A framework to detect digital changes in the mangrove forests

Abstract

Despite the particular importance in marine ecosystems and food chain, mangrove forests are subject to destruction due to rapid population growth, poor planning, and inconsistent economic development. Identifying changes in these ecosystems is the first step in their sustainable management. Therefore, in this study, we have tried to determine the appropriate method for determining the changes in mangrove forests using satellite data and for identifying appropriate thresholds for revealing these changes. Based on the surveying of temporal variation of these forests, the tidal conditions are the first problem in determining the digital change of these forests. Accordingly, in order to determine the appropriate digital change detection method, OLI images in 2015 and ETM + images for 2001 were obtained in the same tidal conditions. The preprocessing operations included geometric, and radiometeric correction was done on these data. Then, to enhance spatial resolution, the fusion is done in a way that does not affect the output histogram. In the next step, first the images were enhanced by applying the spectral indexes, then the hybrid classification was applied to the images to extract the mangrove forest. At this stage, the changes in these forests were determined by post-classification comparison. In the next step, by combining the mangrove forest area at both time intervals, the total forest mask was obtained. Then, the NDVI spectral index was appropriately considered by analyzing the coefficients of variation of the spectral vegetation indices. Then, in the mask area of the mangrove forest, the NDVI was used to perform algebra change detection methods including image difference, image ratio, regression. Also, to determine the appropriate thresholds for algebra operations, the thresholds were applied based on deviation from the mean for all methods, then accordingly the changes were detected. Finally, by 120 sampling points, areas with the decrease, increase, and no change trends were visited and then overall accuracy and kappa coefficients were determined. Based on the results, the post-classification comparison has the highest accuracy in detecting changes. It was also found that the threshold of twice standard deviation showed the best accuracy in outputs.