (Brassica oleracea L. var. capitata (L.) alef.) in Germany, three irrigation scheduling approaches were tested: (i) three sprinkler irrigation schedules based on soil water balance calculations using different development-dependent crop coefficients; (ii) automatic drip irrigation based on soil water tension thresholds; (iii) irrigation scheduling by real-time application of a partially calibrated mechanistic crop growth model. Multi-objective calibration was applied to derive a fully calibrated model as a diagnostic tool to identify the water loss terms of the individual irrigation strategies.The results of the experiment showed that: (i) high yields can be achieved with relatively low amounts of irrigation water (~100 mm), provided the optimal irrigation strategy and technique are applied; (ii) automated tension threshold-based drip irrigation outperformed soil water balance or crop growth model-based strategies; (iii) the soil water balance calculation approach relying on recommended Kc factors led to an enormous overirrigation; (iv) the application of a partially calibrated crop growth model led to an underestimation of the crop water requirements in conjunction with an incorrect timing of irrigation events and therefore resulted in the lowest yields. The diagnostic model identified percolation, for the wet sprinkler irrigation treatments, and canopy evaporation, for the dry and model-based treatments, as major water loss sources; only minimal additional water was lost in the tension-controlled treatment. Copyright © 2016 John Wiley & Sons, Ltd. key words: irrigation scheduling; soil water tension-based irrigation; crop coefficient; crop growth modelling; multi-objective optimization; white cabbage Received 8 January 2015; Revised 2 April 2015; Accepted 7 April 2015

RESUMÉ Dans une expérience de terrain avec du chou blanc (Brassica oleracea convar capitata var. alba) en Allemagne, trois approches de calendriers d’irrigation ont été testées: (i) trois programmes d’irrigation par aspersion, fondés sur des calculs du bilan hydrique du sol basés sur des coefficients culturaux; (ii) une irrigation au goutte à goutte automatique basée sur des seuils de pression d’eau du sol; (iii) une irrigation par l’application en temps réel d’un modèle mécaniste de croissance des cultures partiellement calibré. Des calibrations multiples objectives ont été conduites pour dériver un modèle entièrement calibré, qui servira d’outil de diagnostic et d’identification des termes de perte d’eau résultant des stratégies individuelles d’irrigation.Les résultats de l’expérience ont montré que: (i) des rendements élevés peuvent être atteints avec des quantités relativement faibles d’eau d’irrigation (~100 mm), à condition que les calendriers optimaux et que les techniques adaptées soient appliquées; (ii) l’irrigation au goutte à goutte à seuil de tension automatique a surperformé les stratégies fondées bilan hydrique du ‹‹sol bas » sur le modèle de croissance des cultures; (iii) les méthodes basées sur le calcul de bilan hydrique du sol à partir des coefficients culturaux Kc ont conduit à une énorme surirrigation; (iv) l’application d’un modèle de croissance des cultures partiellement calibrée a conduit à une sous-estimation des besoins en eau des cultures en conjonction avec une mauvaise synchronisation des tours d’eau et ont donc entraîné les rendements les plus faibles. Le modèle de diagnostic a identifié la percolation, pour les traitements humides d’irrigation par aspersion, et la verrière évaporation, pour les traitements secs et de modèles basés, comme des sources majeures de perte de l’eau, que l’eau supplémentaire minime a été perdu dans le traitement de la tension contrôlée. Copyright © 2016 John Wiley & Sons, Ltd. mots clés: calendrier d’irrigation; sols tension base calendrier d’irrigation; coefficient cultural; la modélisation de la croissance des cultures; chou blanc * Correspondence to: Dr Sabine Seidel, Technische Universität Dresden, Institute of Hydrology and Meteorology, Bergstrasse 66, D-01069 Dresden, Germany. E-mail: [email protected] † Evaluation des stratégies d’irrigation fondée sur expérience appliqué un modéle mechanistic

Copyright © 2016 John Wiley & Sons, Ltd.

