Effects of disturbances on the spatiotemporal patterns and dynamics of coastal wetland vegetation

The coastal wetlands represented by mangroves are vulnerable to disturbances. To clarify the effects of disturbances quantitatively on mangroves, spatiotemporal changes in land use and land cover (LULC) were compared between the inside and outside of mangroves during 1988 and 2020 in Bangladesh by remote sensing. The mangrove region, which had diverse surface greenness due to natural and anthropogenic disturbances, were surveyed. Changes in greenness were estimated by five vegetation indices (VIs). To increase in the accuracy of land use classification, the five VIs were combined under a random forest (RF) classifier. The five VIs were: normalized difference vegetation index (NDVI), green NDVI (GNDVI), soil-adjusted VI (SAVI), green-red VI (GRVI) and enhanced VI (EVI), using data obtained from Landsat TM (30 m resolution) and Sentinel-2A (10 m) via Google Earth Engine. A Moran's I analysis showed a random pattern of training samples of LULC classes for integrated RF classification. The classification accuracy was evaluated using the reference point data with multiple comparisons of VIs and kappa coefficient. The area was classified into eight LULC classes. Although the respective VIs depicted unique characteristics, the accuracies of them were less than 89 %. The integrated classifier improved the accuracy of 96-97 % with kappa of 0.96 in 1988 and 0.97 in 2020. The VI-integrated RF was the most effective classifier for LULC, particularly for separating mangroves from homestead vegetation. The changes for 32 years showed that the dense mangrove occupied 42 % of surveyed area in 1988 but declined ca 622 km2 in 2020, whereas sparse mangrove and aquaculture increased by 487 km2 and 464 km2, respectively, out of the total area of 8,360 km2. The changes in LULC classes were mostly from dense mangroves to sparse mangroves, sparse mangroves to dense mangroves and from cropland to bare land and aquaculture. Inaccessible mangrove degradation should result from natural disturbances, i.e., tropical cyclones, while areas near forest boundaries were distorted by anthropogenic disturbances such as aquaculture development. The restoration of habitat quality, mangrove afforestation and sustainable aquaculture practices were effective indicators for improving the mangrove ecosystems. Since natural and anthropogenic disturbances induce substantial loss of mangrove vegetation, the VI-integrated RF classifier should be applied for monitoring LULC and for developing the sustainable management strategies.