Glacier facies characterization using optical satellite data: Impacts of radiometric resolution, seasonality, and surface morphology

Abstract

The spectral characterization of geographic landscapes is vital for their accurate mapping using remote sensing data. This can be done through spectral profiling, as demonstrated here, to characterize the surface facies of the Gangotri and neighbouring glaciers, central Himalaya. The satellite-derived reflectance curves were compared with the in-situ and published (validation) data. The study attempts to understand the influence of certain parameters such as the satellite sensor’s radiometric resolution, timing of data acquisition (seasonality), and surface morphology on glacier/snow–ice facies identification. Results show that the first two parameters complement each other in identifying the snow–ice facies accurately. High radiometric resolution (HRR) data concurred closely with the validation dataset and had higher mean entropy values over the glaciated areas than low radiometric resolution (LRR) ablation data. Presence of seasonal snow and degree of surface melting show considerable influence on satellite-derived reflectances of glacier facies. Our findings assert the usage of HRR ablation data in appraising the interannual and seasonal variability of glacier facies. While HRR post-ablation data overestimates the reflectance of snow–ice facies, LRR post-ablation data have limitations in their discrimination. Certain morphology and resultant features, such as crevasses and shadows, induce underestimation of the satellite-derived reflectances, creating confusion among the snow and ice facies. This spectral confusion can, however, be resolved by the use of ancillary data. Elevation, temperature, and band ratios/spectral indices are helpful in segregating snow–ice facies, while slope, band ratios, temperature, and texture measures effectively discriminate the other facies.

On Drivers of Subpixel Classification Accuracy—An Example From Glacier Facies

“…Spectral overlap among the glacier facies is another factor that lowers the SPC accuracy. In such cases, ancillary data have been widely recognized as enhancers of classification accuracy [18], [22], [33]. Results from experiment III confirm that inputting the ancillary layer(s) in SPC offers significant accuracy improvements.…”

“…class, since our aim was mainly to delineate the snowice/glacier cover facies. 4) The aforementioned ancillary layers (see Section II) were derived from multiple sources for both the AWiFS and MSI data (see Table A.1, Supplementary material) [18]. The applicability of these layers in glacier facies mapping has been explained in Section A.3 (Supplementary material).…”

“…A brief review of prevalent SPC techniques, including support vector machine (SVM) details are given in the Supplementary material (see Section A.1). Glacier facies result from snow accumulation and surface ablation patterns occurring on the glacier surface [18]. Changes in their distribution reflect climatological and hydrological changes [19].…”

“…Whereas the applicability of HRR data like Resourcesat-1/2 AWiFS (10/12-bit, respectively) have been restricted to snow cover mapping only [25], [26]. Recently, the potential of HRR data in the characterization of various snow-ice and other facies has been demonstrated [18]. Excluding the recent MSI and OLI sensors, available HRR data confounds the PPC of glacier facies due to the dominance of mixed pixels (mixels) [25].…”

“…Whereas, certain glacier facies lack spectral variability among them due to their compositional similarity (supraglacial/periglacial debris and valley-rock) or glacier's surface morphology (crevassed/ shadowed snow facies and ice facies). This can induce misclassification errors, which can although be reduced by using spectral plus ancillary data in the classification process [18], [22]. Acquisition time of the RD may highly influence the classification accuracy of the glacier facies due to their dynamicity in the input and RD.…”