Ensemble Adaptive Data Assimilation Techniques Applied to Land-Falling North American Cyclones - Data Assimilation for Atmospheric, Oceanic and Hydrologic Applications - page 559

Maximum Variance Reduction for Temperature Observations

9

normalized
nonnormalized

8

7

6

5

4

3

2

1

0

0 20 40 60 80 100 120 140 160 180 200

Cyclone

Both normalized (nondimensional) and non-normalized (hPa 2 / Fig. 23.7 values of maximum

variance reduction of the response function for all cyclones for observations of temperature.
Normalized values are multiplied by 10 for ease of comparison with non-normalized values

and 40 are offset. This indicates that whereas a larger absolute variance reduction is
estimated from temperature observations at the peak of the non-normalized values,
the response function variance must be somewhat larger for that cyclone such that
the estimated fraction of response function variance is smaller. This indicates that
for specific cases, the impacts of EnKF targeted observations can be viewed with
differing degrees of importance depending on whether these impacts are determined
by the total or the fraction of estimated response function variance reduction.
Another interesting result that can be found by analyzing targeted observations
for many cases is how the location of the most significant targeting locations
vary in time. For a specific high-impact weather event, if the targeting regions
remained constant over many cases, strong support would exist for taking routine
observations in those locations. If targeting regions were not constant, the degree
to which they vary would provide crucial information toward how to best design
an adaptive observing network. Figure 23.8 shows the mean estimated maximum
variance reduction calculated over all cyclones throughout the vertical. Interestingly,
pressure observations show roughly a constant impact throughout the troposphere.
This reveals that although the maximum estimated targeting values tend to occur
in the lower atmosphere for pressure observations (Fig. 23.4 ), values are nearly
constant within the entire troposphere. In turn, there are no preferred targeting
locations in the vertical with regard to pressure observations to improve land-falling
cyclone forecasts. Wind and temperature observation targeting regions, however,
show two distinct peaks in the vertical. Wind targeting regions show peaks near
400 hPa and the surface, whereas temperature targeting regions reveal the 250-hPa
and the 600-hPa levels to be most important. It seems reasonable that the important
temperature targeting locations near 250-hPa are due to large variance near the