215

SIMULATION-BASED EVALUATION OF IRRIGATION SCHEDULING STRATEGIES

INTRODUCTION The worldwide demand for fresh water is increasing steadily, especially in water-scarce areas, producing an unprecedented need for efficient water use in irrigated agriculture. In contrast, irrigation is often conducted as full irrigation to reach soil water content near field capacity (Smittle et al., 1994; Jones, 2004; McKeown et al., 2010) in order to increase yields and improve fruit quality. To determine the timing and amount of applied water, different irrigation scheduling strategies can be employed, varying in their complexity, technical infrastructure and required expertise. One option is irrigation scheduling based on soil water balance (SWB) calculations (Allen et al., 1998). Hereby, the crop water requirement is calculated from the reference evapotranspiration by multiplication with a crop-specific, development-dependent crop coefficient (Kc factor). Daily values of the crop water requirement are accumulated until a certain deficit threshold is reached and the soil water storage is filled by irrigation. The SWB calculation approach is sufficiently robust under a wide range of conditions (Jones, 2004). However, several drawbacks are involved in this approach (Annandale et al., 2000): (i) growth and development are dependent on calendar time, ignoring thermal time and water supply; (ii) the Kc factor curves cannot be transferred to different regions and to different planting dates in the same region; (iii) the approach assumes that water loss is always demand-limited, although it is well known that crop water uptake and evaporation can be limited by either atmospheric evaporative demand or by water supply; (iv) errors accumulate over time which makes data assimilation necessary. Specific crop coefficients for white cabbage in German conditions are provided by the Geisenheim irrigation scheduling approach (Paschold et al., 2010; Kleber and Mayer, 2014), later referred to as GH, serving as one approach in this study. Furthermore, two adaptations of the proposed crop coefficients were also applied in the field experiments and will be evaluated. Another approach is based on soil water tension and soil water content measurements. Soil water status is measured directly to estimate the irrigation water requirement. According to Shock and Wang (2011), a major advantage of soil water tension-based irrigation scheduling is that soil water tension can be closely related to the drought stress experienced by plant tissues. Soil tension-based irrigation is relatively easy to apply and irrigation can be automated which is important for intensive horticultural production (Jones, 2004). On the other hand, extensive measurement programmes can be required in highly heterogeneous soils in order to achieve representative measurements. A third option is the estimation of irrigation water demand by real-time application of mechanistic crop growth models. In this strategy, a model which is able to simulate all relevant crop growth and water response processes, is Copyright © 2016 John Wiley & Sons, Ltd.

updated at regular time intervals and run as a forward model to predict the—crop development and soil water status specific—irrigation water requirement. This approach relies heavily on an accurate and well-calibrated model. For reliable predictions, soil and plant model parameters, which are always site specific, need to be calibrated properly. Angulo et al. (2013) recommend deriving crop model parameters from field experiments where growth and development processes have been measured rather than from aggregated data of regional statistics. Currently, numerous crop models exist but only a few include horticultural crops. The mechanistic model DAISY (Abrahamsen and Hansen, 2000), the performance of which regarding yield estimation was found to be fairly good (Palosuo et al., 2011), provides parameterizations for several horticultural crops including white cabbage. Irrigation scheduling for cabbage has been tested worldwide (Cripps et al., 1982; Ells et al., 1993; Imtiyaz et al., 2000; Tiwari et al., 2003; Šturm et al., 2010; McKeown et al., 2010; Shock and Wang, 2011). However, due to differences in climate, varieties and other growing conditions, it is difficult to apply results from studies conducted outside of Germany (or Central Europe). To test irrigation scheduling under local conditions, white cabbage was cultivated in 2012 on four experimental sites all over Germany including the study site of this investigation. The Kc factors recommended by GH, along with three variations (one higher and two lower levels), were applied to estimate the irrigation water demand of white cabbage and the associated yield reductions caused by drought stress. However, no considerable drought impact on the yields was observed at any of the sites. This calls the recommended Kc values into question and reveals that drought stress effects and their influence on plant development and yields of white cabbage are only poorly understood. However, knowledge about these relationships is important for the estimation of feasible irrigation water demands. To shed some light on these relationships three irrigation scheduling approaches were applied in a field experiment in 2013 and 2014: (i) SWB calculations based on the GH Kc factors and two adaptations at lower levels; (ii) soil water tension-based irrigation scheduling; (iii) scheduling based on real-time model predictions. Along with extensive data collection, comprising soil water and plant development observations, this study compares the applied approaches regarding water productivities, applied irrigation water amount and model-estimated water loss.

MATERIALS AND METHODS Site description and experimental set-up The experimental site is located near Dresden in Germany (51° N, 13.9° E and 120 m altitude). Average annual precipitation and temperature are about 667 mm and 10.4 °C, Irrig. and Drain. 65: 214–223 (2016) DOI: 10.1002/ird

216

S. J. SEIDEL ET AL.

respectively. The loamy sand soil is composed of 35% sand, 39.5% silt and 25.5% clay (soil depth from 0 to 60 cm), with the sand content increasing at deeper soil depths (at about 1 m). The groundwater table of the site is located deeper than 6 m. Figure 1 shows a schematic representation of the measurement and irrigation set-up at the field site. Over three years (2012–2014), field trials with white cabbage were established. The experimental design included four randomized sprinkler irrigation treatments with each treatment being replicated three times. A linear move irrigation system (Gierhake, Germany) was used to sprinkler irrigate the crops of the SWB-based treatments. Moreover, an NMC-Pro drip irrigation system (Netafim, Israel) was installed in 2013 and 2014 in an adjacent plot for soil water tension-based drip irrigation treatment. The drip irrigation system has a discharge rate of 7.1 l h-1 m-2 and a drip line and drip emitter spacing of 75 and 30 cm, respectively.

Nl

g

Bl Bs

Bh

LAI H

0 - 29

White cabbage seedlings were transplanted into the fields on 15 May 2012 (variety Kilaton F1), 16 May 2013 (variety Typhon F1) and on 15 May 2014 with a planting density of 3 plants per m2. The plants were fully fertilized according to soil sampling and quantification of mineral nitrate.

