Validation of satellite based SWE
Validation of satellite-based SWE can be done by calculating the correlation coefficient, root-mean-square error and bias. Contingency table statistics cannot be applied because SWE is not a nominal measure (yes/no event), but a quantitative measure.
Satellite-based SWE validation with ground data was performed at three meteorological stations in the Baltic States: Kuusiku (Estonia), Dobele (Latvia) and Vilnius (Lithuania) (Fig. 2). Statistical evaluation of the satellite data's accuracy varied between the regions and stations (Table 1). The largest bias (-4.1 mm) and RMSE (28.0 mm) were determined at Dobele. However, the sample size at this location was very small due to the gaps in ground measurements and satellite observations. Correlation coefficient between satellite-based SWE and in-situ measurements was 0.64 for Dobele, 0.80 for Vilnius, and 0.91 for Kuusiku.
Figure 2: The meteorological stations which were used for satellite-based SWE validation.
Table 1: Summary of satellite based SWE performance over the Baltic States.
| Station | RMSE (mm) | Bias (mm) | Corr. Coeff | Sample size |
|---|---|---|---|---|
| Dobele, Latvia | 28,0 | -4,1 | 0,64 | 35 |
| Kuusiku | 17,9 | 3,7 | 0,91 | 79 |
| Vilnius, Lithuania | 11,6 | 4,9 | 0,80 | 85 |
Validation results indicate that GlobSnow-2 data has a tendency to overestimate SWE (bias in Vilnius was 4.9 mm and Kuusiku 3.7 mm). This overestimation may be related to the documented bias of GlobSnow project data to overestimate SWE under shallow snow conditions (Luojus et al., 2010). Figure 3 shows a comparison of the results of satellite-based SWE and ground measurements at the Vilnius meteorological station.
 |
 |
Figure 3: Satellite-based SWE performance compared to the in-situ data at the Vilnius meteorological station. On the left: scatter plots of SWE estimates. On the right: histogram of the differences. GlobSnow-2 product shows a tendency to overestimate the SWE value.