Abstract
Accurately distinguishing plantation forests from natural forests is a crucial prerequisite for assessing forest ecosystem services, carbon sink functions, and the effectiveness of ecological restoration. However, in arid regions such as Xinjiang, China, long-term, high-resolution mapping of plantation and natural forests remains highly uncertain due to fragmented forest patches, pronounced topographic heterogeneity, and the relative scarcity of ground reference samples. In this study, we developed a 30-m classification and mapping framework tailored for plantation and natural forests in Xinjiang based on the Google Earth Engine platform. This framework organically integrated Landsat series remote sensing imagery, forest resource inventory data, and visually interpreted samples from Google Earth high-resolution imagery, while extracting multi-source features including topographic, spectral, and textural variables. A total of 7621 forest-type sample points were collected across seven periods, namely 1990, 1995, 2000, 2005, 2010, 2015, and 2020. Subsequently, forty core features were selected from 269 candidate variables, and a Random Forest classifier was employed to finely identify plantation and natural forests. The results showed that the proposed classification model exhibited excellent stability and reliability. The overall accuracy for all periods exceeded 86%, with Kappa coefficients ranging from 0.67 to 0.88 and F1-scores all surpassing 0.70. From 1990 to 2020, the area of plantation forests in Xinjiang significantly increased from 1.778 × 105 ha to 5.344 × 105 ha, with its expansion mainly occurring along oasis margins, river valleys, and shelterbelt construction zones. Concurrently, the area of natural forests increased from 1.4875 × 106 ha to 2.2363 × 106 ha, and these forests were primarily distributed in alpine zones and river source regions. The exploratory structural equation model results indicate that plantation forest and natural forest dynamics may follow different associative pathways: natural forest change is more closely related to climate–hydrological conditions, particularly precipitation, whereas plantation forest expansion is more strongly associated with population growth and socioeconomic activities. Ultimately, this study not only provides 30-m spatial-resolution data support for the long-term dynamic monitoring of plantation and natural forests in the arid region of Xinjiang, but also offers a vital scientific basis for regional forest carbon sink accounting, ecological engineering assessment, and scientific forest resource management.