Ecological Indicators (Oct 2024)
Towards interpreting machine learning models for understanding the relationship between vegetation growth and climate factors: A case study of the Anhui Province, China
Abstract
The prediction of vegetation evolution and the understanding of its relationship with climate factors are essential for environmental protection, land use management, and policy planning. It is crucial to investigate accurate prediction methods for vegetation evolution and explore the impacts of climate factors. In this study, we developed machine learning (ML) based vegetation prediction methods by denoting vegetation status using normalized difference vegetation index (NDVI) and quantified the impacts of climate factors using interpretable methods. For the study region in this paper, i.e. the Anhui province in China, the proportions of areas with improved, stable and degraded vegetation status are 86.03 %, 8.13 % and 5.84 % respectively, and the NDVI evolution for the whole study region exhibits annual growth rate of 0.0031y-1. ML-based NDVI predictors exhibit R2 value exceeding 0.89 and MAE value below 0.1 for all lead times, which indicates the effectiveness of the ML-based prediction approaches. SHapley Additive exPlanation (SHAP) and Permutation Importance (PI) methods were utilized to provide insights into the black-box ML-based predictors. The results reveal that three temperature variables (minimum, maximum, and mean temperature) and precipitation are the key factors influencing vegetation growth. The increase of precipitation corresponds to an increase in vegetation, while higher minimum temperatures lead to a decrease in vegetation. When considering the combined contribution of minimum temperature and precipitation, it is shown that higher minimum temperature and larger amount of precipitation result in vegetation growth. On the contrary, lower minimum temperature and insufficient precipitation have negative impacts on vegetation. This work promotes the development of ML-based NDVI prediction approaches with transparency by taking advantages of interpretable methods. It provides understandings on how climate change influences vegetation growth in the Anhui Province.
