<![CDATA[Land-use/land-cover (LULC) changes in new satellite towns such as Gading Serpong, Tangerang, Indonesia, represent a major driving force of spatial transformation and landscape fragmentation in metropolitan outskirts. Rapid urban development has converted agricultural lands, open spaces, and local water bodies into massive residential and commercial built-up areas. The objective of this study is to analyze a 10-year period of spatio-temporal LULC dynamics (2017–2026) using multispectral Sentinel-2 satellite imagery. For classification, the performance of a traditional pixel-based Random Forest (RF) classifier is compared against two state-of-the-art Deep Learning semantic segmentation models (U-Net and DeepLabv3+). The models successfully mapped the study area into four classes: Water Body, Vegetation, Built-Up Area, and Bare Land. Quantitative results indicate an expansion of the Built-Up Area from 212.0 Ha in 2017 to 2583.4 Ha by 2026. The primary converted land covers were Vegetation, which declined from 1759.2 Ha to 818.8 Ha, and Bare Land, which decreased from 1568.7 Ha to 193.8 Ha. The highest performance was achieved by the DeepLabv3+ algorithm, yielding an Overall Accuracy of 88.40% and a mean IoU of 76.24%. These results significantly outperform U-Net (63.69% mIoU) and the Random Forest baseline (44.52% mIoU) under complex urban spectral mixing conditions.]]>
Copyrights © 2026