Plant and soil observations Plant data collection in 2013 and 2014 included approximately weekly measurements of the leaf area index (LAI, measured with AccuPAR LP-80, Decagon Devices Inc.), plant height, leaf-N content (SPAD Chlorophyll-Meter, Decagon Devices Inc.) and stomatal conductivity (SC-1 leaf porometer, Decagon Devices Inc.), usually with 10 replicates. In order to provide an estimate of above-ground biomass, head samples were collected three times during crop growth and once at harvest. The samples were partitioned into leaves, stems and heads. About 50 heads (in the dripirrigated treatment only 25 heads) were collected at harvest from each treatment to estimate cabbage yield. The fresh heads were weighed and a subsample of five heads per treatment was dried down to a constant weight. Only centre rows were evaluated to avoid border effects. Water productivity (WP, in kg m-3) is here defined as dry matter head yield divided by the sum of applied irrigation water and rainfall during the growing season. Additionally, continuous measurements (every 10–15 min) of soil water tension using tensiometers (T4e, UMS, Germany) were taken in two treatments in 2013 (T and D, treatment descriptions see below) and in four treatments in 2014 at three soil depths. Standard climate data were collected hourly at the research site and the precipitation measured by a Hellmann precipitation gauge was corrected according to Richter (1995).

- 60

Model description - 90

-200 Figure 1. Soil characterisation at the experimental site in Pillnitz, Dresden showing the four distinct soil layers identified by soil sampling. The plant variables measured during the field campaign were the leaf-N-content (Nl), the biomass of the stems, the leafs and the heads (denoted by Bs, Bl and Bh, respectively), the leaf stomatal conductivity (g), the plant height (H) and the leaf area index (LAI) Copyright © 2016 John Wiley & Sons, Ltd.

In this study the one-dimensional mechanistic soil–plant– atmosphere system model DAISY was applied to simulate crop growth and the evolution of the water balance components at the field site. DAISY solves the Richards equation to simulate soil water flow by application of the retention and conductivity model of van Genuchten (1980). Input data comprise hourly data of wind speed, relative humidity, temperature, global radiation and precipitation. The model was set up according to the available soil data (layering, soil types) and the management during the irrigation experiments. The soil profile in the model has a depth of 2 m and ‘free drainage’ is applied as a lower boundary condition. The spatial discretization ranges between 0.125 cm at the top of the profile and 0.5 cm at the bottom and is combined with an adaptive time-stepping scheme leading to stable numerical solutions and minimization of numerical diffusion effects. Irrig. and Drain. 65: 214–223 (2016) DOI: 10.1002/ird

217


Model calibration Two different model set-ups have been calibrated, one acting as a forward model which is used in the real-time irrigation scheduling approach D using the experimental data of 2012 (described in detail below), and one as the final model which is calibrated using the experimental data of 2014. The model calibration for the real-time scheduling in 2013 and 2014 was done to fit observations from the sprinkler irrigation experiments in 2012. The observational data in 2012 were rather limited and comprised only dry matter yields. Therefore, only a few plant parameters were calibrated manually and the soil hydraulic parameters were estimated by laboratory evaporation experiments on 250 cm3 soil cores. To complicate model applicability for real-time scheduling, the white cabbage variety was changed in 2013, as the cabbage variety was affected by root flies in 2012. Hence, the real-time irrigation scheduling model is affected by a lot of uncertainties and has to be regarded as ‘partially’ calibrated. The second model set-up was calibrated applying experimental data of 2014, by a multi-objective approach comprising altogether 9 objective functions and over 40 parameters, controlling the soil water flow, root and plant development, photosynthetic efficiencies and development-dependent biomass partitioning. The multi-objective optimization problem was solved with the AMALGAM algorithm (Vrugt and Robinson, 2007). The algorithm fits the model parameters to all objectives and their trade-offs simultaneously, giving some additional information about the structural deficiencies of the model in the reproduction of the measurements and allows the selection of a suitable compromise solution after the optimization procedure (Werisch et al., 2014). The objectives comprised the soil water tensions at three depths, the evolution of the above-ground biomass divided into leaves, stem and head parts (three cuts), the plant heights, LAI and the overall dry matter yield based on the field data of 2014 (Seidel and Werisch, 2014). This second model calibration can be considered to be much more robust than the ’partial’ calibration, as the model performance was evaluated on different system components simultaneously and is

therefore considered to be a suitable tool to assess sources of water loss of the individual irrigation strategies.

Irrigation scheduling In 2013 and 2014 three different irrigation scheduling approaches were tested throughout the growing period, namely the soil water balance method (SWB), the real-time scheduling approach on the basis of the ‘partially’ calibrated model DAISY (D) and an automatic drip irrigation approach based on continuous observations of soil water pressure heads (T). Three different SWB methods were applied based on different assumptions about the underlying Kc factors. Detailed information is given in Table I.

