Geoscience Reference
In-Depth Information
noise, and synthetic albedo observations were taken from the truth run with 5 % white
noise applied.
Forecasts were generated by an ensemble of 100 LSM replicates with perturbations
applied to forcing and model parameters, which allowed the mean forecast state and the
forecast error covariance to be determined from the ensemble statistics. Synthetic albedo
observations were assimilated daily at 1 pm and passive microwave observations at 1 am
to mimic MODIS and AMSR-E overpass times.
The regular assimilation of SSM/I frequencies alone significantly reduced both bias and
root mean square error (RMSE) of SWE by approximately 85 % relative to the open-loop
simulation. The EnKF approach also allowed an assessment of the contribution of each
channel, which indicated that the majority of the SWE improvement occurred due to the
assimilation of the 37-GHz channel at both polarisations. The 89-GHz channel appeared to
marginally worsen the SWE analysis by nudging it away from truth; however, it signifi-
cantly improved the grain size analysis, which was vital for the brightness temperature
simulations.
Having demonstrated the assimilation approach using a synthetic experiment, the later
work of Durand et al. ( 2008 ) used data from the University of Tokyo's Ground Based
Microwave Radiometer (GBMR-7) and snowpits at NASA's Cold Land Processes Experi-
ment (CLPX) to test the performance of the MEMLS radiative transfer model. Furthermore,
they were able to identify accuracy criteria for the snow state variables. They determined that
simulated optical grain size should be accurate within ±0.045 mm and the density of melt-
refreeze layers within ±40 kg m -3 in order for predicted brightness temperature errors to be
small enough that the assimilation procedure improves the analysis.
Further work has considered the effect of spatial scaling on the analysis, with different
spatial resolutions in LSMs and microwaves explored in De Lannoy et al. ( 2010 ), while
Andreadis et al. ( 2008 ) discuss how to account for snow's spatial variability in an
assimilation scheme.
3.2 Globsnow
3.2.1 Methodology
The European Space Agency (ESA) Globsnow project's aim is 'production of global long
term records of snow parameters intended for climate research purposes on hemispherical
scale' (Finnish Meteorological Institute 2012 ). The Globsnow SWE product is a system
where the prior state is estimated from field observations of snow depth, with updates
related to the satellite-observed brightness temperature difference (DT B ; V ) at vertical po-
larisation between channels near 19 GHz (T19 V) and 37 GHz (T37 V).
The use of a brightness temperature difference reduces the sensitivity of the satellite
observations to absolute temperatures; if non-snow surfaces are in the field of view and
they have the same emissivity at both 19 and 37 GHz, then their effect on the measured
brightness temperature difference is dependent only on the area they cover and is inde-
pendent of their temperature.
Globsnow produces maps of SWE across the Northern Hemisphere on a 25-km Equal
Area Scalable Earth (EASE) grid, with areas defined as too watery ([50 % open water) or
too mountainous (standard deviation of elevation [200 m) masked out. Largely based on
the approach of Pulliainen ( 2006 ), its methodology is explained in detail in Takala et al.
( 2011 ) and proceeds as follows:
Search WWH ::




Custom Search