Vegetation mapping of Yunnan Province by integrating remote sensing, field observations, and models

Vegetation maps are crucial for ecologists and decision-makers, providing essential information on the spatial distribution of various vegetation types to support ecosystem exploration and management. Despite advancements in Earth observation and machine learning enabling large-scale vegetation mapping, creating detailed and accurate maps in biodiversity hotspots remains challenging due to significant environmental heterogeneity and frequent human disturbances. The lack of sufficient ground-based data and complex climate-vegetation interactions further limits mapping accuracy. In this study, we developed an integrated framework for multi-source data fusion to enhance vegetation mapping and validation in Yunnan Province, a global biodiversity hotspot region in Southwest China. The mapping process involved four key steps: (1) vegetation classification using random forest and Landsat imagery, (2) boundary calibration based on a locally calibrated static climate-vegetation model, (3) patch correction with independent forest inventory data, and (4) validation using adequate field observations. This approach enabled the mapping of 17 vegetation types and 44 subtypes in Yunnan Province (1:50000), categorized based on the growth-form composition of dominant species of the community. The overall accuracies were 0.747 and 0.710 for natural vegetation types and subtypes, and 0.905 and 0.891 for artificial types and subtypes. This high-resolution map enhances our understanding of vegetation distribution and ecological complexity in this region, offering valuable insights for policymakers to support conservation efforts and sustainable management strategies.