This study reassessed the spatio-temporal dynamics of horticultural land in the Upper Jeneberang Watershed, South Sulawesi, Indonesia, using a fully regenerated and auditable 2015-2025 Landsat 8 workflow. Landsat Collection 2 Level-2 imagery was restricted to a common June 1-October 31 seasonal window for 2015, 2020, and 2025, and six spectral bands, NDVI, NDWI, NDBI, elevation, and slope were classified using year-specific Random Forest models (500 trees; seed = 42). For 2015 and 2020, the archived 300 historical reference points for each year were stratified into independent training and validation subsets before model fitting; 2025 used a polygon-level split of 127 reference polygons before pixel extraction. Held-out validation produced Overall Accuracy/Kappa values of 67.03%/0.522 (2015; n = 91 points), 74.44%/0.654 (2020; n = 90 points), and 83.78%/0.799 (2025; n = 37 polygons). Horticultural Producer's/User's Accuracy was 60.00%/60.00%, 52.94%/52.94%, and 91.67%/100.00%, respectively. Using a common valid-data mask, mapped horticultural area was 3,968.47 ha in 2015, increased to 4,650.97 ha in 2020, and declined partially to 4,374.77 ha in 2025, yielding a decade-scale mapped net increase of 406.29 ha (10.24%). Forest-to-horticulture was the largest mapped source pathway during 2015-2025 (1,192.01 ha; 58.63% of gross horticultural gain), followed by shrubland (347.92 ha; 17.11%). Landscape metrics showed increasing patch number and edge density, declining mean patch area, and lower aggregation by 2025, indicating proliferation of smaller, more edge-rich horticultural patches rather than a simple increase in connectivity. All class and transition areas are reported as mapped estimates rather than error-adjusted estimates of true area.