Snowmelt forecasting as a contribution to operational flood warning – a system integrating remote sensing data and meteorological model output

Stefan Voigt¹, Hannes Kleindienst², Michael F. Baumgartner³
University of Bern, Remote Sensing Research Group, Hallerstr. 12, 3012 Bern, Switzerland.
Tel.: ¹+41-31-6318552, ²+41-31-6318552, ³+41-31-6318020, Fax: +41-31-6318511,
E-mail: ¹voigt@giub.unibe.ch, ²kliendienst@giub.unibe.ch, ³baumgartner@giub.unibe.ch

Abstract:
In alpine regions like Switzerland snowmelt is a governing factor for runoff generation. In combination with specific weather conditions such as excessive rainfall on melting snow for example, it may even be a major cause for floods. In addition, also other hazardous processes such as landslides or debris flow can be caused or triggered by rapid snowmelt. Prediction of snowmelt and its runoff can therefore contribute to specific warning systems.

In this paper we present the design of an operational system for snowmelt and snowmelt runoff forecast. It is being developed within the frame of the European project HYDALP (Hydrology of Alpine and High Latitude Basins). First runs are demonstrated for the Rhein-Felsberg river basin (Switzerland). The system is based on the output of a high-resolution weather forecast model ("Swiss Model") and NOAA-AVHRR (Advanced Very High Resolution Radiometer) satellite observations providing frequent snow cover maps. Snowmelt is predicted by applying the Snowmelt Runoff Model (SRM) developed by MARTINEC and RANGO. The model is based on the degree-day method and is driven by input of air temperature, precipitation and snow coverage of individual elevation zones.

If the cloud condition permits, near real-time information on the spatial extent of snow cover is derived from daily satellite observations using NOAA-AVHRR afternoon passes. The high repetition rate and the medium resolution of the AVHRR instrument (1.1 km at nadir) allow operational snow cover monitoring of hydrological basins ranging from a few hundred square kilometres up to continental scale. AVHRR data are received at a local ground station using the direct readout mode of NOAA satellites. Snow maps are then generated in an off-line processing chain, which takes about four to five hours. The process steps are pre-processing, calibration, geo-referencing and snow cover classification. Snow is separated from clouds using AVHRR channels 3, 4 and 5 (thermal wavelengths) whereas snow classification relies on the spectral albedo in the visible and near-infrared part of the spectrum (AVHRR channels 1 and 2). A statistical post-classification procedure using a modified Kriging approach is utilised to interpolate snow coverage for pixels, which could not be classified due to cloud cover. Finally, these snow cover maps are imported into a Geographical Information System (GIS) to compute the statistical information on snow coverage of each elevation zone.

The meteorological input for the SRM is taken from the Swiss Model forecasts. They are automatically received twice per day and range up to 48 hours into future. It is planned also to apply medium range forecast data taken from general circulation models (GCM). The Swiss Model output data are provided as a grid with a cell size of 15x15 km˛. Therefore, they need to be resampled in order to represent the elevation zones as required by the SRM. Whenever new meteorological and hydrological data are available they are automatically incorporated to update the model database and a new simulation run is initialised.