Accounting for temporal and spatial autocorrelation to examine the effects of climate change on vegetation greenness trend in China

Trend and attribution analysis of vegetation greenness is crucial to explain and predict ecosystem responses to climate change. The common practice to detect and explain greenness pattern from remote sensing time series is mostly based on pixel-by-pixel analysis, which often fails to account for spatial autocorrelation and may lead to spurious patterns. Here we applied the Partitioned Autoregressive Time Series (PARTS) method to the Normalized Difference Vegetation Index-3rd generation (NDVI3g) data and multiple climate datasets, and examined the climate effects on greenness trend in China. This method considers temporal and spatial autocorrelation structure, and aggregates pixel information to rigorously test the hypotheses about regional patterns. The results showed that greenness trends were strongly impacted by climate change, environmental background and their interactions. In regions with lower greenness, higher temperature, more precipitation and soil moisture, and lower vapor pressure deficit (VPD), the greening rate tends to be higher. For the whole China, long-term trends of temperature (P < 0.05) and soil moisture (P < 0.05) made significantly negative effects on greenness trend, while trend of precipitation (P < 0.05) and VPD (P < 0.001) made significantly positive impacts. But their effects strongly interacted with environmental background. The overall positive VPD impact was significantly enhanced with an increase in VPD level (P < 0.001), which was also supported by the significantly positive VPD impact in the northwestern arid regions (high VPD) and the significantly negative impact in the tropical and subtropical areas (low VPD). In the cold ecosystems, the change in soil moisture made significantly negative effect on greenness trend. This study provides new insights into the driving mechanisms of greenness change, which is useful to inform ecosystem modeling to make accurate predictions. Moreover, the analysis framework with PARTS method could be effectively applied to other regions or to analyzing other ecosystem responses to climate change.