Site Specific Results email - June 28, 2001
Subject: Performance of WFO Site-Specific Hydrologic Model
Date: Thu, 28 Jun 2001 12:28:07 -0500
From: "Bob Cox" <Bob.Cox@noaa.gov>
To whom it may concern (and may be interested) ...
Since another quarter has flown by, I thought this might be
a good time to
let everyone know where we are with development of the WFO site-specific
hydrologic forecast model. As you know, MBRFC has been working with
WFO-EAX for several months now to develop, implement, and test the
functionality of the model, with the goal of delivering it to the
AWIPS Build 5.1.3. To date, Russ Erb has been primarily responsible
software development at OHD and has been able to provide new, improved
versions of the executables to us on a fairly regular basis. He
great strides toward creating a tool which should become extremely
to the WFOs.
At MBRFC, we have been busy defining the functionality of the
performing the necessary hydrologic development, and testing the
a variety of circumstances. We now have the model up and running
system for the thirty locations for which we currently provide headwater
guidance in the WFO-EAX hsa. Accomplishing this task required
(re)development of 1-hour unitgraphs for all sites, then defining
unitgraphs, rating curves, basin boundaries, etc. in the IHFS DB.
we finally have everything functional on our system, we are ready
the necessary software and parameters to the EAX system for more
operational testing. Meanwhile, we will begin expanding our development
efforts into our remaining WFO areas.
Initial testing of the model performance looks promising. We
the model against existing procedures, such as OFS and headwater
are reasonably satisfied with the results. Most of this testing,
has been for relatively moist soil conditions ... things just haven't
out much in the EAX area this spring. We do intend to continue testing
the dry side of things, however, hopefully later this summer.
I have also tested the performance of the model for about a
rainfall events over the past couple of months, with good results.
nearly every case, the model has produced a crest forecast within
a foot or
less of the observed value, provided that an accurate estimate of
rainfall time series was input. Typically, however, the StageII
have been significantly lower than the gage reports and had to be
upwards before an accurate forecast was possible.
I have attached the results of four case studies from rainfall
last week. These results are typical of what we have been seeing
We have also produced a slide presentation of our experiences with
so far and will update this as necessary. You can find it on our
(This is essentially the same
slideshow which Larry presented at the WR SOO/DOH conference last
month.) (The slide show currently is not available, please check back at a later time)
That's all I know. If you have questions or comments, feel free.
Performance of the WFO Hydrologic Forecast Model
For the Rainfall Events of June 20-21, 2001
Over the past couple of months, MBRFC has been testing the performance
of the newly developed WFO hydrologic model against existing procedures
such as OFS and flood advisory (headwater) tables and, sporadically
at least, during some actual rainfall events. The attached charts
depict the performance of the model for four locations in and around
Kansas City during the rainfall events of June 20-21, 2001, and
are very indicative of the accuracy we have encountered with the
model to date.
In the attachments, the hydrographs shown in red are the result
of applying the Stage II estimates from the EAX radar, without any
further adjustments, to compute the hourly MAPs for each forecast
location. A significant under-simulation is evident, producing crest
forecasts which, on average, would have been 3.7 feet too low.
The hydrographs shown in blue in the attachments are the result
of adjusting the Stage II estimates by factors ranging from 1.17
to 1.75, based on the average bias demonstrated at locations where
storm total precipitation reports were available. This improved
the simulations dramatically, resulting in a mean absolute forecast
error at crest of 0.6 foot. The event at Easton, Kansas, was noteworthy.
This constituted a new flood of record for this location and forced
the evacuation of over 200 residents. The model (adjusted) produced
a crest forecast within 0.1 foot of the observed crest, nearly 9
feet over flood stage.
In all four cases, the timing of the crest forecasts were within
an hour of that observed. All computations for these events were
performed by applying hourly MAPs to the API-MKC runoff model, keying
to the hourly FFH value valid for 12Z of June 19th. Although the
model allows for baseflow adjustment based on a pre-existing river
stage, this option was not employed because that adjustment had
already been accounted for in the computation of FFH values.
To summarize, when provided accurate estimates of precipitation,
these modeling techniques seem to be performing exceptionally well.
The advantage of the one-hour computation interval is evident. However,
without adequate ground-truth by which to determine the extent that
the WSR-88D is underestimating precipitation, accurate hydrologic
forecasts are not consistently possible. Like every other tool we
deal with, it is only as good as the data that drives it.
at Easton, Kansas
at Mosby, Mo
Blue River at
at Overland Park, Kansas