Spatiotemporal dynamics of land cover before, during, and after the earthquake, using Sentinel-2 imagery and random forest classification

This study aimed to analyze the spatiotemporal dynamics of land cover in Cianjur Regency, Indonesia, before (2021), during (2022), and after (2023) the 2022 earthquake, using multitemporal Sentinel-2 imagery and a Random Forest classifier implemented in Google Earth Engine.Six Sentinel-2 spectral bands and six spectral indices (NDVI, EVI, mNDWI, NDBI, NDBaI, and NBR) were used as classification variables.Land cover was classified into six major classes, namely built-up land, water bodies, wet agricultural land, open land, secondary dryland forest, and mixed dryland agriculture and shrubland.Accuracy assessment yielded overall accuracy values ranging from 72.12% to 83.65%, with kappa coefficients indicating substantial agreement.The results revealed that secondary dryland forest remained the dominant land cover class but decreased consistently from 236,170.58 ha in 2021 to 217,503.11ha in 2023.During the earthquake year, land cover changes were characterized by increased open land and reduced vegetation cover, whereas the post-earthquake period was marked by significant increases in built-up land and wet agricultural land accompanied by a decline in open land.Spatial analysis indicated that land cover changes were more intensive in the northern part of Cianjur Regency, where the earthquake epicenter and major settlement areas were located.The findings suggest that post-earthquake landscape transformation occurred through two distinct phases, namely a disturbance phase during the earthquake year and a reorganization phase during the recovery period.This study highlights the importance of multitemporal remote sensing approaches for capturing post-disaster landscape dynamics and provides valuable information for environmental rehabilitation, spatial planning, and disaster risk reduction.