Comparison of Modeling Grassland Degradation with and Without Considering Localized Spatial Associations in Vegetation Changing Patterns

Grassland ecosystems worldwide are confronted with degradation. It is of great importance to understand long-term trajectory patterns of grassland vegetation by advanced analytical models. This study proposes a new approach called a binary logistic regression model with neighborhood interactions, or BLR-NIs, which is based on binary logistic regression (BLR), but fully considers the spatio-temporally localized spatial associations or characterization of neighborhood interactions (NIs) in the patterns of grassland vegetation. The BLR-NIs model was applied to a modeled vegetation degradation of grasslands in the Xilin river basin, Inner Mongolia, China. Residual trend analysis on the normalized difference vegetation index (RESTREND-NDVI), which excluded the climatic impact on vegetation dynamics, was adopted as a preprocessing step to derive three human-induced trajectory patterns (vegetation degradation, vegetation recovery, and no significant change in vegetation) during two consecutive periods, T1 (2000–2008) and T2 (2007–2015). Human activities, including livestock grazing intensity and transportation accessibility measured by road network density, were included as explanatory variables for vegetation degradation, which was defined for locations if vegetation recovery or no significant change in vegetation in T1 and vegetation degradation in T2 were observed. Our work compared the results of BLR-NIs and the traditional BLR model that did not consider NIs. The study showed that: (1) both grazing intensity and road density had a positive correlation to vegetation degradation based on the traditional BLR model; (2) only road density was found to positively correlate to vegetation degradation by the BLR-NIs model; NIs appeared to be critical factors to predict vegetation degradation; and (3) including NIs in the BLR model improved the model performance substantially. The study provided evidence for the importance of including localized spatial associations between the trajectory patterns for mapping vegetation degradation, which has practical implications for designing management policies to counterpart grassland degradation in arid and semi-arid areas.