Mapping Bamboo Forest Dynamics with Long-Term Landsat Stacks and Samples Migrated from Percentile-Based Head/Tail Break

Mapping bamboo forest dynamics is essential for forest resource management and understanding vegetation succession, but it is hindered by the lack of historical samples. To address this challenge, this study developed a percentile-based head/tail break (P-HTB) method to reconstruct historical samples and produced bamboo forest maps for Jiangxi Province at five-year intervals from 1990 to 2020. Compared with the OTSU and original head/tail breaks (HTB) methods, P-HTB performed better in automatic threshold determination across most target years. Using the migrated samples and random forest algorithm, bamboo forest dynamics in Jiangxi Province from 1990 to 2020 were mapped at 5-year intervals. The results showed that the bamboo forest area more than doubled over the 30-year period, from 57.74 × 104 ha to 120.78 × 104 ha. Expansion was most pronounced in hilly and lower-elevation areas, driving the elevation centroid of bamboo forests downward by 89 m, yet increases were also observed at high elevations. This expansion came largely at the expense of evergreen broadleaved forests. This study provides a reliable long-term bamboo forest dataset for Jiangxi Province and a transferable methodology for large-scale vegetation monitoring, thereby offering support for sustainable forest management and improved land cover mapping.