Modelling Daily Runoff from Snow and Glacier Melt Using Remote Sensing Data

In this paper we present runoff simulations for snow and icemelt for the basins of Rhine-Felsberg, RhôneSion and Ticino-Bellinzona. We use high resolution multispectral remote sensing data to map snow and ice cover of selected years. Further, we use GIS based extrapolation techniques to evaluate cloud and forest covered areas with respect to snow cover. We use the deterministic hydrologic model SRM+G to simulate the daily runoff of selected years. Conditions for a norm year in terms of normalized daily values for snow cover depletion, temperature and precipitation for the time period 1961 to 1990 have been established. Based on norm year conditions we did climate change simulations for the scenarios 2030 and 2100, which are characterized by increasing temperatures during winter and summer and increased precipitation during winter. The results show different characteristic behaviour for the three basins. The highly glaciated basin Rhône-Sion shows an influence of enforced summer icemelt, the low glaciated Rhine-Felsberg and Ticino-Bellinzona an earlier snowmelt. INTRODUCTION Multispectral optical remote sensing data are particularly qualified for monitoring the extension of the alpine snow and ice cover (1,2,3). The optical satellite sensors are not capable to penetrate clouds, so images with more 20% cloudiness can not be used for interpretation. However, these problems with partly cloud covered images can widely be mastered since Ehrler showed a method to extrapolate cloud covered areas (4). Combining the remote sensing derived snow and ice cover maps with a hydrologic runoff model the daily runoff can be calculated (4,5). Snow and icemelt are important contributors to the total yearly runoff volume in high alpine basins. The seasonal snow melts progressively during spring and summer. The snow cover in lower regions of the Alps disappears in March – April, in higher regions in May – August. Sometimes summer new snow delays the ongoing snowmelt. The glacier icemelt results after the glacier areas become snowfree, starting in summer and ending with the autumn snowfall events in September – October. Alpine glaciers react very sensible on temperature changes because their temperatures lie close to melting point and the refreezing point of water (6). From this point of view it is important to investigate individually the runoff resulting from snow and ice in a warmer climate. Contributing to the research related to snowmelt runoff modelling under the influence of climate change (7,8), we include the glaciermelt in climate change simulations. The study is carried out for the three Swiss test basins Rhine-Felsberg (3241 km, 575 – 3614 m a.s.l.), Rhone-Sion (3371 km, 488 – 4634 m a.s.l.) and Ticino-Bellinzona (1515 km, 192 – 3402 m a.s.l.) located as shown in Figure 1. The basins are characterized with different topographic, physiographic and climatic conditions. Rhône-Sion is 17% glaciated, Rhine-Felsberg contains just 1.9% and TicinoBellinzona has almost no glacier areas (0.5%). Figure 1: Location of the three alpine basins Rhine-Felsberg, Rhône-Sion and Ticino-Bellinzona in Switzerland. METHODS 1. Snow and Ice Cover Mapping Using Remote Sensing For each alpine basin as mentioned above the snow and ice covered areas have been monitored during the melting season from March – October for various years. Primarily cloud free data sets with 4 – 12 images per season has been choosen from various high resolution satellite sensors like Landsat-MSS, Landsat TM and SPOT-XS. The imagery was geometrically and radiometrically corrected. Due to the enormous topographic range of the test basins, we did the geometric orthorectification with a digital elevation model (DEM 25m resolution). The max. error amounts to 2 pixels with Landsat and to 4 pixels with SPOT data (in the case of a 19o side-looking sensor). Figure 2 shows an example of an orthorectified Landsat-TM image of the basin Ticino-Bellinzona. The radiometric correction eliminates atmospheric and topographic induced radiation errors and is based on additional scene information and the DEM. We performed this correction with programs of Sandmeier (9). Further, we masked out the cloud and forest covered areas. The forested areas are problematic for snow cover interpretat ion, most often the snow is underestimated. We used supervised (maximum likelihood) and unsupervised (K-means clustering) multispectral classification methods to label the pixel, and merged the results to “snow”, “ice”, and “snowfree non-ice”. A first visual inspection pointed to wrong declarations of ice and rocks and, especially in TicinoBellinzona, with snow pixel within sparse forest. In a second approach the results have been compared with snow-measurements of SLF (10). Figure 3 shows a result of the classification process in basin Ticino-Bellinzona. After classification of the images we extrapolated the cloud and forest covered parts in a geographic information system (GIS) taking advantage of a method developed by Ehrler (4). All images are segmented into snow cover units (SCU), being defined as regions of equal snow coverage. Subsequently the forested and cloud covered units are assigned to carry the same snow coverage as the corresponding forest-free and cloud-free units. Final result is a set of complete classified images for each basin. Figure 2: Detail of a Landsat-TM image, channels 3-2-1, from 25-May-1994 showing basin Ticino-Bellinzona (red line). EURIMAGE 1994. Figure 3: Result from multispectral classification. Green the snow-free areas, white the snow covered areas, gray appear the previously masked clouds. 2. Runoff Model for Snow and Glaciermelt The results from remote sensing and GIS were then linked to a hydrologic runoff model. We used the SRM+G model to simulate daily runoff. This model is a further development of the snowmelt runoff model (SRM) (5) and can calculate the glaciermelt as well. SRM+G is a deterministic model with spatial distribution in elevation zones and subbasins. The model works with the remote sensing derived snow and ice cover, daily temperature and precipitation measurements and a set of 11 physically derived parameters. The melt computation is based on the degree-day approach, and uses the size and distribution of glaciers within the basin as basic input. To illustrate the procedure, the modified model formula from Martinec (5) can be arranged as follows (11): where Q is the average daily discharge [m s] with index referring to rain, newsnow, snowmelt and icemelt. Newsnow indicates snow falling in the summer on snowfree area whereas snowmelt is the melt of the seasonal snow cover fallen before 1. April. n is an index indicating the sequence of days, c the runoff coefficient expressing the losses through evaporation, interception and sublimation as a ratio (runoff/precipitation) with index referring to rain, snow and glacier. as is the degree-day factor for snow [cm oC d], ag the degree-day factor for ice [cm oC -1 d]. While as varies during the melt season accordingly to changing snow density, ag stays constant over the season. A is the area of a zone [km ] with index referring to glacier, noglacier and total area. The altitude range of the zones is approx. 500 m. T is the air temperature at the mean hypsometric elevation of a zone [oC] and is extrapolated from a network of climate stations with an average temperature gradient of -0.65 oC per 100 m. Pr is the precipitation as rain according to the critical temperature at the mean hypsometric elevation of a zone or new snow falling in the summer on snowfree area [cm d]. According to the complex precipitation situation of the central alpine valleys it was a challenge to adapt the precipitation gradient to the various climate conditions of the three basins. The precipitation gradient changes from 0 upto 4.5% per 100 m. S is a ratio of the snow covered area to the total area. The recession coefficient k indicates the decline of discharge in a period without snowmelt or rainfall and depends on the storing capacity of the basin. As can be seen from the Eq., k is a function from the constants x and y, which are determined for a given basin from previous years. There are further parameters not l isted here, which are described elsewhere (11,12). We tested the model in several basins and found high accuracy even in basins with 67% glacier