Tracing mangrove forest dynamics of Bangladesh using historical Landsat data

Abstract: Present, accurate, and reliable estimation of mangrove forests in Bangladesh is limited. Former estimation of mangroves extent and density has been more or less restricted to Sundarbans and do not represent the whole country. In this study, a time series analysis was performed using Landsat images from four epochs, namely: 1976 (Landsat MSS), 1989 (Landsat TM), 2000 (Landsat ETM+) and 2015 (Landsat L8 OLI) to accurately quantify mangroves extent and density change/variation in the study area. An atmospheric correction using the Dark Object Subtraction method and a radiometric normalization using pseudo-invariant features was performed to reduce haze and sedimentation effects of the images. A standard approach to study changes in forest characteristics is to perform a time series analysis using images that have undergone a supervised classification. Results are indicating a gradual increase of forest area in most parts of the study area. Overall, the areal coverage increased by 3.10% (58140 ha) from 1976 to 2015, where 1.79% (58140 ha) of this increase took place from 2000 to 2015. The Sundarbans area turned out to be an exception. There, the mangrove forest area remained almost unchanged, although a little change (decrease by 1.03%) was found between 2000 and 2015. The study also claimed that one of the oldest mangrove forests in Bangladesh (Chakaria Sundarbans) had lost 4135 ha (32%) of forest area between 1976 and 1989. Similarly, a vegetation index (NDVI) analysis suggested that not only the area but also the density of the mangroves has changed over the years. The forests seem to have been denser in 1976 than in 1989. In 2000 the density appeared to have increased again, while decreased again in 2015. The study also found a substantial increase between 1989 and 2000 while a considerable density decreases in the Sundarbans region between 2000 and 2015. However, Mangroves area change was not significant in the context of classification uncertainty. A little error source was found due to the similar spectral reflectance between mangroves and non-mangrove vegetation, for example, in Patuakhali-Bhola. An accuracy assessment was performed using confusion matrices, showed maximum likelihood algorithm produce a better result for mangrove classes than other Land use/ Land cover classes. The overall accuracy of Landsat 8 OLI, ETM+, TM, and the MSS classified images (five classes) were found to be 97%, 87%, 80%, and 80% respectively, with Kappa values of 0.96, 0.82, 0.73, 0.74. Several possible factors such as cyclones, sedimentation and erosion, deforestation, shrimp and salt farming, and mangrove plantation were identified, which might be responsible for mangrove variations/changes in Bangladesh.