Geoscience Reference
In-Depth Information
Fig. 3 Map of the 1,800,000 km
2
Mackenzie River Basin including GEOS-5 topography, sub-basin
delineation, and GRDC observation locations (solid dots). Adapted from Forman et al. (
2012
)
Additional work was conducted to analyze modeled river discharge estimates against
ground-based gauging stations. The findings (not shown) suggest that the assimilation of
GRACE observations causes little or no change in the mean difference and RMSE of
modeled river discharge, but that small, statistically significant improvements in the
anomaly correlations were found. Improvements in the modeled river runoff anomalies are
attributed to a redistribution of the water mass from the snow pack during the accumulation
phase into the subsurface during the subsequent ablation and runoff phase. This redistri-
bution of water by the assimilation framework effectively retains water within the
hydrological basin for a longer period of time, which results in small but statistically
significant improvements in modeled estimates of river discharge.
Investigation of the analysis increments can provide valuable insights into the behavior
of the assimilation procedure and track how much and at what time water is being added to
or removed from the individual TWS components. The thin, solid line in Fig.
5
shows the
increments made to the subsurface water component. Averaged over the Mackenzie River
basin and the 7-year experiment period, a total of 12.5 mm of water has been added into
the subsurface by the assimilation procedure. This is most evident during the spring and
summer. The thick, dashed line in Fig.
5
shows the increments for SWE. Averaged over
time and space, SWE is removed during the accumulation phase with a small amount
added back during the ablation and runoff phase for a total SWE increment of -45.1 mm.
Acting together, the analysis increments to the subsurface water and SWE serve to reduce
mass during snow accumulation and then increase the mass during ablation and runoff.
These two phenomena essentially constrain the amplitude of the modeled TWS dynamics
to achieve better agreement of the model estimates with the GRACE retrievals.
