Spatio-temporal distribution of NDVI and its influencing factors in China

Normalized Difference Vegetation Index (NDVI) is an important indicator reflecting the state of regional climate and environment, which is affected by precipitation, temperature, soil water content, and so on. This study analyzed the influence of different factors on NDVI through the spearman correlation coefficient (SCC) and standard regression coefficient (SRC), carried out climate zoning in China through principal component analysis (PCA) and cluster analysis, simulated the NDVI value through deep learning algorithm, and analyzed the spatial mutation positions of different factors through concept drift detection. The research results show that the main factor that affects NDVI in most areas of China is surface soil moisture (SSM), followed by temperature, while in arid areas root zone soil moisture (RZSM) and profile soil moisture (PSM) are the main influencing factors. The climate zones of China can be roughly divided into northern, southern and northwestern inland areas, while there are transitional zones between different main zones. The deep learning algorithm can simulate the NDVI value very well, and the Nash efficiency coefficients (NSE) during training and verification periods are 0.893 and 0.833, respectively. However, the details of some areas are rough. This shows that the selected impact factors can basically determine the status of NDVI, and these impact factors can be used to predict the changes in NDVI. The environmental drift areas discovered by concept drift detection are mainly in the Heilongjiang River Basin in Northeastern China, Yunnan, Sichuan, and Southern Tibet in Southwestern China, and Gansu, Qinghai and Xinjiang in Northwestern China. This study provides an important reference for the analysis of environmental regional zoning, climate change detection and simulation.

Automated summary

NDVI is influenced mainly by surface soil moisture and temperature, China's climate zones are divided into northern, southern, and northwestern inland areas, the deep learning algorithm can simulate NDVI values well, and concept drift detection identified environmental drift areas.