Recent Advances in Remote Sensing and Geoinformation Processing for Land Degradation Assessment – Röder & Hill (eds) pp.313-322 © 2009 Taylor & Francis Group, London, ISBN 978-0-415-39769-8
Satellite Image Processing and Geo-statistical Methods for Assessing Land Degradation around Watering Points in the Ust-Urt Plateau, Kazakhstan A. Karnieli & U. Gilead The Remote Sensing Laboratory, Jacob Blaustein Institutes for Desert Research, Ben Gurion University of the Negev, 84990, Israel, e-mail:
[email protected]
T. Svoray Department of Geography and Environmental Development, Ben Gurion University of the Negev, 84105, Israel,
ABSTRACT: Desertification around watering points has been well observed by satellite images in many drylands around the world. It can be recognized as radial brightness belts fading as a function of the distance from the wells. The primary goal of this study was to characterize the spatial and temporal land degradation/rehabilitation in Central Asia drylands, in terms of vegetation and soil patterns, during different time periods, with respect to socio-economic changes before and after the collapse of the Soviet Union. More specific objectives of the study were: (1) to develop a geo-statistical model, based on the kriging interpolation technique and using high-resolution satellite image processing to assess spatial and temporal land cover patterns in three key time periods (mid 1970's, late 1980’s, and 2000); (2) to conduct a change detection analysis based on the interpolation products to assess the direction and intensity of changes between the study periods; and (3) to link the previous findings to the socio-economic situations before and after the collapse of the Soviet Union that influenced the grazing gradients and hence the landuse/land-cover state of the study area. The Tasseled-Cup's Brightness Index was found as the best spectral transformation for enhancing the contrast between the bright degraded areas close to the water wells and the darker surrounding areas far and in-between these wells. Empirical variograms were computed for each of the images and the exponential models were fitted. The Kriging geo-statistical technique utilized the variograms for creating brightness maps. The maps demonstrate the grazing gradient as levels of degrading belts around the wells. Change detection analysis, based on the Kriging maps, reveals some land rehabilitation between the 1975 and the 1987 images. However, mixed results, degradation and rehabilitation, were observed between the 1987 and the 2000 images. Degradation of the area occurs due to recent exploration and exploitation of the gas and oil reserves in the region. Consequently, large areas went through intensive 'technological desertification' that means utilizing large amount of heavy-duty equipments, large-scale plants, and vehicles that damage the soil surface. The rehabilitation of the rangelands can be explained by some historical events of the last decades. Following independence of the former Soviet states in 1991 and the imposition of difficult economic conditions with transition reforms, several major socio-economic changes occurred that caused drastic declines in livestock populations, with the major drop in the number of sheep and goats, and hence vegetation recovery and land rehabilitation. 1 INTRODUCTION According to the United Nations Environment Program (UNEP) the term overgrazing refers to a practice of allowing a much larger number of animals to graze at a location than it can actually support. As a result, overgrazing by different types of livestock is perhaps the most significant
anthropogenic activity that degrades rangelands and causes desertification in terms of plant density, plant chemical content, community structure, and soil erosion (Manzano & Návar 2000). In arid and semi-arid environments, land (soil and vegetation) degradation is mainly related to area surrounding point-sources of water, either natural or artificial, such as wells or boreholes (Lange 1969). Pickup & Chewings (1994) were the first who defined the term 'grazing gradient' as "spatial patterns in soil or vegetation characteristics resulting from grazing activities and which are symptomatic of land degradation". Domestic animals (sheep, goats, cattle, camels, yaks, and horses) prefer to graze in vicinity to a watering point. When food is depleted in this area they move away from the source of water but return regularly for drinking. Consequently, higher number of individuals is frequently concentrated around the watering points; density that decreases gradually with increasing distance from water (Pickup et al. 1993; Friedel 1997). The livestock grazing distance is limited, depends on the water demand of different animals, season and weather conditions, and quality of the forage. Typical distance is 4-6 km that can increase to 10 or even 20 km under extreme conditions (Hodder & Low 1978). Since the radial pattern around watering points is well observed from space, most of the recent studies are based on the interpretation and modeling of remotely sensed data, which can be analyzed in a semi-automated and repeatable way over vast and remote areas. Various remote sensing based models have been developed to estimate the spatial distribution of the different variables around the watering points, these are the PD54 (Pickup et al. 1993), Normalized Difference Vegetation Index (NDVI) calculated from Advance Very High resolution Radiometer (AVHRR) (Hanan, et al. 1991), or Probabilistic linear spectral mixture model, called AutoMCU, based on Monte Carlo analysis (Harrise & Asner 2003). The above-described ground and spaceborne observations have demonstrated not only that the grazing gradients are characterized by concentric circles around the watering points, but also the spatial nature of the measured biotic, abiotic, and environmental variables that are distributed in a common fashion. Each variable (e.g., vegetation cover, grass and annuals production, bush encroachment ?, soil pH, organic content, phosphate, and nitrate, soil nutrient concentrations, particularly potassium and phosphorus, and track density) has low (or high) value near the center and changes continuously as the distance increases. Moreover, most of the observations show that the rate of improvement (or decline) of each variable does not change after several kilometers (usually 5 km as mentioned before) from the watering point. The current paper presents another approach for assessing and mapping the grazing impacts around watering points. The primary goal of the study was to characterize the spatial and temporal land degradation/rehabilitation in Central Asia drylands, in terms of vegetation and soil patterns, during different time periods, with respect to the socio-economic changes before and after the collapse of the Soviet Union. More specific objectives of the study were: (1) to develop a geo-statistical model, based on the kriging technique and using high-resolution satellite image processing in order to assess spatial and temporal land cover patterns in three key different time periods (mid 1970's, late 1980’s, and 2000); (2) to conduct a change detection analysis based on the geo-statistical products in order to assess the direction and intensity of changes between the study periods; and (3) to link the previous findings to the socio-economic situations that influenced the grazing gradients and hence the land-use/ land-cover state of the study sites before and after the collapse of the Soviet Union. 2 METHODOLOGY 2.1 Study site The study site is located in the Ust-Urt desert plateau, ca. 160,600 km2, between the Caspian and Aral seas in Central Asia and occupies the southern part of Kazakhstan, the northern part of the Karakalpak Republic, and Turkmenistan (Fig. 1). It rises to between ca. 150 and 300 m AMSL. It consists primarily of stony desert with grey-brown soils. The area has a semiarid
continental climate with hot summers and cold windy winters. Vegetation consists on Artemisia terrae-albae, Anabasis, and salsae. Its semi-nomadic population raises sheep, goats, and camels. The North Caspian basin is a petroleum-rich region with large oil and gas reserves but hardly explored yet.
Figure 1. (A) Location of the study area in the Ust-Urt plateau, Kazakhstan (image courtesy of Google Earth). (B) The specific study site on Landsat-TM image. 2.2 Image processing Three Landsat images of the study area were used. They were acquired in 1975, 1987, and 2000 respectively by different sensors – MSS, TM, and ETM+. From the original images, subsets of the study sites were made, bounding an area of 1,184 km-2. Image processing started by converting digital numbers to reflectance values (Markham & Barker, 1985; http://ltpwww.gsfc.nasa.gov/IAS/handbook/handbook_htmls/chapter11/chapter11.html). Reflectance values were used to calculate several vegetation indices for each image subset. Reflectances were used for calculating several vegetation indices for each image subset. These include: NDVI (Rouse et al. 1974), SAVI (Huete 1988), MSAVI (Qi et al. 1994), PVI (Richardson & Wiegand 1977), and the Tasseled Cap-derived Greenness and Brightness indices (Kauth & Thomas 1976). Performance analysis, in terms of standard deviation and stretch of the index values within the dynamic range, was applied to all indices. These analyses revealed that the Tasseled Cap-derived Brightness Index (BI) had produced the best contrast and consequently had been selected for further analysis. The BI has the general form of
BI = α1( B1) + α 2( B 2) + ..... + α n( Bn)
(1)
where Bn is the spectral band number and αn is the appropriate BI coefficients of each sensor – MSS, TM, and ETM+ (Crist et al. 1984, Kauth & Thomas 1976, Huang et al. 2002, respectively).
