|
|
|
|
|||
|
|
|
|
|||
|
|
GODAE OceanView Observing System Evaluation Dr Peter R. Oke And Dr. Gilles Larnicol Latest Results - Long-term Goal - Analysis
self-sensitivities Past Results - Publications - Workshops Coming soon The goal of
the Observing System Evaluation Task Team, under GODAE OceanView, is to
quantify the impact of ocean observations on Operational ocean forecast and
analysis systems. See www.godae.org/OSSE-OSE-home.html.
Every
assimilation system combines a model background field with a set of observations
to produce an analysis.
The degree to which the analysis fits the observations depends on the assumed
observation errors, the assumed background field errors, and the degree to
which the background innovations project onto the assumed background error
covariance functions. The background error covariance functions are typically
either analytical functions, such as Gaussian functions, or numerical
functions, from an ensemble or an adjoint and tangent linear model. The
importance of every assimilated observation for a given analysis can be
quantified by analysis self-sensitivities. Analysis self-sensitivities can be
computed for any assimilation system, as follows: Step
1: Compute an analysis using the available observations and
store the vector of assimilated observations o, and the resulting analysis at the
observation location Ha; Step
2: Perturb the observations according to their assumed
observation errors and compute a second analysis; storing the vector of
perturbed observations o*, the resulting analysis at the observation location
Ha*,
and the vector of observation error variance e2; Step
3: Compute the analysis self-sensitivities, stored as a
vector HK
(using matlab notation): HK
= (o*-o).*(Ha*-Ha)./(e2) So
there is an element of HK for every assimilated observation. Where HKi is small (large), the
analysis is insensitive (sensitive) to ith observation. For
more robust results, multiple realizations of the analysis self-sensitivity
should be computed, by repeating steps 2 and 3; and the results averaged, and
uncertainty assessed (e.g., via the standard error). Suppose
an observation is assumed to have a typical error of 1; and the observation
is perturbed by a value of 1; and the analysis changes at that observation
location by a value of 1. In this case, the analysis self-sensitivity is 1,
and the analysis was sensitive to a change in that observation. So, according
to the assimilation system used, and given all other available observations,
the observation is important. Suppose
an observation is assumed to have a typical error of 1; and the observation
is perturbed by a value of 1; and the analysis changes at that observation
location by a value of 0.1, or even zero. In this case, the analysis
self-sensitivity is 0.1, and the analysis was not very sensitive to a change
in that observation. So, according to the assimilation system used, and given
all other available observations, the observation is unimportant. Coming soon Publications The relevant references for analysis sensitivity are: Cardinali, C., S.
Pezzulli, E. Andersson, 2004:
Influence-matrix diagnostic of a data assimilation system. Quarterly
Journal of the Royal Meteorological Society, 130, 2767-2786. Chapnik, B., G.
Desroziers, F. Rabier, O. Talagrand, 2006: Diagnosis
and tuning of observational error in a quasi-operational data assimilation
setting. Quarterly Journal of the Royal Meteorological Society, 132, pp.
543-565 doi: 10.1256/qj.04.102. Rabier, F., P. Gauthier,
C. Cardinali, R. Langland, M. Tsyrulnikov, A. Lorenc, P. Steinle, R. Gelaro,
K. Koizumi, 2008: An update on THORPEX-related research in data assimilation
and observing strategies. Nonlin. Processes Geophys.,
15, 81-94. Workshops Details
of past GODAE workshops on Observing System Evaluation can be found on the
official GODAE OceanView
website. |
|
|||
|
|
Last updated 27/11/06 | Legal Notice and
Disclaimer |
Copyright |
|
|||