Forecasting vegetation index based on vegetation-meteorological factor interactions with artificial neural network

Vegetation index derived from remote sensing measurement servers as the significant reference for crop growing monitor and agricultural disaster forecasting. Vegetation index forecasting at long lead time and appropriate spatial scale is critical for decision making to mitigate the impacts from agricultural disaster. In previous studies, vegetation index forecasting has been studied and implemented through different methodologies. Nevertheless, either the accuracy or confidence of forecasting result was affected by model forecasting capability. Artificial neural network (brief as ANN) is capable to model highly complex non-linear patterns of ecological processes. Moreover, its implementation is less restricted by strict assumption regarding variable distribution and model specification. In this paper, an ANN-based forecasting model was proposed to predict the NDVI of grassland within the U.S Great Plains. The forecasting model finds its ground in the historical patterns of lagged correlations between vegetation index and meteorological factors. NDVI forecasting performance at different lead time (i.e., 16, 32, 48 and 64 days) and at different regions (i.e., four ecological zones) is evaluated through cross-validation and comparing with naîve forecasting model. According to forecasting results, the forecasting performance is satisfied with high accuracy. Results of this study will contribute to the knowledge of vegetation stress monitoring and also serve as references for agricultural drought forecasting in the future study.