Download the paper:
Abstract:
We assess the importance of horizontal error correlations in
background (that is, model forecast) fields for large-scale
soil moisture estimation by comparing the performance of
one- and three-dimensional
ensemble Kalman filters (EnKF) in a twin experiment.
Over a domain centered on the United States Great Plains we use gauge-based
precipitation data to force the ``true'' model solution, and reanalysis
data for the prior (or background) fields.
The difference between the two precipitation data sets
is thought to be representative of errors we might encounter
in a global land assimilation system.
To ensure realistic conditions the synthetic observations of surface soil
moisture match the spatio-temporal pattern and expected errors of
retrievals from the Scanning Multichannel Microwave
Radiometer (SMMR) on the Nimbus-7 satellite.
After filter calibration, average actual estimation errors in the (volumetric)
root zone moisture content are 0.015 m3/m3 for the 3D-EnKF,
0.019 m3/m3 for the 1D-EnKF,
and 0.036 m3/m3 without assimilation.
Clearly, taking horizontal error correlations into account
improves estimation accuracy.
Soil moisture estimation errors in the 3D-EnKF are smallest
for a correlation scale of 2 degrees in model parameter and forcing
errors, which coincides with the horizontal scale of difference fields
between gauge-based and reanalysis precipitation.
In our case the 3D-EnKF requires 1.6 times the computational effort
of the 1D-EnKF, but this factor depends on the experiment setup.