Mapping crop yield spatial variability using Sentinel-2 vegetation indices in Ethiopia

Abstract

Crop yield prediction before harvest is a key issue in managing agricultural policies and making the best decisions for the future. Using remote sensing techniques in yield estimation studies is one of the important steps for many countries to reach their agricultural targets. However, crop yield estimates rely on labor-intensive surveys in Ethiopia. To solve this, we used Sentinel-2, crop canopy analyzer, and ground-truthing data to estimate grain yield (GY) and aboveground biomass (AGB) of two major crops, teff and finger millet, in 2020 and 2021 in Ethiopia’s Aba Gerima catchment. We performed a supervised classification of October Sentinel-2 images at the tillering stage. Among vegetation indices and leaf area index (LAI) used to predict teff and finger millet GY and AGB, the enhanced vegetation index (EVI) and normalized-difference VI (NDVI) provided the best fit to the data. NDVI and EVI most influenced teff AGB (R2 = 0.87; RMSE = 0.50 ton/ha) and GY (R2 = 0.84; RMSE = 0.14 ton/ha), and NDVI most influenced finger millet AGB (R2 = 0.87; RMSE = 0.98 ton/ha) and GY (R2 = 0.87; RMSE = 0.22 ton/ha). We found a close association between GY and AGB and the satellite EVI and NDVI. This demonstrates that satellite images can be employed in yield prediction studies. Our results show that satellite and crop canopy analyzer-based monitoring can facilitate the management of teff and finger millet to achieve high yields and more sustainable food production and environmental quality in the area. The results could be reproducible under similar study catchment conditions and boost crop yield. Extrapolation of the models to other areas requires local validation. To improve crop monitoring for farmers and reduce expenses, we suggest integrating time series Sentinel-2 images along with LAI obtained from crop canopy analyzers collected during the cropping season.