Retrieving leaf area index and crop yield from Landsat and Sentinel 2 earth observation satellites in Lake Tana sub basin, northwestern Ethiopia

Abstract

Crop growth monitoring during the growing season is important to understand crop growth status and productivity, essential in an effort to address food security problems particularly in smallholder farmers system. Conventional approaches are time-consuming, expensive and unable to cover large geographical areas. Integrating remote sensing data with ground measurement has been used a valuable opportunity to estimate leaf area index (LAI) for crop growth monitoring. The reliability of Earth observation derived LAI and yield is however influenced by small field size in fragmented smallholder farmers environment. The objective of this study was to estimate crop LAI from vegetation indices (VIs) from Sentinel-2 and Landsat 8 OLI data and test the potential to estimate crop yield of maize and rice crops in Lake Tana sub basin, Ethiopia. Regression equations were evaluated to estimate LAI from normalized difference vegetation index (NDVI), soil adjusted vegetation index (SAVI) and the two band enhanced vegetation index (EVI2). The results showed that LAI derived from Landsat 8 and Sentinel 2 exhibited strong nonlinear relationships with in-situ LAI for the two crops. In relatively waterlogged rice growing area the high spatial resolution of Sentinel 2 does not improve prediction accuracy where best performance was found from Landsat derived SAVI R2=0.88 and RMSE=0.38 m2m−2). In the highly fragmented maize growing environment Sentinel 2 derived EVI2 and SAVI showed better LAI prediction accuracy. Sensitivity analysis showed that NDVI has the highest sensitivity to LAI. The overall results provide sensors synergy where the lack of data in one sensor could substitute the other in data scarce agricultural landscape regions. However, models need to be tested under different environmental before applying on a larger scales. Further study should focus of assimilation of derived LAI in process-based and machine learning crop models to estimate LAI and yield. Graphical Abstract