Soil water balance—Geisenheim For irrigation scheduling in Germany, the water balances can be calculated using the ‘Geisenheim irrigation scheduling’ (GH) approach (Paschold et al., 2010). The approach is a SWB method which uses three to four developmentstage-dependent Kc factors for the vegetation period defined by the BBCH scale (Meier, 2001). These Kc factors were determined in lysimeter and field experiments in Geisenheim (Germany). For white cabbage, four Kc values for the following development stages are recommended (in spring 2013): Kc of 0.7 after BBCH 13 (transplanting), Kc of 1.1 after BBCH 18 (8-leaf stage), Kc of 1.7 after BBCH 111 (11-leaf stage) and Kc of 1.8 after BBCH 41 (heads begin to form). According to Paschold et al. (2010), the GH approach involves the following three steps: firstly, the soil has to be filled to about 90% of field capacity. For cabbage, only the soil layer from 0 to 30 cm soil depth is considered until BBCH 18, from 0 to 60 cm until BBCH 41 and from 0 to 90 cm soil depth afterwards until harvest. Secondly, the amount of water to be distributed at each irrigation event needs to be estimated. The amount depends on the soil characteristics, the vegetable and its developmental stage. Thirdly, the daily water balance has to be estimated. Thus, the daily potential reference evapotranspiration is multiplied by the development-stage-dependent Kc value. The water balance will then be summed up on a daily

Table I. Irrigation treatments conducted with white cabbage in 2013 Treatment

Irrigation scheduling strategy

Crop coefficient ( Kc ) BBCH 13–18

SWB1 SWB2 SWB3 D T

BBCH 18–111

BBCH 111– 41

BBCH 41– 47

GH irrigation scheduling approach 0.7 1.1 1.7 GH irrigation scheduling approach 0.7 1.1 1.4 GH irrigation scheduling approach 0.7 0.8 0.8 Real-time scheduling using the partially calibrated DAISY model; irrigation is triggered if the simulated soil water head is below 250 hPa Automatic drip irrigation when soil water tension is below 250 hPa, application of 15 mm per event


1.8 1.5 0.8 pressure

Irrig. and Drain. 65: 214–223 (2016) DOI: 10.1002/ird

218

S. J. SEIDEL ET AL.

basis until the amount of water to be distributed at each irrigation event is reached. The water balance is reduced by the amount of precipitation. In 2013, Kc ranged from 0.7 (transplanting) to 1.8 (head formation) for the highest level (SWB1, recommended by GH) and from to 0.7 to 0.8 for the lowest level (SWB3), with recommended irrigation amounts of 10 mm (before BBCH 18) and 20 mm (after BBCH 18) per event.

Model-based real-time scheduling The partially calibrated model was applied to estimate the irrigation schedule on a weekly basis in 2013. Once a week on a predefined day, the observed weather data and the irrigation events of the last 7 days were assimilated into the model. This set of initial conditions was used to run the model for the near future. Boundary conditions for the prediction period were derived from historic weather data representing average conditions without rainfall serving as input. Every time the simulated soil water pressure head reached 250 hPa an irrigation event with 15 mm was triggered in the model. The sum of all simulated irrigation events in a 7-day period defined the water requirement for the real experiment in the next week. Observed precipitation within this week was subtracted from the simulated water requirement.

Sensor-based drip irrigation In the automatically drip-irrigated treatment (treatment T), 15 mm were irrigated automatically when the soil water pressure head fell below 250 hPa at 29 cm soil depth. An irrigation event was followed by a break of 2 h to allow for the redistribution of the irrigation water in the soil. The automated drip irrigation started on 16 June 2013.

Evaluation of irrigation strategies The aim of every irrigation strategy is to use the applied water as effectively as possible, which means here that the applied water is plant available and is not lost by percolation, runoff or interception. In crop production, only transpiration, which is directly linked to biomass production, can be seen as productive (although stored water in the soil may be available for the next crop). In this study, the different irrigation strategies are evaluated regarding (i) the observed yields, (ii) the amount of applied irrigation water and (iii) their crop water productivities, and (iv) the irrigation efficiency (IE), which is calculated as the ratio of applied water and plant available water. Furthermore, the unproductive water losses of the individual irrigation strategies—defined as water which is evaporated from the canopy (interception), from soil or ponded water, percolated out of the root zone, and run off during the growing period—are estimated in a simulation study with the fully calibrated model to gain a more detailed overview Copyright © 2016 John Wiley & Sons, Ltd.

of the water losses. The water losses were predicted for the five irrigation schedules tested in 2013 and a hypothetical rain-fed treatment.

RESULTS In 2013, May and June were cool and wet with 383 mm of precipitation from transplanting (16 May) until the end of June. Cumulated precipitation during the whole growing period (16 May–1 October; 138 days) was 482 mm. In June, ponding was observed on the soil surface. Runoff from the plots was not observed as the field site is not sloping. The months of July and August can be characterized as dry.

Yield and water productivity Table II shows the fresh head yields of the 2013 experiments, which were harvested on 1 October; dry matter yields, applied water and corresponding water productivities are also shown. A significant yield reduction (according to Tukey’s honest significant difference (HSD) test (Tukey, 1949), was observed for the strategies SWB3 and D compared with SWB1 and 2. The observed head yields between SWB1 and SWB3 and between SWB1 and D were significantly different. The yield of SWB2 but also the yield of SWB3 and D did not differ significantly. Water productivity was highest for treatment T (1.45 kg m-3), which achieved the highest yields with a minimum of applied water. The lowest WP was achieved by treatment SWB1, due to the very high irrigation volumes.

