District-wise Statistical Yield Modelling of Wheat using Weather and Remote Sensing Inputs

Abstract

Accurate yield estimation has always been a matter of challenge to the scientific community especially so in the recent times due to the heightened risk of climatic variability. This study explored the statistical technique of fixed effect panel regression for estimation of the district-wise wheat yield using weather as well as satellite remote sensing indices. As wheat crop is sensitive to heat, temperature during the reproductive stage was used for modelling. Along with that trend adjusted vegetation condition index (VCITadj), temperature condition index (TCI) and vegetation health index (VHI) during the thermo-sensitive reproductive phase (TSP) was also used for modelling of wheat yield. The results show that, models developed with extreme temperature and remote sensing indices could capture the broad variation in district-wise wheat yield. The error was higher for extreme temperature based model as compared to the remote sensing based models. Among the remote sensing based models, VHI based one outperformed both the TCI and VCITadj based models which may be due to the reason that VHI combines both the information about greenness as well as temperature stress in it. The error in estimated yield varied based on the model but it was below 10% for all the districts for VHI based model. Further, it was seen that the accuracy was good for first year of prediction but it decreases for the second year. It indicates that the model should be used in a rolling mode, updating the parameters in each year before using it for next year.