Global Natural and Planted Forests Mapping at Fine Spatial Resolution of 30 m

Planted forest expansion often encroaches upon natural forests, leading to numerous environmental and social problems and altering the carbon sequestration capacity. Mapping natural and planted forests accurately is pivotal for achieving carbon neutrality and combating climate change. However, global mapping of natural and planted forests at fine spatial resolution remains an unmet requirement, mainly due to the insufficient number of training samples often needed in land cover mapping methods. This study presents a novel approach for automatically generating training samples and for accurately mapping the global distribution of natural and planted forests at 30-m spatial resolution in 2021. More than 70 million training samples were generated based on the distinct disturbance frequency of planted and natural forests across the 30-m Landsat images from 1985 to 2021 derived using a well-established time-series change detection method. These training samples encompass diverse Landsat and auxiliary data features, including spectral, structural, textural, and topographic attributes. Subsequently, locally adaptive random forest classifiers were trained using these samples and achieved an overall accuracy of 85% when validated against independent visually interpreted reference data. Based on the produced map, the proportions of the natural and planted forests for all the continents and countries were consistent with the Global Forest Resources Assessment 2020 statistics, indicated by regression slopes of 1.0050 and 1.2432, respectively. The generated training samples can be employed to update the global map of natural and planted forests. The produced map is expected to enhance our comprehension of variations in carbon sequestration, biodiversity maintenance, climate change mitigation, and other factors between natural and planted forests. Data presented in this study is publicly available at https://doi.org/10.5281/zenodo.10701417 .