Evaluating performance of prediction of vegetative drought using classic and recent sequence-based models

Drought is one of the deadliest and costliest natural disasters. Water deficit leads to vegetative drought, including low precipitation and temperature-caused evapotranspiration. Vegetation response to drought depends on climate, vegetation ecology, terrain, soil hydrology etc. environmental factors. Data based approach is useful as well as exploring physical mechanism of vegetation response to drought. While classic time series analysis and neural network methods can be used to predict vegetative drought, deep learning neural networks have great potentials in modeling the complex nonlinear characteristics between vegetative drought and putative environmental factors in view of their success in analysis of sequence data of other fields. In view of lack of evaluation on classic and recent sequence-based models, taking the CONUS as the study area, this article constructs models and evaluates their performance in predicting NDVI from some environmental variables such as meteorological and soil ones. The preliminary results show that recently developed deep neural network takes advantage over the classic models such as time series models and Elman neural network. Further investigation should be made in wider and latest models like space-time regression, GRU, BiLSTM, etc.