An accuracy assessment of satellite-based cotton yield estimation using panel data regression: a case study of Uzbekistan

Abstract

Satellite-based yield estimation is crucial for spotting potential deficits in crop yields at an early stage, supports farm-level decision-making and early-warning systems, and is a prerequisite for index insurance markets. Precise satellite-based yield estimations are already established for important food crops like maize and wheat. However, for many cash crops like cotton, the accuracy of satellite-based yield estimation has not been scientifically tested, mainly due to their low biomass-yield correlation. This paper contributes to exploring the suitability of multiple vegetation indices based on Sentinel-2 imagery to estimate farm-level yields for one of these cash crops, cotton. We estimated various vegetation indices conjugated with the cotton crop phenology for the selected study area and compared them with farm-level panel data (n = 232) for the years 2016–2018 obtained from a statistical agency in Uzbekistan. Overall, we tested the suitability of the Normalized Difference Vegetation Index, the Modified Soil Adjusted Vegetation Index 2, the Red-Edge Chlorophyll Index and the Normalized Difference Red-Edge Index (NDRE). Among these indices, the NDRE index shows the highest fit with the actual cotton yield data (R2 up to 0.96, adj R2 = 0.95 and RMSE = 0.21). These results indicate that the NDRE index is a powerful indicator for determining cotton yields. Based on this approach, farmers can monitor crop growth, which in turn avoids crop loss and thereby increases productivity. This research highlights that a satellite-based estimate of crop production can provide a unique perspective which should improve the possibility of identifying management priorities to improve agriculture productivity and mitigate climate impacts.