A new approach for monitoring spatial and temporal changes in forest types in subtropical regions with sample migration and multi-source remote sensing data

Abstract

Accurate and cost-effective monitoring of forest-type distribution and dynamics is crucial for managing the rich resources and complex ecosystems of subtropical regions. However, the region’s complex terrain and climate conditions pose significant challenges to precise forest classification and change detection. This study leverages the Google Earth Engine (GEE) platform to analyze long-term Landsat imagery and diverse environmental data for mapping forest/non-forest areas and five distinct forest types in Yunnan Province from 2000 to 2020. We then conducted a change detection analysis using a post-classification comparison method. We propose a novel workflow for processing ground survey data to generate stable reference samples, which are then used with a transfer learning approach to construct a multi-year sample library. Subsequently, a high-dimensional feature set was developed using multi-source data, employing classifiers like random forest (RF), support vector machine (SVM), and Extreme Gradient Boosting (XGBoost) to classify forests/non-forests and five forest types, with subsequent accuracy assessment. Finally, a spatiotemporal filtering process was applied to optimize the multi-temporal classification results, thereby enhancing their accuracy for the final analysis of spatiotemporal changes. The classification consistently achieved over 93% overall accuracy for forest/non-forest mapping and over 76% for classifying specific forest types. Change analysis revealed that 97.21% of forests remained unchanged from 2000 to 2020, with a 16.26% probability of non-forest converting to forest, indicating an overall increase in forest area in Yunnan Province. Although forest types underwent changes, they remained relatively stable, with evergreen broadleaf forests rarely converting to other types, maintaining 90.98% stability. Changes in forest types demonstrated strong spatial consistency, correlating with the environmental conditions conducive to their growth. This research offers a robust framework and practical guidance for large-scale forest mapping and change monitoring, with findings that hold direct implications for sustainable forest resource management and development in Yunnan Province. Furthermore, the proposed methodology can be adapted to identify more diverse vegetation types and extended to other mountainous forest ecosystems worldwide.

Keywords

• Forest type mapping in Yunnan Province
• landsat imagery
• sample migration
• machine learning
• change detection
• Yunnan