A multi-scale spatiotemporal modeling approach to explore vegetation dynamics patterns under global climate change

Given the complexity of vegetation dynamic patterns under global climate change, multi-scale spatiotemporal explicit models are necessary in order to account for environmental heterogeneity. However, there is no efficient time-series tool to extract, reconstruct and analyze the multi-scale vegetation dynamic patterns under global climate change. To fill this gap, a Multi-Scale Spatio-Temporal Modeling (MSSTM) framework which can incorporate the pixel, scale, and time-specific heterogeneity was proposed. The MSSTM method was defined on proper time-series models for multitemporal components through wavelet transforms. The proposed MSSTM approach was applied to a subtropical mountainous and hilly agro-forestry ecosystem in southeast China using the moderate resolution imaging spectroradiometer enhanced vegetation index (EVI) time-series data sets from 2001 to 2011. The MSSTM approach was proved to be efficient in characterizing and forecasting the complex vegetation dynamic patterns. It provided good estimates of the peaks and valleys of the observed EVI and its average percentages of relative absolute errors of reconstruction was low (6.65). The complexity of the relationship between vegetation dynamics and meteorological parameters was also revealed through the MSSTM method: (1) at seasonal level, vegetation dynamic patterns are strongly associated with climatic variables, primarily the temperature and then precipitation, with correlations slight decreasing (EVI–temperature)/increasing (EVI–precipitation) with altitudinal gradients. (2) At inter-annual scale, obvious positive correlations were primarily observed between EVI and temperature. (3) Despite very low-correlation coefficients observed at intraseasonal scales, considerable proportions of EVI anomalies are associated with climatic variables, principally the precipitation and sunshine durations.