Geoscience Reference
In-Depth Information
Fig. 1 Schematic of the
distributed (''three-dimensional'')
EnKF update used for the
assimilation of coarse-scale snow
observations. See text for details.
Adapted from De Lannoy et al.
( 2010 )
covering it were unavailable—the two neighboring SWE retrievals (dark gray shading)
would still contribute to the update. The connection between the neighboring SWE retri-
evals and the model grid cell in question relies on horizontal model error correlations that
are due to, for example, errors in large-scale model forcing fields such as snowfall or air
temperature.
To assimilate SCF, the Noah model snow depletion curve acts as the observation
operator that converts fine-scale modeled SWE into SCF estimates. Unlike binary indi-
cators of snow presence, the continuous SCF observations used here can thus be assimi-
lated with an EnKF, taking advantage of the distribution of SCF values across the
ensemble. Snow-free or fully snow-covered conditions in the model-forecast ensemble
were addressed by supplementing the EnKF with rule-based update procedures (De
Lannoy et al. 2012 ). If at a given time and location all members of the model-forecast
ensemble are snow-free but the SCF observation indicates the presence of snow, then a
nominal amount of snow is added to the model forecast. If all forecast ensemble members
have full snow cover and the observed SCF indicates less than full cover, then the model-
forecast SWE and snow depth are reduced by a fixed fraction.
Figure 2 shows several observed and modeled snow fields for one snow season. The top
row shows the coarse-scale (25 km) AMSR-E SWE retrievals, with data missing when the
satellite swath does not fully cover the study area. MODIS fine-scale estimates of SCF,
shown in the second row, are available only for clear-sky conditions. The bottom four rows
of Fig. 2 show that the assimilation of coarse-scale AMSR-E SWE and fine-scale MODIS
SCF observations both result in realistic fine-scale spatial SWE patterns.
Through a quantitative validation of the assimilation results with independent mea-
surements at individual SNOTEL and COOP sites over the course of 8 years, De Lannoy
et al. ( 2012 ) demonstrate improvements from the assimilation of SWE and/or SCF retri-
evals in shallow snow packs, but not in deep snow packs (not shown). The validation also
shows that joint assimilation of SWE and SCF retrievals yields significantly improved
RMSE and correlation values. For example, the RMSE for SWE versus COOP site mea-
surements was reduced by 21 % (from 78 to 62 mm) through the joint assimilation of
satellite SWE and SCF retrievals. Furthermore, SCF assimilation was found to improve the
Search WWH ::




Custom Search