Simulation study Model calibration and validation. The performance of the partially and the fully calibrated models regarding reproduction of observed yields is satisfying. The partially calibrated model (PC) simulated a dry matter yield of 7.50, 7.44 and 7.41 t ha-1 for the treatments SWB1–3, respectively, while 7.60, 7.45 and 6.96 t ha-1 were observed in Table II. Observed head yield as fresh (YFM) and dry matter (YDM) (mean ± standard deviation of four replicates; treatments sharing the same letter are not significantly different at the chosen level of P = 0.05), irrigation water applied (I) and water productivity (WP) in 2013 Treatment

SWB1 SWB2 SWB3 D T

YFM

YDM

I

WP

(t ha-1)

(t ha-1)

(mm)

(kg m3)

105 ± 3ac 103 ± 2c 97 ± 5bc 94 ± 5bc 112

8.3 ± 0.3 8.3 ± 0.2 7.9 ± 0.2 7.5 ± 0.5 8.5

410 306 108 106 105

0.93 1.05 1.33 1.27 1.45


219


Soil water pressure [cmWs]

Precipitation & Irrigation [mm]

the experiments. The fully calibrated model (FC) simulated a dry matter yield of 10.37 which is very close to the 10.4 t ha-1 that was observed in the SWB1 treatment of 2014, which served as calibration treatment. Additionally, the observed and by both models predicted pressure head time series at 29 and 52 cm soil depth of a drip-irrigated treatment of 2014 are shown in Figure 2. The fully calibrated model captures the observed dynamics far better than the partially calibrated model which uses soil hydraulic parameters estimated from soil core evaporation experiments. As a consequence it can be stated that the partially calibrated model performs well regarding the observed yields, but fails to simulate the correct partitioning of the above-ground biomass (see also Figure 3), which results in rigorous underestimation of the leaf and stem biomass. Furthermore, results are based on false soil water dynamics. As a consequence it underestimates the required irrigation water amounts and drought stress effects on yield. The fully calibrated model was validated against four other treatments of 2014. The head yield was very well predicted (NRMSE = 1.4%), but also the other plant variables are simulated accurately. The validation results for plant height (of treatment D of 2014) and LAI (of treatment T of 2014) and the corresponding model fits are shown in Figure 3. The overall performance of the FC model is very satisfying considering the numerous objectives in the optimization and the high number of parameters. However, there are still small deviations between the model simulations and the observations, but the model performs well enough to be applied as a diagnostic tool to analyse the loss terms of the individual irrigation strategies.

Evaluation of the irrigation scheduling strategies tested in 2013 To assess the effectivity of the irrigation strategies the following aspects are analysed: (i) the yields; (ii) the overall above-ground biomass; (iii) the applied irrigation water; (iv) the resulting water productivity and the water loss components of the water balance, namely evaporation and percolation as predicted by the fully calibrated model. Table III provides an overview of these parameters for the individual irrigation strategies applied in the experiments of 2013. As a baseline scenario, a synthetic rain-fed treatment (a treatment without irrigation) was simulated with the fully calibrated model to determine the minimum values for the individual components. Table III shows the water loss terms components evaporation from the canopy, soil evaporation and deep percolation. Additionally, their sum as a measure of the total water loss of the individual irrigation strategies and the differences compared with the rain-fed treatment are given in brackets to quantify the additional losses caused by irrigation. The irrigation efficiency is defined as the percentage ratio of transpiration and applied irrigation water. All variables where calculated for the time period between planting and harvest for the year 2013. Comparing the individual irrigation strategies with respect to the yields for the 2013 experiment, treatments can be classified into two groups, one with yields over 8 t ha-1, including treatments SWB1 and SWB2 and T, and treatments SWB3 and D with yields less than 8 t ha-1. This is not a surprising result as the applied amounts of irrigation water decrease from 410 mm (SWB1) to 106 mm (D) and

0 20 Prec Irr

40 60

0 −100 −200 −300 −400 −500 −600 −700

- 29cm

50 obs PC FC

0 −50 −100

- 52cm

20/06

30/06

10/07

20/07

30/07

09/08

19/08

29/08

08/09

18/09

28/09

Figure 2. Observed precipitation (Prec), irrigation water applied (Irr) and the measured soil water pressure head (obs) at soil depths of 29 cm and 52 cm compared to the model predictions of the partially (PC) and the fully calibrated (FC) model for a drip irrigated treatment in 2014 Copyright © 2016 John Wiley & Sons, Ltd.