Besides the statistical significant results, and that this index was originally designated to examine soil properties, its advantage is in the ability to compare between the different sensors that have different spectral bands, as it reduces the different spectral bands to one normalized layer of BI values. The resolution of the images was reduced by a factor of 3 and 6, for the MSS and TM/ETM+ images, respectively. The resulting pixel size of 171 m is processed easily, as the sub-scene size was reduced from approximately 1.5 million to 40,000 pixels. 2.3 Geostatistical analysis The ordinary kriging interpolation was used to model and map the spatial variation of surface brightness values in the study sites. The key concept of geostatistics is that of the regionalized variable (Matheron 1971), defined as a variable that can be characterized from a number of measurements that identify spatial structure. The basic assumption underlying this theory is that when assuming spatial continuity, samples that are located adjacent to one another tend to be more similar than samples located in remote areas. This spatially dependent variation may be treated statistically and described through a number of parameters derived from a semivariogram that is the function relating the semi-variance to the directional distance between two samples. The semi-variance is defined as half the mean squared difference between two samples in a given direction and distance apart (Eq. 2). The direction and distance are defined by the vector h that is commonly referred as the lag:
1 N (h) ( z xi − z xi + h ) 2 γ ( h) = ∑ 2 N (h) i =1
(2)
where γ(h) is the semi-variance at lag h, N(h) is the number of sample-pairs a distance h apart, and Zi is the value of the regionalized variable at location i. In addition to the lag, the variogram is characterized by three other parameters – the nugget, range, and sill. The nugget is variability at zero distance and represents sampling and analytical errors. The range of influence designates the extent, say a distance a, beyond which autocorrelation between sampling sites is negligible. The sill represents the variability of spatially independent samples. An empirical semi-variogram can be calculated from the given set of observations and then fitted to several common theoretical models (Delhomme 1979). Once the theoretical semivariogram is chosen, several criteria should be applied to determine the correctness of the model and to adjust its parameters. The current study uses four of these criteria: (1) Quantile-Quantile (Q-Q) plot – this is a visual comparison of two distributions while the quantiles from the two distributions are plotted versus one another. Thus a Q-Q plot of two identical distributions will result in a straight line while departure from this line reveals where they differ. To test the normality of the BI values, they were plotted versus standard normal distribution data (provided by ArcGIS Geostatistical Analyst tool). (2) Cross-validation scatter plot – in this plot, the measured and predicted data are regressed and the cloud of points is compared to a 1:1 line and to a line of best fit. (3) Mean kriged estimation error:
1 n 1 n ( Z xi −Z ∗ xi ) = ∑ ε i ≈ 0 ∑ n i =1 n i =1
(3)
where ε is the difference between the kriged and the known point value (this term should approach 0). (4) Mean standardized squared estimation error:
1 n 1 n ∗ ∗ 2 [( ) / ] (ε i / si ) 2 ≈ 1 Z − Z s = ∑ ∑ xi xi i n i =1 n i =1
(4)
where s*i is the estimation standard deviation (this term should approach 1). The rationale for choosing the kriging technique was the similarity in spatial structure of most of the above-mentioned variables, gradually increasing (or decreasing) as a function of the increasing distance from the watering point until reaching the limit of no grazing effects, and the typical shape of the variogram. Curran (1988) and Woodcock et al. (1988a, 1988b) introduced the semi-variogram to remote sensing and discovered that the parameters of the variogram can be directly related to a feature in an image. The kriging technique has recently become very common for analyzing spaceborne data (Oliver et al. 2000). 2.4 Image differencing change detection analysis Post-processing change detection method, namely BI Differencing (Yuan et al. 1998), was implemented to compute the degree and direction of the changes in each site and between the imaging periods. The general function of the change can be considered as:
⎧⎪0 if BI t − BI t +1 ≤ T change = ⎨ (5) ⎪⎩1 if BI t − BI t +1 > T where t and t+1 represent the two time periods and T is the threshold value. A common way to assess changes is based on determination of thresholds in terms of standard deviation levels below and above the mean of the difference between the BI values ( ΔBI ) of the images under study. In this manner, one can distinguish between changed and unchanged pixels as well as between negative and positive changes (Jensen 1986). In the current study, one standard deviation from the ΔBI was defined as the threshold and steps of one standard deviation beyond this threshold determined the magnitude and direction of the change. 2.5 Semivariance for change detection Addink (2001) introduced the Absolute Normalized Difference at lag h, AND(h), to calculate differences between two sets of semivariances, γ(h), from different dates, t and t+1:
AND ( h ) =
1 2 1 2
γ t +1 ( h ) − γ t ( h )
(γ ( h ) + γ ( h ) ) t +1
(6)
t
The AND(h) algorithm can produce values ranging from 0 to 1, indicating no and total change, respectively. 3 RESULTS AND DISCUSSION Subsets of the study site were extracted from Landsat MSS, TM, and ETM+ of the years 1975, 1987, and 2000, respectively. Figure 2 represents the BI products as calculated by Equation 1 with the appropriate BI coefficients. Watering points can be recognized as light spots spread over mostly the MSS and TM images, but the ETM+ one (Fig. 2C). In the latter image many of the watering points disappeared, however, a wide bright area exists in the center of the image. The brightness levels were equally stretched for the different sensors and ranging between 0.65 and 0.96. High values correspond to bare soil while low values indicate vegetation.
