Irrigation mapping in Africa and the Near East: physical-based versus supervised classification

Monitoring irrigated areas has received considerable attention given the importance for e.g. food security, and for the management of water resources. However, the current extent of irrigated areas at continental to global scale is still uncertain. Existing maps, especially covering large areas, are mainly derived from country-level statistics.

Remote sensing has proven to be a valuable tool to map agricultural production areas, yet mapping irrigated areas has proven to be challenging. On the one hand, agricultural areas are difficult to map in general, while on the other hand the step from land cover (cropland) to land use (irrigation practice) requires additional data sources other than the optical satellite data.

Recently, a generic irrigation mapping method was developed in the framework of the FAO WaPOR data portal (https://wapor.apps.fao.org), to provide irrigation maps at a continental scale for Africa and the Near- East on a seasonal and annual basis. The method combines information on (i) the land cover, (ii) the phenology as obtained from time series analysis, (iii) actual Evapotranspiration (ETa), and (iv) precipitation. In short, the Water Deficit Index (WDI) is computed as the ratio of the seasonal precipitation over the seasonal ETa, with values below one when the water consumption is larger than the water availability on a seasonal basis. Although very good results were obtained with a WDI threshold of 0.9 on a continental scale, some issues remain, especially in areas where the water availability and the consumption are very similar, or where only supplementary irrigation is applied.

In addition to this physically-based approach, supervised classification methods have proven to be a suitable irrigation mapping method as well, yet suffer from the drawback that they require a large amount of reference information. To evaluate which method is best suited to distinguish between irrigated and rainfed agriculture, a comparison between both is made for three different irrigation schemes: (i) the Nile delta in Egypt, which receives full irrigation, (ii) the Bekaa valley in Lebanon where both irrigated and rainfed croplands are present, and (iii), the Koga region in Ethiopia, which is rainfed during the rainy season, but irrigated in the dry season.

From the analysis, it is clear that for regions with very little precipitation (Nile delta), the physical-based method is very well suited to map the irrigated areas, without the need for ground reference information. However, in more complex systems, such as the Bekaa valley, the confusion between irrigated and rainfed areas is quite substantial for the physical-based method, and the supervised classification method obtained more promising results.

This may suggest that for continental or global irrigation mapping, a combination of both methods is desirable. Irrigated areas in regions with low precipitation could best be mapped based on the WDI approach, while in more complex areas, supervised classification methods may be required. This would strongly reduce the need for detailed reference data over large areas, while maintaining a high mapping accuracy.