220

S. J. SEIDEL ET AL.

a

b

10

60

9 50

plant height [cm]

leaf area index [ ]

8 7 6 5 4 3 2

obs PC FC

1 0

0

20

40

60

80

100

40 30 20 obs PC FC

10 0

120

0

20

40

days after planting

60

80

100

120

days after planting

Figure 3. Observed (obs) leaf are index development in treatment T (left) and plant heights of treatment D (right) shown as box plots, in which the filled dot marks the median, the boxes the 25th and 75th percentile, the whiskers extend either to 1.5*IQR (interquartile range) or the respective minimum or maximum value. Data points outside the interquartile range are marked by circles. Although in the interpretation of the classical box plot they are considered to be outliers, they are not in this case. The predictions of the fully calibrated (FC) and partially calibrated (PC) model are depicted by lines

Table III. Average measured dry matter head yields (Yobs), simulated yields (Ysim), as well as measured (Bobs) and simulated (Bsim) biomass above ground at harvest for all treatments in 2013 using the FC model: RF indicates a hypothetical rain-fed baseline scenario. Furthermore, the applied irrigation water (I) is given along with the water loss terms components, namely: evaporation from canopy (Ec), soil evaporation (Es), deep percolation (DP) and their sum (∑L). The irrigation efficiency (IE) is given to express the amount of irrigation water that can be effectively used by the crop as percentage value. The precipitation during the growth period was 482 mm Yobs

Ysim

Bobs

Bsim

I

Ec

Es

(t ha-1) SWB1 SWB2 SWB3 D T RF

8.3 8.3 7.9 7.5 8.5 –

9.6 9.6 9.3 9.3 9.4 6.5

15.8 15.3 16.2 15.7 16.6 –

∑L

(mm) 16.4 16.4 16.1 16.1 16.1

410 306 108 106 105 0

the yield reduction appears to depend on the applied irrigation water. In contrast to this, the highest yields and the highest above-ground biomass were achieved in the tension-based treatment, which also required the least irrigation water. These findings already point out some of the main advantages of tension-based irrigation strategies: the water is automatically applied at the moment when it is required and timing seems to play an important role here. Furthermore, the tension-based irrigation treatment was irrigated by drip irrigation which had lower discharge rates compared to the sprinkler irrigation treatments, resulting in longer residence times within the root zone and a decreased probability of irrigation water entering soil macropores and thereby bypassing larger volumes of the soil matrix. A third advantage of water application by drip irrigation is application directly on the ground, which circumvents evaporation and interception losses and places the water directly underneath the cabbages where the irrigation plume is shaded from radiation-reducing soil evaporation. In a sprinkler treatment, water is lost due to wind drift, evaporation within the air and interception on the plant leaves. The water Copyright © 2016 John Wiley & Sons, Ltd.

DP

158 (+79) 147 (+68) 109 (+30) 107 (+28) 75 (–4) 79

92 (+9) 92 (+9) 89 (+6) 89 (+6) 88 (+5) 83

IE (%)

462 (+185) 380 (+103) 289 (+12) 289 (+12) 277 (–) 277

712 (+273) 619 (+180) 487 (+48) 485 (+46) 440 (+1) 439

33.4 41.2 55.5 56.6 99.0

reaching the ground is located where the soil is not covered by leaves and is therefore prone to evaporation. Thus, the remarkable water productivity of the T treatment does not rely just on the irrigation scheduling strategy itself but also on the application technique. The lowest yield was observed in the D treatment, which is the sprinkler-irrigated treatment with the lowest irrigation water amount. Treatment D appears to support the importance of the right timing of irrigation events, as almost equal amounts of water were used in the SWB3 treatment which achieved slightly higher yields. As both treatments are based on the same water application technique timing is the only variable able to explain this small difference. This underlines the importance of robust and reliable model calibration by inverse modelling. In our case the wrong timing of the irrigation events originates mainly from the wrong soil water dynamics which overestimate the soil moisture and the insufficiently modelled plant development (LAI) which furthermore underestimates the model-estimated plant water requirements. The fully calibrated model was subsequently applied to simulate the individual treatments to allow a more detailed Irrig. and Drain. 65: 214–223 (2016) DOI: 10.1002/ird

221

rainfall and irrigation [mm]

a

0

60

10

50

20

40

30

30

40

20

50

10

60

0

20

40

60

80

100

120

deep percolation [mm]


0 140

rainfall and irrigation [mm]

b

0

60

10

50

20

40

30

30

40

20

50

10

60

0

20

40

60

80

100

120

deep percolation [mm]

days after planting

0 140

days after planting Figure 4. Simulated deep percolation (black bars on bottom) of treatment SWB1 (upper Figure) and treatment T (lower Figure) in 2013. The bars on top denote rainfall (blue) and irrigation water applied (red)

investigation of the water loss terms. Comparing the simulated and observed above-ground biomasses shows a good agreement of the model (NRMSE of 4%), while the simulated and observed yields exhibit larger, although acceptable, differences (NRMSE of 17%). This is due to the fact that the harvest index (ratio of yield and total above-ground biomass) was at 60% in 2014 (calibration period) higher than the 53% observed in 2013. Analysing the water losses identifies two major loss terms: (i) evaporation from the plant canopy which is increased for all sprinkler treatments between +28 mm (D) and +79 mm (SWB1); (ii) deep percolation which increases up to +185 mm (SWB1) for the wettest treatment and is also high for SWB2 (+103 mm), underlining the massive overirrigation in those two treatments. Treatments SWB3 and D lost only a small amount of the irrigation water by deep percolation, while no additional percolation was Copyright © 2016 John Wiley & Sons, Ltd.

simulated for the tension-based treatment. A graphic example of the difference of the deep percolation between treatments T and SWB1 is given in Figure 4. Calculating the irrigation efficiency (IE) as the percentage ratio of water that can be used by the crop (ergo water which is not lost) and the applied irrigation water allows an effective comparison of the individual strategies. The tensionbased treatment achieved an IE of 99% while in the wetter treatments SWB2 (41.2%) and SWB1 (33.4%) less than 50% of the applied water could effectively be used.

