'Hot spot' assessment of land cover change and land degradation in CWANA using AVHRR satellite imagery

Land degradation is one of the most serious threats to the prosperity of rural populations in dryland areas. A major problem, in terms of combating land degradation in ICARDA's mandate region, is a shortage of reliable basic data on its extent and severity. In a dryland region as huge and diverse as CWANA, with limited reliable ground-based resource inventories and monitoring systems, remote sensing is a valuable tool for getting to grips with the highly complex issue of land degradation.
     Since 1982, the Advanced Very High Resolution Radiometer (AVHRR) satellite system has acted as a platform for the continuous space-based monitoring of the world's vegetation cover. Although covering a relatively short period of time and having a rather coarse resolution (8 km), it is the only consistent dataset to permit the detection of trends in land use/land cover change at global and regional scales. At the level of CWANA, a time series of AVHRR imagery could thus be used to identify 'hot spots' of land use/land cover change. This would overcome both the problems of the system's short time series and low spatial resolution, and the difficulties involved in distinguishing genuine trends in the land cover from short-term fluctuations in biomass (a result of year-to-year weather variations). ICARDA has now developed a specific methodology to achieve this.
     Six-hundred and twelve 10-daily composites of 8 km-AVHRR reflectance data, covering the period from January 1982 to December 2000, were downloaded from the relevant NASA website for band 1 (0.58-0.68 mm) and band 2 (0.725-1.1 mm). The data, consisting of separate subsets for Africa and Asia, were imported and merged to give complete coverage of the CWANA region. The Normalized Difference Vegetation Index (NDVI) was calculated and aggregated into monthly NDVI composites in order to reduce the effects of cloud cover. Additional corrections were made for 'noise' and sensor drift.
     In order to convert this temporal NDVI dataset into a land use/land cover classification, ICARDA developed two procedures. The first (described in the ICARDA Annual Report 2000, pp. 49-52) consisted of a hierarchical decision-tree, based on the average values of the mean and maximum NDVI, to take into account average weather conditions. The second procedure adjusted the NDVI thresholds of the hierarchical decision-tree for different agroclimatic zones, to take into account the fact that, in any particular year, the actual weather can differ substantially from the average.

     Using these procedures, 17 annual land-cover maps were produced for each year of the period 1982-1999. This dataset was further condensed into four maps, showing the majority land-cover classes for the key periods 1982-1984, 1987-1989, 1992-1994, and 1997-1999 (Fig. 18 gives an example). Depending on the value and sequence of the majority land-cover types, the following kinds of change were allocated to each pixel: 'noise', 'stable land use/land cover', 'stable land use/land cover mosaic' and 'change pattern'. Seventeen stable classes were recognized, as well as 66 change patterns. The latter were regrouped into 22 change classes, and four change trends:

1. 'intensification of agriculture' (e.g. a change from rainfed to irrigated agriculture)
2. 'intensification of natural vegetation' (an increase in vegetation biomass/density, e.g. a change from bare soil to grassland, or from grassland to forest);
3. 'retrenchment of agriculture' (a change from a more-intensive to a less-intensive formof agriculture);
4. 'retrenchment of natural vegetation' (a decrease in vegetation biomass/density).
   On the basis of this hierarchical classification 'change maps' were prepared for the CWANA region (Fig. 19), and areas belonging to individual change combinations, change classes and change trends were calculated.
   On the basis of this exploratory assessment, it was concluded that, in terms of area, the most dramatic changes in land cover in CWANA have occurred in the Sahel, followed by North Central Asia (Fig. 20). In the former region, approximately 75 million hectares changed from one land cover to another, in the latter 43 million hectares. In relative terms, the regions where the most change occurred are the Middle East and the Sahel, where about 14% of the land cover changed. But, even in the regions with the highest change in land cover, most of the land has remained stable during the period 1982-1999.
   At the onset of the study it was expected that, within CWANA, two major trends in land-use/land-cover change would occur: intensification of agriculture and land degradation, as indirectly evidenced by declines in biomass in both natural vegetation and agriculture. However, a more complex picture emerged, showing some remarkable and often unexpected trends in land-use/land-cover change in different subregions of CWANA. In most subregions intensification of agriculture is a major, if not the predominant trend (Fig. 21). The Near East in particular has experienced intensification of agriculture to a remarkable degree, mainly by the conversion of rainfed into irrigated croplands. However, retrenchment of agriculture and natural vegetation (potential indicators of land degradation) and intensification of natural vegetation are also important trends throughout CWANA, with significant differences being exhibited between subregions.

Fig. 18. Majority land-cover classes for different years: northwest Syria (on the left, the land-cover classes for 1987, 1988, 1989; on the right, the majority land-cover class for the period 1987-89, obtained by superimposing the 1987, 1988 and 1989 images upon each other).

Fig. 19. Example of a 'change map' showing the spatial distribution of land-cover change trends in the Near East and the Caucasus.

Fig. 20. Stability of land cover/land use.