Rice Modeling Using Long Time Series of High Temporal Resolution Vegetation Indices in Nepal

Quick prediction and forecasting of crop yield during the growing season and before harvest are of important value in supporting decision makers in agriculture and food security at large. Indices derived from remote sensing have been approved rapid approach of monitoring and detecting crop growth conditions. The correlation between vegetation indices and crop yield has been well recognized and applied in many yield estimation study. High temporal resolution time series of vegetation index maps are generated up to daily coverage using multiple source remote sensing data at different spatial resolution. The result index maps are up-scaled to match the rice statistic at district level. The condition profiles represented in crop condition indices are smoothed using Best Index Slope Extraction (BISE) and double sigmoid model. Regression models have been trained over different periods of data since 2000. The condition profiles for the year under study is estimated during stages and used to estimate the yield. Validation of different machine learning models has been tested with different periods. The results showed that the profile series of recent years yield better estimation than using all the years since 2000. This may be caused by the yield increase over the years due to other factors such as rice farming technology development.