Geoscience Reference
In-Depth Information
takes on a role that is conceptually similar to that of the observation operator used for the
partitioning of TWS information into its water cycle components (Sect. 3.2 ).
The fourth and final challenge addressed in the paper discusses the selection of the types
of observations that are most relevant for the analysis of poorly observed variables. For the
analysis of one such variable, root zone soil moisture, the use of gauge- and satellite-based
precipitation observations along with active and passive surface soil moisture retrievals
was investigated (Sect. 3.4 ). It was shown that the MERRA-Land surface reanalysis
provides better estimates of root zone soil moisture than MERRA due to the use of gauge-
based precipitation observations in MERRA-Land. Next, the potential skill gained from the
assimilation of surface soil moisture retrievals was investigated. It was demonstrated that
improved root zone soil moisture estimates can be obtained even where the skill of the
assimilated surface soil moisture retrievals is somewhat poorer than that of the model
estimates of surface soil moisture. For maximum coverage and accuracy, both active and
passive retrievals should be assimilated. Finally, it was shown that the use of precipitation
observations and the assimilation of surface soil moisture retrievals contribute significant
and largely independent amounts of root zone soil moisture information. Therefore, future
reanalyses should use both of these observation types. This finding is consistent with the
general expectation that using more observations in a data assimilation system will
improve its output.
In some cases (for example, Sects. 3.1 and 3.2 ), the appropriate observation operator
and assimilation system configuration entail that neighboring grid cells (or land model
tiles) are no longer computationally independent in the assimilation system, even if they
are independent in the land model (Reichle and Koster 2003 ). These computational
dependencies arise through spatially correlated perturbation fields or spatially distributed
analysis update calculations. Such ''three-dimensional'' land data assimilation systems
therefore necessitate greater computational resources than more simplistic, ''one-dimen-
sional'' assimilation systems where all model grid cells (or tiles) are treated independently.
It is assumed here that the purely technical challenge of computational demand can be
overcome with sophisticated software engineering and the increasing availability of
affordable and massively parallel computing architectures.
5 Conclusions and Outlook
The present paper focused on the seeming mismatch between satellite observations and the
water cycle variables of interest, and how a mismatch can be overcome through careful
design and application of a land data assimilation system. Responding to the challenge
questions of Sect. 1 , we find that, if designed properly, a land data assimilation system can
enable
1. the horizontal downscaling of coarse-scale satellite observations,
2. the partitioning of vertically integrated satellite measurements such as TWS into their
water cycle components,
3. the direct assimilation of satellite radiances for soil moisture or snow analyses, and
4. the propagation of information from observed fields such as precipitation and surface
soil moisture into variables such as root zone soil moisture, that are of great interest
but are not directly observed by satellites.
Naturally, many challenges still lie ahead. State-of-the-art land data assimilation
algorithms are only now emerging in operational systems. Much of the recent progress has
Search WWH ::




Custom Search