Figure 2. Brightness index products of sub-images used for the geostatistical analysis. (A) Landsat-MSS (1975) image; (B) Landsat-TM (1987) image; (C) Landsat-ETM+ (2000) image. Bright areas indicate degraded regions due to grazing and/or technological desertification. The three BI images were used for the kriging analysis. The first step in this course of action was to establish the empirical semi-variogram based on ca. 40,000 pixels in each image, for the three periods. Subsequently, several theoretical models were examined and the exponential model was selected due to the best cross validation results. Thus, all empirical models were fitted with an equation of the form: ⎛
⎛ h
⎝
⎝
γ ( h ) = C0 + C1 ⎜1 − exp ⎜ −
⎞⎞ a ⎟⎠ ⎟ ⎠
(3)
where a is the range, h is the lag, C0 is the nugget, and C0+C1 equals the sill. The fitted exponential models are illustrated in Figure 3. All variograms were processed with 16 lags of 1,000 m each. Visually, the 1975 and 1987 variograms look quite similar, having a typical variogram shape. Note that when the shape of the variogram is more round it can be referred to more symmetrical features in the image. Contrary, the 2000 variogram reaching about the same sill level only after 16,000 km and its shape is more linear than round.
Figure 3. Modeled variograms for the Brightness Index values of the Ust-Urt Plateau in 1975; 1987; and 2000.
The Q-Q plot (Fig. 4a) shows that in principle the distribution of the brightness values does not deviate much from normal distribution. A small deviation is observed in the region of the upper quantile and in the lower regions. However, these deviations are considered to be relatively small and thus can be ignored. The cross validation graph (Fig. 4b) shows a very good correlation between the measured and predicted values (R2 = 0.92). The slope coefficient is very close to unity and the intercept coefficient is very close to zero. This means that the point distribution best fit is very close to the desired 1:1 line.
Figure 4. Cross validation of the exponential model semi-variogram fittings. (A) Q-Q plot; (B) Predicted vs. Measured plot. In the next step, the kriging interpolation maps were performed based on exponential models. Since the grazing impact was assumed to be an isotropic feature and the direction has no influence on the spatial variation, the maps were derived from the omni-directional variograms. Figure 5 depicts the final products for the distribution of the BI values for the three periods. In the 1975 and 1987 maps, one can observe the belts around the watering points, indicating progressive land degradation radiating from the wells, i.e., the grazing gradient. The dark-red areas in the images are related to zones where the grazing impact is the predominant feature that has a strong effect on the spatial variation. These areas can be considered as the center of the grazing impact, denoted as the 'sacrifice zone' by Perkins & Thomas (1993). The surrounding light red and yellow belts represent a mixed zone where grazing impact and natural variability overlay each other or create a stable balance. This zone can be compared to an edge zone of the grazing impact and highlights the principal migration routes of livestock. The zone colored by blue tones is considered to be an area where natural variability overbalances the grazing impact, denoted as 'grazing reserve'. The radial pattern, related to the grazing gradient, is not seen in the 2000 map. Instead, the dominant feature in the middle of the scene is colored in red tones.
Figure 5. Kriging interpolation maps for images shown in Figure 2 based on the Brightness Index values. Dark tones indicate land degradation due to grazing and/or technological desertification. (Color figure at the end of the volume|).
Image differencing change detection analysis, based on the BI values, was performed on the two pairs of kriging maps – 1987 vs. 1975, and 2000 vs. 1987. Results are illustrated in Figure 6. Figures 6A and 6C are the change maps while Figure 6B and 6D are the respective frequency histograms of the change categories. The SD lines are also presented. The difference map computed from 1987 and 1975 shows that despite of the area that was considered as 'no change' (64.0%), more area underwent rehabilitation process than degradation (22.1% vs. 13.8%). As opposed to this favorable land cover change, degradation process characterizes considerable portion of the area (30.7%) during the second period (2000 vs. 1987), while rehabilitation occurred only in minor part (12.5%).
