Field Data Collection Methods Strongly Affect Satellite-Based Crop Yield Estimation

Crop yield estimation from satellite data requires field observations to fit and evaluate predictive models. However, it is not clear how much collection methods matter for performance. To this, we used maize estimates obtained with seven (two farmer estimates, two point transects, three crop cut methods) the “true yield” measured a full-field harvest 196 fields in districts Ethiopia 2019. We combination of nine vegetation indices five temporal aggregation growing season Sentinel-2 SR as predictors linear regression Random Forest Crop-cut-based models had highest model accuracy, similar that full-field-harvest-based When were training data, prediction gain was negligible, indicating very little advantage using remote sensing predict when quality low. Our results suggest estimate should be cuts or comparable high-quality measurements, which give better than low-quality sets, even larger numbers such are available.