Subgridding High Resolution Numerical Weather Forecast in the Canadian Selkirk range for local snow modelling in a remote sensing perspective

Abstract

Abstract. Snow Water Equivalent (SWE) is a key variable in climate and hydrology studies. Current SWE products mask out high topography areas due to the coarse resolution of the satellite sensors used. The snow remote sensing community is hence pushing towards active microwaves approaches for global SWE monitoring. However, designing a SWE retrieval algorithm is not trivial, as multiple combinations of snow microstructure representations and SWE can yield the same radar signal. The community is converging towards forward modeling approaches using an educated first guess on the snowpack structure. Yet, snow highly varies in space and time, especially in mountain environments where the complex topography affects atmospheric and snowpack state variables in numerous ways. Automatic Weather Stations (AWS) are too sparse, and high-resolution Numerical Weather Predictions systems have a maximal resolution of 2.5 km × 2.5 km, which is too coarse to capture snow spatial variability in a complex topography. In this study, we designed a subgridding framework for the Canadian High Resolution Deterministic Prediction System. The native 2.5 km × 2.5 km resolution forecast was subgridded to a 100 m × 100 m resolution and used as the input for snow modeling over two winters in Glacier National Park, British Columbia, Canada. Air temperature, relative humidity, precipitation and wind speed were first parameterized regarding elevation using six Automatic Weather Stations. Alpine3D was then used to spatialize atmospheric parameters and radiation input accounting for terrain reflections and perform the snow simulations. Modeled snowpack state variables relevant for microwave remote sensing were evaluated against profiles generated with Automatic Weather Stations data and compared to raw HRDPS driven profiles. Overall, the subgridding framework improves the optical grain size (OGS) bias by 0.04 mm, the density bias by 2.7 kg · m−3 and the modelled SWE by 17 % (up to 41 % in the best case scenario). Overall, this work provides the necessary basis for SWE retrieval algorithms using forward modeling in a Bayesian framework.

Evaluation of high resolution snowpack simulations from global datasets and comparison with Sentinel-1 snow depth retrievals in the Sierra Nevada, USA

“…To address the mismatch in spatial resolution between reanalyses datasets and snow distribution, previous studies used downscaling algorithms based on a digital elevation model before running a snowpack model on a finer grid (Armstrong et al, 2018;Baba et al, 2018;Billecocq et al, 2023;Mernild et al, 2017;Weber et al, 2021). This approach enables estimation of high resolution SWE and snow depth without ground data.…”