Figure 6. Change detection results shown as maps and histograms. HR = high rehabilitation; LR = low rehabilitation; HD = high degradation; LD = low degradation. (A, B) difference between the 1987 and 1975 images; (C, D) difference between the 2000 and 1987 images. (Color figure at the end of the volume). Two different processes govern the landuse and land cover changes in the Ust-Urt Plateau. On the one hand desertification processes have developed due to recent exploration and exploitation of the gas and oil reserves in the region (Ulmishek 2001). Consequently, large areas went through intensive technological desertification that means utilizing large amount of heavy-duty equipments, large-scale plants, and vehicles that damage the soil surface. On the other hand, following independence of the former Soviet states in 1991, and the imposition of difficult economic conditions with transition reforms, several major socio-economic changes occurred. Strong centralized government subsidy programs terminated, including the practice of guaranteed supplemental forage in cold winters and drought years. Farmers could no longer feed the livestock during the harsh winters, water wells were demolished, pumps were stolen or broken, and there were no longer transportation means to convey the animals to the markets in the central cities (Antonchikov et al. 2002). Consequently, drastic decline in livestock
populations were observed after 1991 that resulted in much less grazing pressure and hence recovery of the natural vegetation and rehabilitation of the land. The AND analysis that is illustrated in Figure 7 reveals that during the first period most of the change between 1987 and 1975 occurred within 5 km, more likely from the watering points, with minimal change in further lag. These results support the grazing gradient approach related to the average grazing distance of livestock from the drinking source. However, during the second period, this pattern is not evident. The change decreases gradually as a function of the lag distance up to approximately 16 km.
Figure 7. AND results plotted against distance from the water source. 4. CONCLUSIONS The Tasseled Cap-derived Brightness Index (BI) was selected to describe the spatial surface patterns since (1) it produced the best contrast; (2) it was originally designed to examine soil properties; and (3) it is able to compare between the different sensors with different spectral bands, as it reduces their different spectral bands to one normalized layer of BI values. Geostatistical analysis, based on the ordinary kriging interpolation technique, was found to be a suitable method for modeling the spatial patterns of land-use/land-cover, especially around watering points in arid and semi-arid regions. The reason is the similarity between the shape of the variogram and the directional change of many biotic, abiotic, and environmental variables along the grazing gradient. Temporal changes were effectively conducted by the index differencing technique. The study demonstrates the ability of spaceborne image analysis to follow after land-use/land-cover changes caused by dramatic socio-economic changes as occurred in Kazakhstan after its separation from Russia. Different effects are observed. One part of the area was degraded due to recent exploration and exploitation of the gas and oil reserves, while another part underwent rehabilitation processes due to dramatic reduction of the grazing pressure.
REFERENCES Addink, E. 2001. Change Detection with Remote Sensing. Relating NOAA-AVHRR to Environmental Impact of Agriculture in Europe. Wageningen University: PhD Thesis. 133 pp. Antonchikov, A.N., Bakinova, T.I., Dushkov, V.Y., et al. 2002: Desertification and Ecological Problems of Pasture Stockbreeding in the Steppe Regions of Southern Russia. Moscow: IUCN – World Conservation Union, Crist, E.P. & Cicone, R.C. 1984. A physically-based transformation of Thematic Mapper data the TM Tasseled Cap. IEEE Transactions on Geoscience and Remote Sensing 22: 256263. Curran, P.J. 1988. The semivariogram in remote sensing: an introduction. Remote Sensing of Environment 24: 493-507. Delhomme, J. P. 1979. Spatial variability and uncertainty in groundwater flow parameters: a geostatistical approach. Water Resources Research 15: 269-280. Friedel, M.H. 1997. Discountinous change in arid woodland and grassland vegetation along gradients of cattle grazing in cental australia. Journal