Can we propagate informations from sparse snowpack observations into space using a Particle Filter?

Abstract:

The seasonal snowpack of mid-latitude mountainous areas has a strong spatio-temporal variability. Snowpack models accuracy is inherently limited by the errors of meteorological models that they use as input, uncertainties in the modeling of snow processes and representativeness issues due to the high local variability of the snowpack. Data assimilation of remotely-sensed snowpack observations such as surface reflectances, and airborne lidar snow depths is a promising avenue, but these data generally have a partial spatial coverage and are affected by errors. In-situ observations of snow depth or snow water equivalent are much more accurate, but usually lack spatial coverage and representativeness.

In this work, we present an ensemble data assimilation using the Particle Filter to ingest snowpack observations. We developed innovative versions of the PF enabling to propagate information from observed areas to non observed areas. We demonstrate the theoretical efficiency of this method by assimilating synthetic observations of snow depth and reflectance. Then, we investigate the performance of this system in a network of about 300 snow depth stations in the French Alps and Pyrenees. Results show that the ensemble reliability is improved by up to 30%. We also show that the best results are obtained in the denser areas.