SUMMARY AND CONCLUSIONS Crop irrigation management is a complex and difficult task in which three questions need to be answered: (i) How much water should be applied? (ii) When should it be applied? Irrig. and Drain. 65: 214–223 (2016) DOI: 10.1002/ird

222

S. J. SEIDEL ET AL.

(iii) How should it be applied? Irrigation scheduling is conventionally based on either experience, SWB calculations, crop growth simulation models, soil water measurements, or on sensing of the plant’s response to water deficits. White cabbage was cultivated at the experimental site between 2012 and 2014 to test and evaluate different irrigation strategies: (i) SWB calculations using the ‘Geisenheim irrigation scheduling’ (GH) approach (Paschold et al., 2010) which provides specific Kc factors for white cabbage in Germany, (ii) demand estimation by real-time model application, and (iii) automated drip irrigation based on soil water tension measurements. Irrigation based on the partially calibrated mechanistic crop growth model resulted in the lowest yields (but yield did not differ significantly from treatment SWB3) due to the underestimated soil water dynamics and crop development, but still performed better than the sprinkler irrigation treatments (SWB1 and SWB2) regarding water productivity or irrigation efficiency. Performance of the model can be drastically increased if more measures (higher resolution and further plant variables) of crop growth and soil water dynamics are included in the optimization task, resulting in better prediction of the required irrigation water amounts and application timing. However, extensive data collection is required and disqualifies this approach for everyday irrigation practice in farming. Nevertheless, fully calibrated models serve as powerful tools for detailed investigation of irrigation strategies and associated potential water losses, which are becoming more and more important under the increasing demand for high-yielding water-efficient horticultural production. With these challenges ahead, common approaches should be evaluated properly and new approaches need to be tested. To improve simulation reliability and enhance model calibration and validation results, longlasting experiments with the same variety are required. Considering that the treatments SWB3 and D estimated realistic irrigation water demands but resulted in the lowest yields and the sprinkler treatments with higher yields required far more water, this proved the importance of correct timing, application rate and application technique to achieve highly efficient irrigation strategies. Moreover, the target wetness of the soil in the root zone in the ‘Geisenheim’ methodology (90% of the available water capacity) could be too high and neglects the fact that plant water uptake is mainly driven by pressure gradients between the soil and the root system (Jury et al., 2011). The soil water tension at a certain soil water content on the other hand depends strongly on the site-specific soil and can already be high (more sandy soils), causing reduced uptake rates, or be still very low (more clay content). However, an extended observation strategy is required to analyse possible reasons in further detail and to make sitespecific adjustments. These extended observations should Copyright © 2016 John Wiley & Sons, Ltd.

include measurements of evapotranspiration using lysimeters, Bowen ratio or eddy covariance approaches to define the real transpiration rates and support model calibration. Additional soil water content measurements could also help to define the water balance, but measurements with three different capacitance probes failed in the 2013 experiments, probably due to the high background electrical conductivity in the clayey soil layers. A fully calibrated mechanistic crop growth model was applied as a diagnostic tool to analyse the effectivity of the tested irrigation strategies and define the major sources of water losses. Most of the water is lost by percolation for the wet sprinkler treatments SBW1 and SWB2, while evaporation from the crop canopy is the main source in the treatments SBW3 and D. Percolation water losses, especially, are not only costly, but lead at the same time to negative off-site effects like fertilizer leaching or groundwater pollution which should be avoided. Water losses by canopy evaporation can only be circumvented by microirrigation strategies (Madramootoo and Morrison, 2013), such as drip irrigation, which proved to be superior in the 2013 experiments as almost no irrigation water was lost and the highest yields and above-ground biomasses were achieved. The performance of the fully calibrated model as a predictive tool for real-time irrigation scheduling needs to be tested in future experiments. Furthermore, the shortcomings of the models need to be analysed in order to gain further knowledge about the main driving processes controlling the drought stress response of white cabbage under German growing conditions.

ACKNOWLEDGEMENTS The authors are very grateful for the help of Verena Wommer, Anne Hartman, Patrick Röhm, Ulrike Grießbach, Sebastian Kloss, Peter Stange and Marika Wolf and the numerous assistants of the research site of the Sächische Landesamt in Pillnitz for their valuable help with the field experiments and the harvest. These investigations are part of the research project ’SAPHIR— Saxonian Platform for High Performance Irrigation’ funded by the EU ESF ‘Nachwuchsforschergruppen’ programme under grant no. 100098204.

