Warm and dry fohn winds on the Antarctic Peninsula (AP) cause surface melt that can destabilize vulnerable ice shelves. Topographic funneling of these downslope winds through mountain passes and canyons can produce localized wind-induced melt that is difficult to quantify without direct measurements. Our Fohn Detection Algorithm (FonDA) identifies the surface fohn signature that causes melt using data from twelve Automatic Weather Stations on the AP, used to train a machine learning model to detect fohn in 5km Regional Atmospheric Climate Model 2 (RACMO2.3p2) simulations and in the ERA5 reanalysis model. We estimate the fraction of AP surface melt attributed to fohn and possibly katabatic winds and identify the drivers of melt, temporal variability, and long-term trends and evolution from 1979-2018. We find fohn wind-induced melt accounts for 3.1% of the total melt on the AP but can be as high at 18% close to the mountains where the winds are funneled through mountain canyons. Fohn-induced surface melt does not significantly increase from 1979-2018, despite a warmer atmosphere and more positive Southern Annular Mode. However, a significant increase (+0.1Gt y-1) and subsequent decrease/stabilization occurred in 1979-1998 and 1999-2018, consistent with the AP warming and cooling trends during the same time periods. Fohn occurrence more than fohn strength drives the annual variability in fohn-induced melt. Long-term fohn-induced melt trends and evolution are attributable to seasonal changes in fohn occurrence, with increased occurrence in summer, and decreased occurrence in fall, winter, and early spring over the past 20 years.

To better resolve the bed elevation of Antarctica, we present DeepBedMap - a deep learning method that produces realistic Antarctic bed topography from multiple remote sensing data inputs. Our super-resolution deep convolutional neural network model is trained on scattered regions in Antarctica where high resolution (250 m) groundtruth bed elevation grids are available, and then used to generate high resolution bed topography in less well surveyed areas. DeepBedMap takes in a low resolution (1000 m) BEDMAP2 dataset alongside other high spatial resolution inputs such as ice surface elevation, velocity and snow accumulation to generate a four times upsampled (250 m) bed topography map even in the absence of ice-thickness data from direct seismic or ice-penetrating radar surveys. Our DeepBedMap model is based on an Enhanced Super Resolution Generative Adversarial Network architecture that is adapted to minimize per-pixel elevation errors while producing realistic topography. We show that DeepBedMap offers a more realistic topographic roughness profile compared to a standard bicubic interpolated BEDMAP2, and also run model inversions to compare the basal traction of our DeepBedMap_DEM with other bed elevation models.