The use of multi-sensor satellite imagery to analyze flood events and land cover changes using change detection and machine learning techniques in the Barito watershed

Abstract

Indonesia is one of the countries in the world that is frequently affected by floods. Flood disasters can have various negative impacts and therefore need to be analyzed to determine prevention and mitigation measures. This study examined land cover change, flood detection, and flood distribution using multitemporal Sentinel-1 and Landsat-8 satellite imagery in the Barito watershed. A combination of change detection and the application of the Otsu algorithm was used to detect floodplains from Sentinel-1 imagery. Land use/land cover (LULC) changes are detected using a combination of change detection and machine learning in the form of a random forest algorithm. The overlay technique was used to analyze the distribution of floodplains. In this study, the floodplain in the study area was mapped to 109,623 ha. The change detection method detects a decrease in the areas of primary forest, secondary forest, fields, rice fields, shrubs and ponds, respectively, by 13,020 ha, 116,235 ha, 259 ha, 146,696 ha, 47,308 ha, and 9,601 ha. Settlements, bare land, plantations and water bodies increase by 14,879 ha, 64,830 ha, 218,916 ha, and 34,768 ha, respectively. Flooding was mainly found in the classes of rice fields, water bodies and primary forests.