New computationally efficient 100 m statistically downscaled Greenland products using the MAR regional climate model and various remote sensing observations.
Abstract
The Modèle Atmosphérique Règional (MAR) is a regional climate model able to simulate the physical processes and feedbacks modulating and controlling the surface mass balance of the Greenland ice sheet (GrIS). Improvements in numerical schemes have reduced the computational effort of running regional climate models over large areas for given spatial and temporal resolution. However, the horizontal spatial resolution is still limiting a factor in terms of running time. Statistical downscaling techniques are a suitable solution to obtain higher spatial resolution outputs from the original coarse model predictions without paying too much effort on the running time side. Here we present the updates of our work aimed to develop a statistically downscaling algorithm, building on the approach originally reported by Noel et al., 2016. We statistically downscale runoff, surface mass balance, surface temperature and liquid water content generated by the regional climate model MAR 3.11.2 (6 km) over the GrIS at a spatial resolution of 100 m. The approach is based on the local dependency to elevation of the parameter to be downscaled (e.g., temperature) expressed as a linear regression. The regression parameters are then applied to the high-resolution grid, obtained from the GrIS digital elevation model from the Polar Geospatial Group. Our algorithm is parallelized across grids to be run in a multi-core environment using OpenMP and gdal spatial libraries. Two different versions have been developed: The first is optimized for limited spatial locations but for very long time-series (e.g. daily for 20 years), and one for thousands of spatial locations for only one or a few days. A typical example of this latter version is used to downscale all pixels associated with a LANDSAT scene. We assessed the statistically downscaled outputs through in-situ observations and remote sensing data. Surface mass balance measurements were taken from the PROMICE dataset while air temperature data from the Greenland Climate Network (GC-Net). Surface temperature images and spatial distribution of surface meltwater are obtained from LANDSAT over two selected region of interest, the Jakobshavn and Helheim glaciers. Our results indicate that the statistically downscaled outputs generally perform better than the original, coarse resolution outputs. Moreover, a semi-variogram based analysis shows a stronger agreement of the downscaled product with satellite data in terms of spatial autocorrelation and scale break. Possible applications of the new high-resolution dataset include, for example, the combination with the recently launched ICESAT2 (e.g., removing compaction signal from elevation changes), the assessment of the MAR outputs with high-resolution remote sensing datasets (e.g., Sentinel) and the possibility to better spatially constrain the transport of meltwater to the ocean at specific locations.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2021
- Bibcode:
- 2021AGUFM.C45B1004C