of Arid Environment 37: 145-164. Hanan, N.P., Prevost, Y., Diouf, A., et al., 1991: Assessment of desertification around deep wells in the Sahel using satellite imagery. Journal of Applied Ecology 28: 173-186. Harrise, A.T. & Asner, G.P. 2003. Grazing gradient detection with airborne imaging spectroscopy on a semi-arid rangeland. Remote Sensnsing of Environment 55: 391-404. Hodder, R.M. & Low, W.A. 1978. Grazing distribution of free-ranging cattle at three sites in the Alice spring district, Central Australia. pp. Australian Rangeland Journal 1: 95-105. Huang, C., Wylie, B., Yang, L., Homer, C. & Zylstra, G. 2002. Derivation of a tasseled cap transformation based on Landsat 7 at-satellite reflectance. International Journal of Remote Sensing 23: 1741-1748. Huete, A.R. 1988. A Soil-Adjusted Vegetation Index (SAVI). Remote Sensing of Environments 25: 295-309. Jensen, J. R. 1986. Introductory Digital Image Processing, A Remote Sensing Perspective. New Jersey: Prentice-Hall. 379 pp. Kauth, J. & Thomas, G. S. 1976. The tasseled Cap - A Graphic Description of the SpectralTemporal Development of Agricultural Crops as Seen by LANDSAT. Proceedings of the Symposium on Machine Processing of Remotely Sensed Data (pp. 4B-41 - 4B-51). Purdue University of West Lafayette, Indiana. Lange, R.T. 1969. The piosphere: sheep track and dung patterns. Journal of Range Management 22: 396-400. Manzano, M.G. & Návar, J. 2000. Process of desertification by goats overgrazing in the Tamaulipan thornscub in north-eastern Mexico. Journal of Arid Environment 44: 1-17. Markham, B.L. & Barker, J.L. 1985. Spectral characterization of the LANDSAT Thematic Mapper sensors. International Journal of Remote Sensing, 6: 697-716. Matheron, G. 1971: The Theory of Regionalized Variables and its Applications, Cahiers du Centre de Morphologie Mathematique de Fontainebleau, No. 5, Ecole Superieure des Mines, Fontainebleau. 211 pp. Oliver, M.A., Webster, R. & Slocum, K. 2000. Filtering SPOT imagery by kriging analysis. International Journal of Remote Sensing 21: 735-752. Perkins, J.S. & Thomas, D.S.G. 1993. Spreading deserts or spatially confined environmental impacts - land degradation and cattle ranching in the Kalahari desert of Botswana. Land Degradation and Rehabilitation 4: 179-194. Pickup, G. & Chewings, V.H. 1994. A grazing gradient approach to land degradation assessment in arid areas from remotely sensed data. International Journal of Remote Sensing 15: 597-617. Pickup, G., Chewings, V.H. & Nelson, D.J. 1993. Estimating changes in vegetation cover over time in arid rangelands using Landsat MSS data. Remote Sensing of Environments 43: 243-263. Qi, J., Chehbouni, A., Huete, A. R., Kerr, Y. H. & Sorooshian, S. 1994. A modified soil adjusted vegetation index. Remote Sensing of Environments 48: 119-126.
Richardson, A. J. & Wiegand, C. L. 1977. Distinguishing vegetation from soil background information. Photogrammetric Engineering and Remote Sensing 43: 1541-1552. Rouse, J. W., Haas, R. W., Schell, J. A., Deering, D. W. & Harlan, J. C. 1974. Monitoring the vernal advancement and retrogradation (Greenwave effect) of natural vegetation. NASA/GSFCT Type III Final Report, Greenbelt, MD, USA. Ulmishek, G.F. 2001. Petroleum Geology and Resources of the North Caspian Basin, Kazakhstan and Russia. Denver: U.S. Geological Survey Bulletin 2201-B. USGS. Woodcock, C. E., Strahler, A. H. & Jupp, D. L. B. 1988a. The use of the variogram in remote sensing: I. Scene models and simulated images. Remote Sensing of Environments 25: 323-348. Woodcock, C. E., Strahler, A. H., & Jupp, D. L. B. 1988b. The use of the variogram in remote sensing: II. Real digital images. Remote Sensing of Environments, 25: 349-379. Yuan, D., Elvidge, C.D. & Lunetta, R.S. 1998. Survey of multispectral methods for land cover change analysis. In: Lunetta, R.S. and Elvidge, C.D. (eds.): Remote Sensing Change Detection, Environmental Monitoring Methods and Applications, pp. 21-39. Michigan: Ann Arbor Press.
Figure 5. Kriging interpolation maps for images shown in Figure 2 based on the Brightness Index values. Red tones indicate land degradation due to grazing and/or technological desertification.
Figure 6. Change detection results shown as maps and histograms. HR = high rehabilitation; LR = low rehabilitation; HD = high degradation; LD = low degradation. (A, B) difference between the 1987 and 1975 images; (C, D) difference between the 2000 and 1987 images.
