Testing the influence of spectral resolution for binary change detection accuracies using simulated multispectral bands resampled from hyperspectral data

The ability from remote sensing data to observe the same areas at different times of acquisition is beneficial for change detection analysis. Various sensors from passive to active sensors have been employed. However, the development of satellite hyperspectral sensors brings the premise of a more accurate change detection analysis. Our study aims to test this premise by conducting the binary change detection analysis at different spectral resolution in Central Java Province. The PRISMA datasets were resampled into the spectral resolution of RapidEye (4-bands), Landsat-8 (8-bands), Sentinel-2 (13-bands), and MODIS (19-bands), apart from the original spectral resolution (237-bands) that were used for detecting change using the Principal Component Analysis and K-Means unsupervised analysis methods (PCA-Kmeans). Our results demonstrated that change detection analysis using the RapidEye and Sentinel-2 spectral resolution produced the highest overall accuracies with both showing the same accuracy of 72.04 . While the original 237 bands produced the accuracy of 65.74 . This indicated that the detection of major changes in the surface cover can be produced using 4 to 13 bands data. However, hyperspectral data are still potential to be used to detect slight changes in the surface cover, or to perform the unmixing based change detection analysis. © The Authors, published by EDP Sciences.