WMOD00275 - Laserfiche WebLink

Home Browse Search

?Jc,,. -. AUGUST 2006 DRESSLER ET AL. 711 the lowest relative differences (1 %-5%). This may sim- ply reflect greater local-scale variability in drier years. Though SNOTEL sites are susceptible to various sensor inaccuracies (Peck 1972; Smith and Boyne 1981; Johnson and Schaefer 2002; Johnson 2004), this was likely not a major factor in the observed differences, because correlations in both this and past studies (Ser- reze et al. 1999) were relatively high. The SNOTEL sites measure daily changes in SWE, but inaccurate measurements can be made because of instrumentation sensitivities and equipment issues, such as ice bridging across the snow pillow (Goodison et al. 1981). Snow courses cover a larger area than SNOTEL measure- ments, making the former more susceptible to environ- mental factors such as snow drifting, wind scour, and falling debris. Though both are located on relatively flat ground, minor differences in aspect, exposure, and veg- etative cover may also affect patterns and produce dif- ferent melt rates during the ablation period (Palmer 1986). The largest differences in interpolated SWE esti- mated from SNOTEL versus snow course data in the three subbasins (Fig. 5) occurred at 2500-3500 m, the elevation range with the most observations (Fig. 2) and the greatest amount of snow. However, in the Colorado basin as a whole, the interpolated SWE estimates dif- fered most at 1500-2000 m (Fig. 7), an elevation range with few SNOTEL sites (Fig. 2). Therefore, the inter- polation using SNOTEL data did not capture the SWE at lower elevations, producing a lower estimate in that area. Elevation ranged from 281 to 3656 m with an average of 1554 m in the Salt-Verde basin, from 1402 to 4229 m with an average of 2655 m in the Gunnison basin, from 1112 to 4098 m with an average of 1985 m in the San Juan basin, and from 0 to 4260 m with an average of 1730 m in the Colorado basin as a whole. The SNOTEL sites were located at a higher average elevation than the snow courses in both the Colorado basin and its subbasins shown in this analysis. There- fore, over- or underestimates in the interpolations of the two data sources relative to each other were eleva- tion independent in the sense that elevation does not bias the SWE estimate when considering either the Col- orado basin or its subbasins. However, the station lo- cation relative to the location of basin snow resources was important in producing low or high SWE estimates from interpolation. Conflicting SWE volume differ- ences were then caused by two factors. First, the eleva- tion distribution of snow course and SNOTEL sites re- porting differs, resulting in different representations of each elevation in the interpolation. Second, there were differences in point SNOTEL versus snow course data even at collocations, as shown in Fig. 4. The number of reporting stations increased until March and then decreased through Mayas the snow melted. The different data combinations showed simi- lar differences in the rmse over the snow season, except during the first and last measurements at the snow course sites. The higher snow course rmse for these dates was due to the small number of reporting snow course stations (Fig. 4a). Interpolated snow course SWE values were greater than those for SNOTEL at all elevations for the Colorado River basin as a whole (_105 km2), but less than SNOTEL values in the sub- basins (-104 km2), indicating that selection of SWE data for particular applications may need to be spatial scale specific. It also implies that other regions of the Colorado basin have measurement biases (e.g., loca- tion) between the data sources, differing from those subbasins studied here. Variogram analysis was used to evaluate differences in the spatial contiIlUity between SNOTEL and snow course data and to determine why interpolated esti- mates were different. Snow course data were correlated over a larger spatial scale than those of SNOTEL. This implies that an interpolation model using SNOTEL data should employ a smaller search radius, up to 300 km, while a model using snow course data should use a search radius of up to 500 km (Table 2). Nugget values were larger overall during wet (1993) versus dry (1999) years and were greater for SNOTEL than snow courses for all study years, indicating larger short-scale variabil- ity or measurement error within both the SNOTEL dataset and during periods of above- and below-aver- age snow. A Gaussian model, known as a model that approximates continuous phenomenon, was the best fit for snow course dataset versus a spherical model for SNOTEL dataset. This reflects the spatial pattern of the snow course as well as the greater number of sites than SNOTEL. 6. Conclusions While SNOTEL sites were implemented to replace the snow courses, measurements at collocated sites gen- erally indicate small differences in SWE at the point scale; but these differences translate to large interpo- lated volume differences at the basin scale. Differences are of most concern in wet years, such as 1993, in which the greatest absolute differences in SWE at the collo- cated sites occur; but large relative differences in dry years could also be a problem when data are used to inform water resources decisions. Interpolation to esti- mate regional SWE produces greater differences in ba- sin-wide water volumes (e.g., 50 rom of SWE over the entire Gunnison basin in 1993 at 3250-m elevation,