Modelling paddy rice yield using MODIS data

Paddy rice is a major source of atmospheric methane, yet vast amounts of rice continue to be grown in order to meet increasing global food demand. Accordingly, paddy rice yield estimation at a large scale is crucial to ensure food security and environmental protection. To address this, we have developed a rice yield estimation model using remote sensing data. First, we created an 8-day NPP model for paddy rice based on the MODIS NPP algorithms and calibrated our models using the MODIS annual NPP product and the more reliable radiation use efficiency (RUE) of rice. Thereafter, we combined our 8-day NPP model and calibrated 8-day NPP models with MODIS GPP products, and integrated these to form crop yield estimation models incorporating RUE and harvest indices (HI). Finally, based on the paddy rice region derived from high-resolution land use data and detailed field calibration, we applied these models to Liling County, China, where paddy rice cultivation is extensive. We evaluated our results with respect to a reference dataset calculated based on the statistical unit rice yield and the percentage of paddy rice area in a 1 × 1 km grid. Our results show that the rice yield estimate obtained from the 8-day NPP model calibrated with RUE = 2.9 g MJ−1 agrees more closely with the reference data than that obtained using the other models, with relative error and RMSE of less than 5% and 5 × 104 kg, respectively. Based on the uncertainty and sensitivity analysis of each input in proposed models, we believe that it is reasonable to improve the accuracy of the rice yield with the supplement of field data, especially for RUE and HI.