Seasonal Evolution of Supraglacial Lakes on Baltoro Glacier From 2016 to 2020

Abstract

The existence of supraglacial lakes influences debris-covered glaciers in two ways. The absorption of solar radiation in the water leads to a higher ice ablation, and water draining through the glacier to its bed leads to a higher velocity. Rising air temperatures and changes in precipitation patterns provoke an increase in the supraglacial lakes in number and total area. However, the seasonal evolution of supraglacial lakes and thus their potential for influencing mass balance and ice dynamics have not yet been sufficiently analyzed. We present a summertime series of supraglacial lake evolution on Baltoro Glacier in the Karakoram from 2016 to 2020. The dense time series is enabled by a multi-sensor and multi-temporal approach based on optical (Sentinel-2 and PlanetScope) and Synthetic Aperture Radar (SAR; Sentinel-1 and TerraSAR-X) remote sensing data. The mapping of the seasonal lake evolution uses a semi-automatic approach, which includes a random forest classifier applied separately to each sensor. A combination of linear regression and the Hausdorff distance is used to harmonize between SAR- and optical-derived lake areas, producing consistent and internally robust time series dynamics. Seasonal variations in the lake area are linked with the Standardized Precipitation Index (SPI) and Standardized Temperature Index (STI) based on air temperature and precipitation data derived from the climate reanalysis dataset ERA5-Land. The largest aggregated lake area was found in 2018 with 5.783 km2, followed by 2019 with 4.703 km2, and 2020 with 4.606 km2. The years 2016 and 2017 showed the smallest areas with 3.606 and 3.653 km2, respectively. Our data suggest that warmer spring seasons (April–May) with higher precipitation rates lead to increased formation of supraglacial lakes. The time series decomposition shows a linear increase in the lake area of 11.12 ± 9.57% per year. Although the five-year observation period is too short to derive a significant trend, the tendency for a possible increase in the supraglacial lake area is in line with the pronounced positive anomalies of the SPI and STI during the observation period.

Spatial and temporal dynamics of aufeis in the Tso Moriri basin, eastern Ladakh, India

“…Therefore, the delineation of icing and aufeis was performed using a random forest (RF) classifier (R randomForest package 59 ). RF is a non-parametric supervised machine learning algorithm and has been frequently applied in various remote sensing studies, e.g., mapping of global lake ice cover, 60 supraglacial lakes 61,62 or wetland inventorying. 63 An RF draws predictions from an ensemble of uncorrelated Classification and Regression Trees (CART) which are built on the basis of randomly selected subsets of training samples.…”

Basal Sliding and Hydrological Drainage at Baltoro Glacier

“…In our approach, a coregistration of the PlanetScope data is not needed since the assignment of the classified lakes over the season is performed using their center coordinate. Sentinel-1 Interferometric Wide Swath Single Look Complex (SLC) C-band and TerraSAR-X ScanSAR Multi-Look Ground Range Detected (MGD) X-band data were processed to Analysis Ready Data (ARD) using the Multi-SAR System (Schmitt et al, 2015(Schmitt et al, , 2020. The mapping of the seasonal lake evolution used a semiautomatic approach which is based on a random forest classifier applied separately to each sensor.…”

“…Detailed processing steps and results for the years 2016 to 2020 were published by Wendleder et al (2021b). For this study, the time series was extended by 2021 and 2022.…”

“…In radar, the Kennaugh Elements describes the polarimetric information and enables the interpretation of physical scattering mechanisms. In the case of optical data, the Kennaugh Elements are a fusion technique to combine multiple multi-spectral bands to one image while enhancing their spectral characteristics (Schmitt et al, 2020). Afterwards, a K-means clustering grouped each image into the two classes "runoff" and "background".…”

“…We mapped the wet snow on the complete Baltoro Glacier which includes the debris-covered and debris-free part of the glacier from the Sentinel-1 Interferometric Wide Swath C-band data. The processing of the SLC to ARD data was performed with the Multi-SAR System (Schmitt et al, 2020). The additional use of image enhancement (specifically, multi-scale multi-looking) resulted in a smoothing of noise and hence a more homogeneous environment and a better wet snow classification (Schmitt et al, 2015;Wendleder et al, 2022).…”