Modulating spectral reflectance and its impact on thematic mapping: an analysis on glacier facies using very high-resolution satellite imagery

Abstract

Analysis of the spectral properties of glacier surface facies representing glaciological zones is an established method for characterizing facies. This necessitates a reliable estimation of spectral reflectance from image pixels using appropriate atmospheric correction methods. Improving spatial resolution by pansharpening may improve visual detection of facies, but the translation of visual perception to algorithmic classification is unclear. In this study, we analyzed the changes in spectral reflectance by comparing one model-based and two image-based methods of atmospheric corrections using very high-resolution (VHR) WorldView-2/3 satellite data for selected glaciers in Svalbard and the Himalayas. Two pansharpening methods were then applied followed by 12 machine learning and three object-based image analysis (OBIA) rule sets. A combination of wavelength ranges representing traditional and recent spectrums were tested using four different spectral band combinations. In total, we compared 8100 surface facies maps to understand the impact of every image processing routine and mapping method on the final thematic categorization of facies. Broad conclusions suggest that VHR data does not require shortwave infrared ranges; OBIA is more accurate and less sensitive to image processing variations; machine learning is more efficient but readily influenced by image processing. A spatial-spectral information OBIA ruleset yielded the highest overall accuracy (0.86). Segmentation parameters were found to be consistent across study regions and may be transferable outside the current study. This study demonstrates the need for testing image processing induced variations for cryospheric applications and may be assessed on open satellite data for long-term standardized mapping of glacier surface facies.