Spatial Distribution of Snow Water Equivalent Observations in Mountainous Terrain

ABSTRACT

The National Weather Service and its cooperators, principally the Natural Resources Conservation Service and California Department of Water Resources, gather SWE and related data for approximately 2500 fixed sites in the western United States and a portion of southwest Canada. Observation sites are exclusively located in mountainous regions and predominately at moderate elevation. As such, the lower and higher elevation ranges are not well represented. Satellite imagery can be used to augment SWE field observation sites, particularly at lower elevations. Analysis of classified imagery (i.e., snow, ground, and cloud) permits an estimate of the spatial distribution of the snow line (zero SWE interface). Gridded estimates of SWE can be obtained by combining the satellite-derived classification and snow line with field observations of SWE in a process that incorporates elevational detrending.

INTRODUCTION

BACKGROUND

In an attempt to quantify the dominant factors that influence the snowpack in the Western U.S., the National Weather Service developed the Snow Estimation and Updating System (SEUS). The purpose of SEUS is to estimate a gridded SWE field that can be used to update a process simulation runoff model resulting in improved streamflow forecasts (Day, 1990, McManamon et.al, 1993). While the accuracy of individual pixel estimates from SEUS may be questionable, the values are reasonable when integrated within a hydrologic basin and provide for improved streamflow simulation. The process of SEUS implementation is somewhat laborious, and to date, only about twenty-five percent of the mountainous Western U.S. can be estimated.

In the meantime, gridded estimates of snow water equivalent can be made with simpler models. It should be noted that simpler models will yield more general, less reliable results. Nonetheless, spatially-distributed estimates of SWE can be derived from point observations of SWE and remotely sensed areal extent of snow cover using techniques described below.

MODIFIED ELEVATIONAL DETRENDING

The application of elevational detrending in our simple model focused on the reasonable representation of a local SWE - elevation relationship. Initial efforts involved only point observations of snow water. The results were not acceptable across the domain of the procedure, namely, the entire Western U.S. In some areas, the observational network was adequate to define the relationship, in most it was not. The key observational network deficiencies included (1) small SWE observation sample size within a (2) limited range of elevations.

The incorporation of satellite estimates of snow cover make elevation detrending possible in the Western U.S. Custom-generated composites of satellite areal extent of snow cover permit the identification of local snow line elevations which "tie down" the low end of the SWE - elevation relationship. Satellite-derived thematic raster images of snow cover are generated locally, on-demand, at the NOHRSC (Hartman, et al 1995).

The process of "modified" elevational detrending operates as follows for each grid cell in the area to be estimated.

1. Point observations of SWE within a user-defined search radius are identified and their average elevation and average snow value calculated (Figure 1).
2. From an areal extent of snow cover raster, the elevation of the nearest snow line is identified (Figure 2). If the snow line is too distant or is non-existent, a user-defined default snow line elevation is used in its stead.
3. A linear relationship between elevation and SWE is established from the snow line and the average elevation and the average SWE observation (sloped line in Figure 3).
4. Elevationally detrended residuals are computed for each SWE point observation using the relationship in (3) (blue lines in Figure 3).
5. An inverse distance weighted mean interpolated residual is computed (red line in Figure 3).
6. An estimate of snow value is computed using the relationship in (3) and the elevation of the pixel being estimated (yellow box in Figure 3).
7. The estimate in (6) is modified by the interpolated residual in (5) yielding the final estimated SWE (red box in Figure 3. Low and high resolution gridded estimates in Figures 4 and 5.)

Figure 1: Snow telemetry (SNOTEL; lightning bolt), snow course (diamond), airborne gamma survey (line), and aerial marker observations within a specified distance of the estimated grid cell (square). The elevation of each observation and of the estimated grid cell is determined from the DEM which is displayed, in this figure, in varying shades of gray.

Figure 2: Areal extent of snow cover (blue) developed from satellite image processing. The arrow indicates the location of the snow line nearest the estimated grid cell. The snow line elevation is determined from the DEM.

Figure 3: Snow value to elevation relationship calculated from the snow line and the mean observation.

Figure 4: 300 arc second gridded snow values.

Figure 5: 30 arc second gridded snow values.

It is recognized that this simple modeling approach greatly generalizes the physical processes that result in the accumulation and melt of the mountainous snowpack. In truth, the observational network probably does not contain adequate information to accurately describe gridded snow water. The introduction of remotely sensed areal extent of snow cover data makes gridded estimation more feasible, but reason should be applied when making use of the results. Very little confidence should be placed on individual pixel values. However, when integrated within hydrologic basins, the results appear to be credible. Comparison of this approach with SEUS, an elaborate physical model of snow accumulation and melt, indicates that this simplified approach captures the spatial distribution and trends exhibited by the snowpack.

Estimation comparisons with SEUS in the Colorado Basin indicated the described procedure may overestimate at higher elevation. Work is underway at NOHRSC to "reshape" the snow water - elevation relationship above the elevation of the highest observation point.

CONCLUSION

Raster and image products developed by this procedure are available and updated at least once a week on the NOHRSC HomePage (http://www.nohrsc.nws.gov).

REFERENCES

Daly, C., and R.P. Neilson, "Digital Topographic Approach to Modeling the Distribution of Precipitation in Mountainous Terrain", In: Interdisciplinary Approaches in Hydrology and Hydrogeology, American Institute of Hydrology, pp. 437-454, 1992.

Day, G.N., "A Methodology for Updating a Conceptual Snow Model With Snow Measurements", NOAA Technical Report NWS 43, National Weather Service, Silver Spring, Maryland, 1990.

Hartman, R.K., A.A. Rost, and D.M. Anderson, "Operational Processing of Multi-Source Snow Data", Proceedings of the Western Snow Conference, pp. 147-151, 1995.

McManamon, A., T.L. Szeliga, R.K. Hartman, G.N. Day, and T.R. Carroll, "Gridded Snow Water Equivalent Estimation Using Ground-Based and Airborne Snow Data", Proceedings of the Eastern Snow Conference, Quebec, Canada, pp. 75-81, 1993.

Phillips, D.L., J. Dolph, and D. Marks, "A Comparison of Geospatial Procedures for Spatial Analysis of Precipitation in Mountainous Terrain", Agricultural and Forest Meteorology, 58, pp. 119-141, 1992.