Search

Geoscientific models are facing increasing challenges to exploit growing datasets coming from remote sensing. Universal differential equations (UDEs), aided by differentiable programming, provide a new scientific modelling paradigm enabling both complex functional inversions to potentially discover new physical laws and data assimilation from heterogeneous and sparse observations. We demonstrate an application of UDEs as a proof of concept to learn the creep component of ice flow, i.e. a nonlinear diffusivity differential equation, of a glacier evolution model. By combining a mechanistic model based on a two-dimensional shallow-ice approximation partial differential equation with an embedded neural network, i.e. a UDE, we can learn parts of an equation as nonlinear functions that then can be translated into mathematical expressions. We implemented this modelling framework as ODINN.jl, a package in the Julia programming language, providing high performance, source-to-source automatic differentiation (AD) and seamless integration with tools and global datasets from the Open Global Glacier Model in Python. We demonstrate this concept for 17 different glaciers around the world, for which we successfully recover a prescribed artificial law describing ice creep variability by solving ∼ 500 000 ordinary differential equations in parallel. Furthermore, we investigate which are the best tools in the scientific machine learning ecosystem in Julia to differentiate and optimize large nonlinear diffusivity UDEs. This study represents a proof of concept for a new modelling framework aiming at discovering empirical laws for large-scale glacier processes, such as the variability in ice creep and basal sliding for ice flow, and new hybrid surface mass balance models. [ABSTRACT FROM AUTHOR]

Within the last decade, the Greenland ice sheet (GrIS) and its surroundings have experienced record high surface temperatures (Mote, 2007; Box et al., 2010), ice sheet melt extent (Fettweis et al., 2011) and record-low summer sea-ice extent (Nghiem et al., 2007). Using three independent data sets, we derive, for the first time, consistent ice-mass trends and temporal variations within seven major drainage basins from gravity fields from the Gravity Recovery and Climate Experiment (GRACE; Tapley et al., 2004), surface-ice velocities from Inteferometric Synthetic Aperture Radar (InSAR; Rignot and Kanagaratnam, 2006) together with output of the regional atmospheric climate modelling (RACMO2/GR; Ettema et al., 2009), and surface-elevation changes from the Ice, cloud and land elevation satellite (ICESat; Sørensen et al., 2011). We show that changing ice discharge (D), surface melting and subsequent run-off (M/R) and precipitation (P) all contribute, in a complex and regionally variable interplay, to the increasingly negative mass balance of the GrIS observed within the last decade. Interannual variability in P along the northwest and west coasts of the GrIS largely explains the apparent regional mass loss increase during 2002-2010, and obscures increasing M/R and D since the 1990s. In winter 2002/2003 and 2008/2009, accumulation anomalies in the east and southeast temporarily outweighed the losses by M/R and D that prevailed during 2003-2008, and after summer 2010. Overall, for all basins of the GrIS, the decadal variability of anomalies in P, M/R and D between 1958 and 2010 (w.r.t. 1961-1990) was significantly exceeded by the regional trends observed during the GRACE period (2002-2011). © 2012 Elsevier B.V.

Mass budget calculations, validated with satellite gravity observations [from the Gravity Recovery and Climate Experiment (GRACE) satellites], enable us to quantify the individual components of recent Greenland mass loss. The total 2000-2008 mass loss of approximately 1500 gigatons, equivalent to 0.46 millimeters per year of global sea level rise, is equally split between surface processes (runoff and precipitation) and ice dynamics. Without the moderating effects of increased snowfall and refreezing, post-1996 Greenland ice sheet mass losses would have been 100% higher. Since 2006, high summer melt rates have increased Greenland ice sheet mass loss to 273 gigatons per year (0.75 millimeters per year of equivalent sea level rise). The seasonal cycle in surface mass balance fully accounts for detrended GRACE mass variations, confirming insignificant subannual variation in ice sheet discharge.

In recent years, the existence of perennial firn aquifers in the Antarctic Peninsula (AP) has been confirmed by in situ observations. Previous studies have suggested that these subsurface aquifers, together with meltwater ponds at the surface, provide a reservoir of liquid water to feed propagating fractures, promoting hydrofracture-driven ice-shelf disintegration. This study maps perennial firn aquifers in the AP from space using C-band Synthetic Aperture Radar imagery from ESA's Sentinel-1 (S1) mission. With these observations, we detect firn aquifers at 1 km × 1 km resolution, for the period 2017 to 2020. Existing methods, that use S1 data and rely on a backscatter intensity difference threshold approach, are prone to misclassify late-melt events as aquifers, when applied to the AP. Therefore, we have developed and evaluated a new approach that is better suited to the Antarctic environment. The new method exploits the characteristic, gradual backscatter increase during the (partial) refreezing of the liquid water in the firn layer after the peak of the melt season. Most firn aquifers are detected in the north and northwest of the AP, as well as on the Wilkins and George VI ice shelves. Aquifer locations detected with the present methodology agree with in situ observations and with model simulations of firn water content. [ABSTRACT FROM AUTHOR]