Settlements extraction and spatiotemporal analysis with SAM and random forest from high-resolution remote sensing

The pace of urbanization has accelerated with the implementation of the rural revitalization strategy, thereby intensifying spatial changes in rural settlements. Consequently, accurate and efficient extraction of settlements has become crucial. Current methods often rely on manual calibration, which is time-consuming and labor-intensive. Remote sensing technology, capable of capturing surface features over a wide range with high accuracy, offers a viable solution. This study utilized high-resolution multisource remote sensing images from the Google Earth Engine (GEE) to extract settlements in Laoling City. A key contribution was the integrated application of the Segment Anything Model (SAM) algorithm and a random forest (RF) classifier for settlement extraction. Finally, the spatiotemporal changes in settlements across 2002, 2012, and 2022 were analyzed. The following results were obtained: (1) July, August, and September were identified as the best classification periods for settlement extraction. (2) Settlement extraction based on the SAM-RF method had the best classification results, with an overall accuracy of 0.98 and a kappa coefficient of 0.97. (3) Most township settlements expanded gradually, with the most remarkable change observed in Shizhong Subdistrict. The remote sensing-based settlement extraction approach in this study is significant for the coordinated development of urban and rural areas. It also provides strong support for the implementation of revitalization, resource management, and environmental protection. • SAM algorithm and Random Forest classification were combined to achieve an accuracy of 0.98 in settlement mapping. • July to September was identified as the optimal period for accurate settlement classification. • Settlement growth was quantified to provide critical insights for urban-rural planning and revitalization.