Authors: Shaobo Zhong*, Beijing Research Center of Urban Systems Engineering, Zhanya Xu, Faculty of Information Engineering, China University of Geosciences, China, Liping Di, CSISS, George Mason University, Ziheng Sun, CSISS, George Mason University, Genong Yu, CSISS, George Mason University
Topics: Hazards, Risks, and Disasters, Agricultural Geography
Keywords: Drought, vegetation index, temperature, soil moisture, deep learning, time series
Session Type: Poster
Start / End Time: 8:00 AM / 9:40 AM
Room: Lincoln 2, Marriott, Exhibition Level
Presentation File: Download
Drought is one of the deadliest and costliest natural disasters. While remote sensing of vegetation is widely applied to monitor drought, there is delay between drought hazards and the monitored vegetation stress. To promptly assess the effect of drought on vegetation, it is necessary to connect the gap between meteorological observations (or forecast) and subsequent vegetation state conditions. Solution to this problem is greatly helpful for efficient and effective agricultural drought monitoring and early warning. In view of historical memory of RNN (recurrent neural network), we explored the prediction of vegetative drought with meteorological observations as input sequences. Taking the contiguous United States (CONUS) as the study area, long-term (more than 30 years) weekly vegetation indices (NDVI) and meteorological and soil moisture variables were applied to train, validate and test several popular deep learning neural network models including Elman, LSTM (long- and short-term memory), bidirectional LSTM, GRU. The results also were compared to several classic time series methods such as ARIMA and ARIMAX. The results should promote our insight into the interaction between vegetation and environments such as climate and vegetation/ecology, and should improve upon current techniques used, depending only on satellite data for vegetation drought monitoring and early-warning, and other climatic impacts. The results also are helpful to provide important clues for physiological and phenological analysis of vegetation.