Quantifying nonlinear responses of vegetation to hydro-climatic changes in mountainous Southwest China

Abstract

Read online

Vegetation plays an essential role in terrestrial carbon balance and climate systems. Exploring and understanding relationships between vegetation dynamics and climate changes in Southwest China is of great significance for ecological environment conservation. Nonlinear relationships between vegetation and natural factors are extraordinarily complex in Southwest China with complicated topographic conditions and changeable climatic characteristics. Considering the complex nonlinear relationships, the Random Forest (RF) and an integration of Convolutional Neural Networks and Long Short-Term Memory network (CNN-LSTM) were used with multi-source data from 2000–2020. Performance of two models were compared with precision indicators, and influence of topographic and hydro-climatic factors on vegetation was quantified based on the optimal models. Results revealed that the Normalized Difference Vegetation Index had a significant negative correlation with elevation and a positive correlation with land surface temperature and evapotranspiration. According to precision indicators, the RF model (RF3) built with longitude, latitude, elevation, slope, temperature, precipitation, evapotranspiration and surface solar radiation as inputs outperformed other models. Relative importance of the eight natural factors was quantified based on the RF3, and results indicated that elevation, temperature and evapotranspiration were major factors that influenced vegetation growth. Responses of vegetation toward climatic variables exhibited significant seasonal change, and there were different decisive factors, which influenced vegetation growth in forests, grasslands and croplands.

Keywords

• random forest
• long short-term memory
• convolutional neural networks
• relative importance
• nonlinear relationships