Geoscience Reference
In-Depth Information
obtain analysis increments, the Kalman gain would be multiplied with innovations in DTb
(that is, the difference between actual AMSR-E observations of DTb and ANN predictions
of DTb).
The Kalman gain computed for February 6, 2003 ranges from -10 to 15 mm K
-1
as
illustrated in Fig.
9
. A gain of 1 mm K
-1
equates to an increase of 1 mm in the posterior
(updated) modeled SWE for a 1 K innovation (that is, for a difference of 1 K between
AMSR-E DTb measurements and ANN DTb predictions). Similarly, a negative Kalman
gain in the presence of a positive-valued innovation would equate to a reduction in
modeled SWE. Most importantly, the results suggest that there is a nonzero error corre-
lation between the model SWE forecasts and the simulated DTb measurements across
much of the North American domain. Overall, the results suggest that the ANN could serve
as a computationally efficient observation operator for radiance-based snow data assimi-
lation at the continental scale.
3.4 Observation Selection for a Root Zone Soil Moisture Analysis
Knowledge of the amount of moisture stored in the root zone of the soil is important for
many applications related to the transfer of water, energy, and carbon between the land and
the atmosphere, including the assessment, monitoring, and prediction of drought (Sene-
viratne et al.
2010
). At the global scale, soil moisture estimates are usually based on two
sources of information: (1) direct observations of surface soil moisture from satellite and
(2) observation-based precipitation forcing driving a numerical model of soil moisture
dynamics. However, neither surface soil moisture retrievals nor precipitation observations
provide direct measurements of soil moisture in the root zone. The selection of the most
relevant types of observations for a root zone soil moisture analysis therefore presents an
important conceptual problem.
A priori, it is not obvious whether the estimation of root zone soil moisture would
benefit more from the use of precipitation observations (as, for example, in the Global
Land Data Assimilation System; Rodell et al.
2003
) or from the assimilation of surface soil
moisture retrievals (as, for example, illustrated by Reichle et al.
2007
). This section pro-
vides examples of both approaches. First, a land surface reanalysis that relies on observed
precipitation is presented, followed by a root zone soil moisture analysis that is based on
the assimilation of surface soil moisture retrievals. Finally, the two sources of soil moisture
Fig. 9 Histogram of the Kalman
gain on February 6, 2003 for
SWE versus
DTb = [Tb
V
(18 GHz) -
Tb
V
(36 GHz)]
