The role of crop classification in detecting wheat yield variation for index-based agricultural insurance in arid and semiarid environments

Abstract

The increasing availability of open-source and high-quality satellite data has facilitated market developments in the index insurance sector. So far, research and industry spheres have used administrative boundaries of units to estimate regional index values for insurance design. In areas with heterogeneous land use or land cover, however, these indices do not provide sufficient accuracy. This study analyzes potential accuracy gains from land-use classification that allow to design indices specifically for croplands and wheatlands. The validity of this approach is tested along conventional satellite-based products, including the Normalized Difference Vegetation Index (NDVI) and Land Surface Temperature (LST), as well as indices that are not yet widely used in crop insurance industry, like the Enhanced Vegetation Index (EVI), Green Chlorophyll Index (GCI) and Leaf Area Index (LAI). The study covers 2060 yield observations from 152 districts across Central Asia and Mongolia with irrigated, mixed and rainfed wheat farming systems. The results show that the majority of these indices are suitable for detecting wheat yield variations in rainfed and mixed agricultural lands, although they remain ambiguous in irrigated lands. Land-use classification and designing indices based on croplands and wheatlands noticeably increases the relationship between indices and wheat yields in rainfed and mixed lands. Notably, the LAI and GCI out-perform other well-known indices. Overall, freely available satellite data could serve as a good source for establishing index insurance products in Central Asia and Mongolia. Nevertheless, a careful assessment and selection of index and land use classification remains essential.

Keywords

• Climate resilience
• Climate adaptation
• Risk reduction
• Cropland mask
• Wheatland mask
• MODIS