Association analysis between spatiotemporal variation of net primary productivity and its driving factors in inner mongolia, china during 1994-2013

Vegetation Net Primary Productivity (NPP) is an important indicator for agriculture production. Understanding spatio-temporal dynamics of NPP and their driving factors have attracted much attention from both academic field and practical applications. In this paper, we coupled spatial statistics and a new approach called accumulated density map analysis (ADMA) to explore spatio-temporal variations in NPP distribution and possible contributing factors to the variations in Inner Mongolia Autonomous Region (IMAR), China. Density of Spatiotemporally Aggregated Clustering (D-STAC) index of NPP distribution, as output from ADMA, was proposed to indicate the impact of local factors on NPP. The study showed that spatially averaged NPP did not exhibit significant changes over time, but inter-annual variability of NPP presented critical spatial heterogeneity. Local spatial association analysis, as a preliminary step for ADMA analysis, detected two positive autocorrelation patterns, namely H-H (high NPP enclosed by high NPP) and L-L (low NPP enclosed by low NPP), and two negative autocorrelation patterns, including H-L (high NPP surrounded by low NPP) and L-H (low NPP surrounded by high NPP), for localized places in the study area. While positive autocorrelation patterns were found to dominate most parts of the study area, D-STAC for negative autocorrelation patterns was closely associated with Neighborhood to Cities (NC), an index for urbanization level. To evaluate the relationship between the NPP variation and possible driving factors, local regression analysis using geographically weighted regression (GWR) revealed that NPP largely had positive correlation with precipitation, but showed substantial spatial variations in the relationships with temperature and NC. We concluded that, through LISA, ADMA and GWR, the associations between the spatio-temporal NPP variations and their driving factors could be examined under localized context.