Simulation model of vegetation dynamics by combining static and dynamic data using the gated recurrent unit neural network-based method

• DOI: 10.1016/j.jag.2022.102901
• Publication Date: 2022-07-06T02:20:56Z
• Authors: Pu Zhang; Zhipeng Li; Heyu Zhang; Jie Ding; Xufeng Zhang; Rui Peng; Yiming Feng
• Keywords: Ecotone; Dynamic data

The simulation of vegetation dynamics is essential for guiding regional ecological remediation and environmental management. Recent progress in deep learning methods has provided possible solutions to simulations. gated recurrent unit (GRU) one the latest algorithms that can effectively process dynamic data. However, static data, which typically coexist datasets changes, are processed indistinguishably. To efficiently extract spatiotemporal patterns improve our ability simulate potential we introduced GRU into further amended original structure according characteristics dataset. new model, model (VDM), independently data using a more appropriate algorithm, thereby improving accuracy. Moreover, presented test applied Luntai Desert-Oasis Ecotone Northwest China compared performance VDM with baseline models. results showed produced 7.51% higher coefficient determination (R2) value, adjusted R2 16.67% lower mean squared error, 10.78% absolute error than those GRU, best model. proposed first GRU-based detect time-order factors by comprehensively considering information affects changes. flexibility VDM, combination wide availability from different sources, aids broader application VDM.