Comparative performance of semi-empirical based remote sensing and crop simulation model for cotton yield prediction

Early prediction of crop yield serves as a strategic plan for policymakers in making suitable import–export policies. In this research, remote sensing driven semi-empirical light use efficiency (LUE) model and a process-based crop simulation model (CSM) approach were attempted to predict cotton yield in the five districts of Maharashtra state, India. Time-series satellite data from different earth observation (EO) missions were retrieved from 2011 to 2017. Satellite-based measurements of fraction of absorbed photosynthetically active radiation (FAPAR) and photosynthetically active radiation (PAR) data were directly incorporated into the LUE model. The LUE model approach used observed minimum (0.07) and maximum (0.12) lint harvest index (HI) values to predict the cotton yield. On the other hand, the CSM model was parameterized with observed field management, soil, weather, and cultivar data for crop yield prediction. The inter-evaluation of models has revealed that the LUE model with HI value 0.07 (HI0.07) performed better when compared to the LUE model with HI value 0.12 (HI0.12) and CSM model. LUE model HI0.07 had shown better accuracy in terms of root mean square error (RMSE) of 154 kg ha−1 and correlation coefficient value of (R2 = 0.46) when compared with LUE model HI0.12 (RMSE 310 kg ha−1 and R2 = 0.41) and CSM model (RMSE 235 kg ha−1 and R2 = 0.38). The predicted yield from the models was found satisfactory for the normal year with acceptable accuracy. However, overestimation of yield was seen for the years affected by drought. This study has shown the potential of geospatial data that can be coupled with crop models to predict the crop yield with limited field observations.