Snow conditions have an important impact on nature, infrastructure and society at large. One third of the annual precipitation in Norway falls as snow. Information about snow is important economically and for recreational purposes. The results from SNOWHOW will benefit hydropower planning, flood forecasting, avalanche risk assessments and transport and construction safety. Monitoring and forecasting of snow conditions are carried out using a combination of computer models and observations. The most commonly used operational snow models are of the regressions type, constructed by fitting parameters and functions. Such models contrast physically based models, which aim to describe the natural processes at play. One reason for the difference in approach is the difference in spatial scale between physically based models, usually quite accurate for a point given sufficient input data - and fitted models used to describe snow conditions over large areas, for example a catchment. Regression type models are, however, 1) not suitable for predicting snow conditions for which they are not fitted, i.e. under climate change and 2) not suitable for updating, i.e. correcting the models from observations.
In SNOWHOW, the Norwegian Water Resources and Energy Directorate, the Metetorological Institute, SINTEF and partners from the hydropower industry (Glommen and Laagen brukseierforening, Hydro Energy, E-CO energy and Trønderenergi), will build snow models that have improved physical realism, a minimum of parameters to be calibrated from data and the ability to be updated from observed snow states such as snow depth, snow covered area and snow water equivalent. This will be achieved by 1) include all available snow information (e.g. snow depth, snow water equivalent and snow coverage) in developing and evaluating operational snow models, 2) reduce the dependence on calibrated parameters by striving for an optimal blend between physically based models and parameterized area-extensive models and 3) provide new projections of snow conditions under a changed climate.
During the project, it became clear that the model development and the model comparison were demanding with respect to resources. Since these tasks were crucial to the project, it was decided to focus on these at the cost of developing operational methods for remotely sensed data and providing new datasets.
Snow simulations for catchments in Southern- and Mid Norway have been used to compare the different snow models, Crocus, DDD_CX, DDD_EB and seNorge. We have run the models on 1hr temporal resolution for the period 1.9.2013-31.8.2016 with identical precipitation and temperature forcing and evaluated the simulated snow parameters; snow depth, snow water equivalent, snow coverage, snowmelt rates and energy balance elements. The results of the model inter comparison have been summarized in Skaugen et al. (2018). We have developed and published a method for estimating the spatial frequency distribution of snow water equivalent from observed spatial variability of precipitation (Skaugen and Weltzien, 2016). We have developed and published a method for updating the snow reservoir in seNorge from satellite images of snow coverage (Saloranta, 2016). For the SINTEF-based snow model GamSnow has developed a new, more physically based routine for simulating albedo (Kolberg, 2018). A first attempt on running the Crocus model regionally with observed input (interpolated meteorological grids) and input from a numerical weather prediction (NWP) model has been carried out. The results show that observed input gives better results, and suggests that Crocus in predictive mode as a tool for avalanche and flood forecasting with input from NWP needs further development. The results are presented in Luijting et al. (2018). The best model from the model inter comparison, seNorge, have been used to analyse snow conditions under a future climate (Saloranta and Andersen, 2018).
The suite of models vary with respect to physical/empirical process description and need for calibration. Analysis of model results show that models with physically based melt algorithms (energybalance) give consistent and good results. Empirically based melt algorithms, calibrated against snow observations and not against runoff observations also provide good results. However, empirical models calibrated against runoff provide varied results and cannot be recommended. These results indicate the primary objective of SNOWHOW; to develop snow models that can be used for climate change projections and for ungauged catchments, is, by large, achieved.
We have organised two internal two-day workshops in order to further concretise the plans and bring our partners in the hydropower industry up to date. SNOWHOW has organised, together with the project ESCYMO, an international workshop at Finse, Norway, 10-12 October, with 33 participants from 10 countries.
Operational models of snow processes have evolved differently in hydrology and meteorology. Models for meteorology and avalanche warning have detailed physically based process representations, while models for hydrology have converged towards simplified and heavily calibrated snow parameterisations. The latter has shown to work well under stationary weather regimes, but will fail under conditions, for which they are not calibrated, i.e under a changed climate. Many experiments have failed to prove that physically based snow models are superior to simpler parameterisations. A challenge is that detailed models tend to be sensitive to poorly known quantities. Further, assumptions which hold for a well-controlled experimental site, are often weakly justified for large, heterogeneous areas. Finally, hydrologic validation relies on statistics from runoff series, which are low in information content compared to the complex processes upstream. This project aims at improve the physical realism in snow models, making them reliable for operational simulation of multiple hydrology- and snow characteristics, (runoff, snow water equivalent, snow covered area, snow surface temperature and snow albedo) at multiple spatio-temporal scales. This will be achieved by 1) including all available sources of snow information in developing, evaluating and updating operational snow models, 2) reducing the dependence on calibration parameters by striving for an optimal blending between physically based point models and parameterized area-extensive models and 3) providing a guide to future observation networks better suited to a new generation snow models. The results of the project will lead to improved flood forecasting, weather forecasting, input to hydropower scheduling, and avalanche risk assessment and will further increase knowledge of seasonal snow dynamics among scientists and water managers.