All-Sky Satellite Radiance Data Assimilation: Methodology and Challenges - Data Assimilation for Atmospheric, Oceanic and Hydrologic Applications - page 473

non-Gaussian all-sky radiance observation errors inherently implies a need for better
handling of nonlinearity. Therefore, if the available data assimilation algorithm is
not very good for nonlinear operators, it is probably good to avoid introducing non-
Gaussian errors.

19.4 Summary and Future

Data assimilation of all-sky satellite radiances is a difficult problem that puts to test
data assimilation methodology and algorithmic solutions that are used today. Since
the information from all-sky radiances is potentially very valuable, using “short-
cut” solutions is not acceptable. We discussed several critical aspects of successful
all-sky radiance data assimilation, with emphasis on forecast error covariance,
Hessian preconditioning, non-differentiability, and correlation of observation errors.
Also, the focus of our presentation was on how variational and ensemble data
assimilation can handle these challenging problems. In conclusion, both methods
have their advantages and disadvantages and likely best approach is to develop
hybrid variational-ensemble methods that can selectively choose the better option.
One can also adopt other methodologies that can possibly address better the
difficulties arising in variational and ensemble methods.
Although we did not describe in detail all issues related to all-sky radiances,
such as verification, or algorithmic details related to a specific methodology, they
also have to be taken into account. There may be research issues that we are not
aware of at present, but will be eventually addressed as all-sky radiance assimilation
research becomes widespread.
One important implication of presented challenges is that development of new
data assimilation methodology that is better suited for all-sky satellite radiance
assimilation has to be comprehensive. For example, solving nonlinear issues cannot
be properly done without addressing non-Gaussian errors or without utilizing the
full power of Hessian preconditioning. Similarly, better definition of forecast error
covariance will not be fully beneficial unless it is combined with superior Hessian
preconditioning that can maximize the benefit of nonlinear minimization. As
suggested here, it may not be always necessary to develop most complex algorithms
to solve the challenges of all-sky radiance assimilation. There are applications that
may accept simple solutions to some of the issues, but it is important not to dismiss
an issue before its impact is well understood. For example, although one can opt
for uncorrelated observation errors at the end, it first needs to evaluate the potential
impact of correlations or to investigate statistics of observations errors.
Development of hybrid variational-ensemble and other new methodologies is an
ongoing effort and will likely produce an improved all-sky radiance assimilation
methodology capable of extracting maximum information from this valuable data.

Acknowledgements This work was supported by the National Science Foundation Collab-
oration in Mathematical Geosciences Grant ATM-0930265, and by NOAA NESDIS Grant