REFERENCES Abrahamsen P, Hansen S. 2000. Daisy: an open soil–crop–atmosphere system model. Environmental Modelling and Software 15: 313–330. Allen RG, Pereira LS, Raes D, Smith M. 1998. Crop evapotranspiration: guidelines for computing crop requirements. Irrigation and Drainage Paper No. 56. FAO: Rome, Italy. Angulo C, Lock RR, Enders A, Fronzek S, Ewert F. 2013. Implication of crop model calibration strategies for assessing regional impacts of Irrig. and Drain. 65: 214–223 (2016) DOI: 10.1002/ird


climate change in Europe. Agricultural and Forest Meteorology 170: 32–45. Annandale JG, Campbell GS, Olivier FC, Jovanovic NZ. 2000. Predicting crop water uptake under full and deficit irrigation: an example using pea (Pisum sativum L. cv. Puget). Irrigation Science 19: 65–72. Cripps JEL, George PR, Oakley AE. 1982. Scheduling irrigation of cabbages using pan evaporation. Irrigation Science 3: 185–195. Ells JE, McSay AE, Kruse E. 1993. Irrigation scheduling programs for cabbage and zucchini squash. HortTechnology 3(4): 448–453. Kleber J, Mayer N. 2014. Geisenheim Irrigation Scheduling 2014 using the FAO-56 Penman - Monteith equation. Tech. rep. Hochschule Geisenheim University, Institut für Gemüsebau, Geisenheim, Germany. Imtiyaz M, Mgadla N, Chepete B, Manase S. 2000. Response of six vegetable crops to irrigation schedules. Agricultural Water Management 45: 331–342. Jones H. 2004. Irrigation scheduling: advantages and pitfalls of plant-based methods. Journal of Experimental Botany 55(407): 2427–2436. Jury WA, Or D, Pachepsky Y, Vereecken H, Hopmanns JW, Ahuja LR, Clothier BE, Bristow KL, Kluitenberg GJ, Moldrup P, Šimůnek J, van Genuchten MT, Horton R. 2011. Kirkham’s legacy and contemporary challenges in soil physics research. Soil Science Society of America Journal 75(5): 1589–1601. McKeown AW, Westerveld SM, Bakker CJ. 2010. Nitrogen and water requirements of fertigated cabbage in Ontario. Canadian Journal of Plant Science 90(1): 101–109. Madramootoo CA, Morrison J. 2013. Advances and challenges with microirrigation. Irrigation and Drainage 62: 255–261. Meier U. 2001. Growth Stages of Mono-and Dicotyledonous Plants. Technical report. Federal Biological Research Centre for Agriculture and Forestry: Oxford, UK. Palosuo T, Kersebaum K, Angulo C, Hlavinka P, Moriondo M, Olesen JE, Patil RH, Ruget F, Rumbaur C, Takac J, Trnka M, Bindi M, Caldag B, Ewert F, Ferrise R, Mirschel W, Saylan L, Siska B, Rötter R. 2011. Simulation of winter wheat yield and its variability in different climates


223

of Europe: a comparison of eight crop growth models. European Journal of Agronomy 35(3): 103–114. Paschold P-J, Kleber J, Mayer N. 2010. Geisenheim Irrigation Scheduling. Technical report. Forschungsanstalt Geisenheim—Fachgebiet Gemüsebau. Richter D. 1995. Ergebnisse methodischer Untersuchungen zur Korrektur des systematischen Messfehlers des Hellmann-Niederschlagsmessers. Technical report. Deutscher Wetterdienst: Offenbach, Germany. Seidel SJ, Werisch S. 2014. Dependence of the accuracy of crop growth model predictions on calibration procedure and data input. In Proceedings of 22nd International Congress on Irrigation and Drainage and 65th IEC Meeting, 14–20 September, Seoul, Republic of Korea. Shock CC, Wang F-X. 2011. Soil water tension, a powerful measurement for productivity and stewardship. Hortscience 46(2): 178–185. Smittle DA, Dickens WL, Stansell JR. 1994. Irrigation regimes affect cabbage water use and yield. Journal of the American Society for Horticultural Science 119(1): 20–23. Šturm M, Kacjan-Marsic N, Zupanc V, Bracic-Zeleznik B, Lojen S, Pintar M. 2010. Effect of different fertilisation and irrigation practices on yield, nitrogen uptake and fertiliser use effciency of white cabbage (Brassica oleracea var. capitata L.). Scientia Horticulturae 125(2): 103–109. Tiwari K, Mal P, Singh A. 2003. Effect of drip irrigation on yield of cabbage (Brassica oleracea L. var. capitata) under mulch and nonmulch conditions. Agricultural Water Management 58(1): 19–28. Tukey J. 1949. Comparing individual means in the analysis of variance. Biometrics 5(2): 99–114. van Genuchten MT. 1980. A closed-form equation for predicting the hydraulic conductivity of unsaturated soils. Soil Science Society of America Journal 44(5): 892–898. Vrugt JA, Robinson BA. 2007. Improved evolutionary optimization from genetically adaptive multimethod search. Proceedings of the National Academy of Sciences USA 104: 708–711. Werisch S, Grundmann J, Al-Dhuhli H, Algharibi E, Lennartz F. 2014. Multiobjective parameter estimation of hydraulic properties for a sandy soil in Oman. Environmental Earth Sciences 72(12): 4935–4956.
