Sentinel-2A MSI and Landsat-8 OLI spectral reflectance comparison in tropical environments: a preliminary research for data fusion

Abstract

The multi-sensor approach has been widely implemented for earth monitoring in many applications to enable regular space-based information. Spectral comparison between sensors is an essential preliminary step before conducting a data fusion for such applications. Comparison of medium spatial resolution data, e.g., Landsat-8 and Sentinel-2 were mostly conducted in subtropical regions, nevertheless, limited in equatorial tropical environments. This study utilised the same day of Landsat-8 and Sentinel-2 data to investigate the spectral changes over relatively unchanged environment conditions during 3-10 minutes of differences acquisitions time of Landsat-8 and Sentinel-2. The study area was the Citarum River Basin area in Indonesia with monsoonal climate type and various landcover, e.g., Lake, agricultural area, rain forest, and settlement areas. Surface reflectance of visible, Near Infrared, Shortwave Infrared and spectral indices, i.e., Enhanced Vegetation Index (EVI), Normalized Difference Vegetation Index (NDVI) and Normalized Difference Water Index (NDWI) generated from a valid pair of pixels, which free from cloud, cloud-shadow, saturated, dark area and cirrus of Landsat-8 and Sentinel-2 images, were compared. Ordinary Least Square and t-test for mean difference analysis were accomplished to perform the relationship and the spectral changes of surface reflectance and spectral indices of Landsat-8 and Sentinel-2. The results showed a significant difference in mean values of surface reflectance and spectral indices of Landsat-8 and Sentinel-2 with a higher value of R² in the dry season (July) than in the rainy season (January). SWIR performed the highest relationships among the other spectra with R² of 0.68 in the wet season and R² of 0.8 in the dry season. However, spectral indices comparison is likely performed well over the dry and wet seasons. A single band of surface reflectance provided a lower relationship (R² < 0.8) than from spectral indices (R² > 0.8). NDWI provided best relationships on water pixels while EVI and NDVI well performed on soil and vegetation pixels, respectively. We found a lower relationships of surface reflectance between Landsat-8 and Sentinel-2 in tropical environments than were reported in subtropical regions. The highest relationship of this spectral comparison was found on water pixels, then decreased on soil pixels and lastly on vegetation pixels. Higher variability of tropical atmospheric conditions and mixed pixels issue may influence this lower relationship, particularly during the rainy season. An improvement of cloud masking product in tropical environment can be investigated further to provide a better relationship of this comparison for further research.