Goals

The project aims first to improve our understanding of the radar interactions with the evolution of a wet snowpack and then to assess the value of this information to improve snowmelt modelling at the catchment scale. In fact, the melting phase is usually not very accurately simulated in distributed snow models, leading to errors in the exact timing of discharge, and any additional measurement that could contribute to reducing the inaccuracies is welcome. SAR remote sensing is a good candidate as it is distributed, available reasonably fast and with a sufficient spatial and temporal resolution.

Since the behaviour of the SAR signal during the melt phase is not fully explored yet, we hypothesize that by comparing snow microstructure properties with the SAR signal during this phase, it could be possible to extract the information needed to interpret this signal and significantly improve the performance of distributed snow cover models. Performance improvements would depend on data availability and on the scale of the application. Our hypothesis is that reasonably large snowmelt-dominated catchments - where the hydrological discharge is of importance for practical applications - are computationally intensive to simulate and affected by data scarcity, meaning that the additional spatial information brought by remote sensing could bring a significant improvement with respect to traditional distributed snow modelling.

To do so, this research project is focused on three different scales:

1. A radar footprint scale that is homogenous and very well instrumented, allowing very detailed comparisons between the SAR signal and the snow properties;
2. An Alpine headwater catchment scale that is well instrumented and suitable for snow hydrology research and snow cover studies;
3. A large, snowmelt dominated catchment that is relevant for operational applications.

Besides experimental measurements, numerical models will play a central role. Thanks to the many detailed measurements available at the radar scale site and the generation of not easily accessible variables through snow cover models, it will be possible to support the understanding of the SAR signal during the melt phase by linking it to the changes in snow properties. This process of understanding will be supported by the results obtained by RTs models, which are running with the real snow observations.

Then, a data assimilation scheme will be implemented into the existing snow models that are only forced by standard AWS data. Meteorological data forcing is often the dominating source of uncertainties in snow-cover model outputs (Largeron, 2020). Moreover, physically based and multi-layer snow models still contain parametrizations for physical processes that are too complex and lack data to be fully characterized. These parametrizations are a source of error in the models and can produce large spread in results between models (Essery, 2013; Rutter, 2009). Remote sensing data can provide a key constraint to reduce such uncertainties. These inputs into either the energy or the mass balance will be the major focus of our data assimilation efforts with the added benefit that focusing on the forcings will also guarantee physical consistency.

Direct insertion has been shown to be quite efficient for physically based snow cover models (Magnusson, 2017) and is computationally very light, but doesn't allow for fully exploiting the information in the measurements and for best constraining the uncertainty. Therefore, we will use direct insertion for the operational catchment scale, but we will investigate at the smaller scales more advanced data assimilation methods relying on already implemented algorithms available in the MeteoIO meteorological data pre-processor.

The major novelty of this approach is to use different kinds of information derived from the SAR signal for data assimilation. We plan to try with different parameters we can derive from radar backscattering: from "simple" indications of presence/absence of LWC at a given time, or at which of the three melting phases (moistening, ripening or runoff) the snowpack is, up to more advanced information. It is envisioned that the SNOWPACK model will provide snow microstructure data to complement the already plentiful measurements while receiving SAR data assimilation capability, while the GEOtop model could be used to further model river discharge from snowmelt water flux as input.