Spatio-temporal dynamics of mangrove cover based on Sentinel-2 imagery and Random Forest in Ketapang District, South Lampung, Indonesia

Mangrove forests fringe many tropical coastlines and provide important ecological and economic benefits, but they remain vulnerable to land conversion.This study analyzed the spatio-temporal dynamics of mangrove cover in Ketapang District, South Lampung Regency, Indonesia, between 2018 and 2026 using Sentinel-2 Level-2A imagery and a Random Forest (RF) classifier implemented in Google Earth Engine.Thirteen predictors were used: nine Sentinel-2 bands (B2, B3, B4, B5, B6, B7, B8, B11 and B12) and four spectral indices (NDVI, MNDWI, NDRE and NDTI).Five land cover classes (mangrove vegetation, non-mangrove vegetation, fish ponds, open land and water bodies) were mapped for 2018, 2020, 2022, 2024 and 2026.Against reference data interpreted from very-high-resolution imagery (2018-2024) and collected by field ground-truthing (2026), overall accuracy ranged from 91.79% to 96.34% and kappa from 0.89 to 0.95; error-adjusted areas with 95% confidence intervals were estimated from the error matrices.The SWIR-2 band (B12) had the highest Gini importance in every year.Mapped mangrove area increased from 196.64 ha in 2018 to 267.97 ha in 2026 (+71.33 ha), and the error-adjusted increase was 82.05 ± 13.42 ha.Decomposition of change showed that in 2018-2020 a small net change (+6.35 ha, within the classification uncertainty) was accompanied by a swap of 59.95 ha (30.48% of the 2018 mangrove area), whereas over the four intervals gross loss decreased from 29.97 to 18.88 ha and persistence increased from 84.76% to 92.70%.Overlay with the official rehabilitation sites (129.00 ha planted in 2020-2023) showed that the net increase occurred almost entirely within these sites (+80.83ha, compared with -9.47 ha outside them) and that their share of the gross gain rose from 16.0% before the main planting to 58.3-77.5% in 2020-2024; this spatial association is consistent with a substantial contribution of rehabilitation planting, although causality was not tested.The results show how frequent Sentinel-2 mapping combined with transition analysis and uncertainty estimation can support the monitoring of mangrove rehabilitation.