Analysis of spatial-temporal variations of grassland gross ecosystem product based on machine learning algorithm and multi-source remote sensing data: A case study of Xilinhot, China

Abstract

The Gross Ecosystem Product (GEP) of grassland ecosystems refers to the total value of the final products and services provided by the grassland of a particular region for human well-being each year. GEP serves as an important indicator for assessing the health of ecosystems. This study focused on Xilinhot, investigating the feasibility of integrating machine learning (ML) algorithms such as stepwise linear regression (SLR), random forest (RF), support vector regression (SVR), and k-nearest neighbor regression (KNN) with multi-source remote sensing data to model grassland GEP. Based on this, the study extracted the spatial-temporal variation characteristics and driving factors of GEP in the study area over the past 20 years. The results: (1) RF demonstrates significant predictive accuracy (R2=0.6409, RMSE=0.15), making it suitable for simulating grassland GEP in the study area. This underscores the potential of employing ML in conjunction with multi-source remote sensing data for grassland GEP estimation. (2) GEP exhibited a distribution pattern that gradually increased from the northwest to the southeast. Over time, there has been a consistent upward trend in GEP, peaking at 23.6 billion CNY in 2020. Among them, the proportion of the material and cultural service values increased annually, while the regulating service value remained stable. This indicates effective protection of the Xilinhot grassland ecosystem. (3) The driving force from climate change is greater than that from human activities, and it predominantly interacts with factors such as terrain and altitude, exerting significant effects on grassland ecosystems. This study can provide technical reference for the evaluation of grassland ecosystem services.