The glacier volume and area are two fundamental data sources on simulating the impacts of glacier change on hrdrologic processes at drainage basin scale. However, the lack of historical satellite imageries significantly limited the compilation of early stage glacier inventories. The rapid changes of glaciers in the source region of Yangtze and Yellow Rivers have large influences on the basin scale hydrological processes, but the shortage of multi-temporaral glacier inventories has largely impeded the studies on the hydrological impacts of glacier change in both river basins. The glacier thickness and area dataset during 1960-2020 in the source region of Yangtze and Yellow Rivers were aimed to fill this data gap. They were produced by the latest glacier model named IGM, which is a deep learnig based glacier model drived by glacier dynamic model and mass balance model, can simulate the century scale glacier evolution and give acceptable illustration on glacier change processe under climate change. The dataset includes the glacier depth with 5-year interval between 1960-2020 of glaciers in the source regions of Yangtze and Yellow Rivers, and the glacier area deduced from the glacier depth, which can fill up the glacier data gaps in both river basins to a certain extent.
IGM (Instructed Glacier Model) was developed by Guillaume Jouvet et al. of University of Zuich, Switzerland, and was formally published in 2021, aimed to simulate 3D glacier evolution. The IGM model was developed using Python with module-wise organization, takes large benefit of existing tools such as OGGM, and uses a horizontal regular grid for numerical discretization. IGM implements mass conservation, high-order 3D ice flow mechanics, an Enthalpy model for the thermic regime of ice, melt/accumulation surface mass balance model, and other glaciological processes. It based on TensorFlow library which facilitate GPU to realize fully vectorized simulation on the ice flow physics, therefore has very high calculation effeciency. The IGM can be used to simulate large domain / high resolution glacier evolution, but can also simulate individual glaciers. When take glacier extent as the input, IGM will automatically collect the DEM and reanalyzed climate data, and simulate the glacier mass balance and dynamic process in specified time periods. The default of IGM output is glacier depth with spatial resolution of 100-1000m and temporaral resolution of month to year, and can describe the glacier evolution with time series of glacier depths.
(1) Using the 1970 glacier boundary in the source regions of the Yangtze and Yellow Rivers as the initial condition and remote sensing-derived glacier boundaries around 1990, 2000, 2010, and 2020 as control conditions for the IGM model, simulate the glacier mass balance and dynamic processes from 1960 to 2020, and output glacier thickness at five-year intervals with a resolution of 100 meters; (2) Employ bilinear interpolation to resample the original 100-meter resolution glacier thickness into 25-meter resolution; (3) Extract the valid (non-zero) simulated area from the resampled glacier thickness, then clip to form the glacier thickness simulation results for each individual glacier; (4) Using a 2-meter thickness threshold to identify glacier from non-glacier areas from the clipped glacier thickness data to generate coded glacier masks, then perform mosaicking and raster to vector conversion to delineate glacier boundaries.
Since the simulated glacier boundaries are generated from glacier thickness simulation results, and the reanalysis climate data driving the model has coarse resolution and high uncertainty, coupled with inherent uncertainties in the DEM used and the model itself, both the simulated glacier thickness and boundaries possess a certain degree of ambiguity and uncertainty, which both have limited comparability with the actual glacier thickness and boundaries. However, during the simulation process, in addition to using the 1970s glacier boundary as the initial constraint, glacier boundaries around 1990, 2000, 2010, and 2020 were also incorporated as control conditions. This approach maximizes the consistency between the simulated boundaries and the real boundaries extracted from remote sensing. Comparative analysis shows that the mean difference between the simulated and actual glacier areas for 1970, 1990, 2000, 2010, and 2020 is approximately 2.8%.
This work is licensed under a
Creative
Commons Attribution 4.0 International License.
CSTR:11738.11.NCDC.GLACIER.DB6957.2025
10.12072/ncdc.glacier.db6957.2025