|
1
Test Bed Leaders
- Martyn Clark
(CIRES, University
of Colorado)
- John Schaake
(NWS Hydrology Laboratory)
2
Test Bed Description
- Basins: TBD;
suggest a subset of the MOPEX basins,
scattered across diverse climate regimes.
- Space/time
scales of
interest: Seek to produce
forecasts at individual stations at lead times of 1-14 days. These would be used to produce streamflow
forecasts at basins ranging in size from ~500 to 20,000 km2
Experimental
Design:
Participants
will be asked to “downscale” the CDC MRF Reforecast Dataset to produce
ensemble
precipitation and temperature estimates at individual stations for the
selected
test basins. Participants should provide
downscaled MRF output in a prearranged format (e.g., CF-compliant
NetCDF files). The format may be a
separate file for each
station that has a structure with two variables (precipitation and
temperature)
with dimensions (forecast initialization time, forecast lead time,
ensemble
member).
The
downscaled MRF output will be used as input to a hydrologic model to
produce
forecasts of streamflow (suggest using the distributed version of the
SNOW-17/ Sacramento
model, but any
model can be used). This model will
include a prearranged method to distribute the downscaled precipitation
and
temperature estimates across a basin. The
model will be configured so it can be run as a “black
box”
by any of
the testbed participants to produce NetCDF ensemble streamflow forecast
output
(i.e., with identical dimensions to the downscaled model input files
produced
by the participants). The hydrological
model can be run by the testbed leaders, if desired.
Test-bed
leaders will compute the following verification statistics:
- Climatological
Bias
- Climatological
space/time/inter-variable correlations
- Reliability,
Discrimination for different thresholds
- Brier Skill
Score for
different thresholds
- Ranked
Probability
Skill Score
- Ranked
Histogram
- Relationships
between
ensemble spread and forecast skill
Note
that participants
have no
responsibility to run the hydrological model and compute verification
statistics themselves. If participants
do perform these tasks as part of their methodological development,
they are
encouraged to document their modifications thoroughly.
3
Key Scientific Questions
What are the advantages
and limitations
of different methods for extracting information from Numerical Weather
Prediction models, for the purposes of forecasting streamflow?
4
Key
Objectives of the Research Project
- Identify the space-time
scales for which forecast skill is present, for different variables,
and develop methods to extract and combine information at different
space-time scales
- Identify the MRF output
variables that can be used to provide sub-grid information—for example,
wind and humidity output can be used in a statistical model to
replicate orographic precipitation processes, and provide local-scale
information that is not present in the raw precipitation output [Clark
and Hay (2004) have a table that summarizes the frequency that
different variables are used in their regression equations].
- Identify of the sample
size required to reliably forecast precipitation, temperature, and
streamflow, for different thresholds.
5
Data Resources
Test-bed leaders will
provide:
- CDC Reforecast model
output for the selected test basins, and example code to read it
- Station data for the
selected test basins, and example code to read it
- Example code to define
the NetCDF output files (i.e., those files produced by the test-bed
participants, that contain ensemble downscaled output at individual
stations)
- The hydrologic model,
with wrappers around it, so it can read the NetCDF output files and
produce forecasts of streamflow
- Verification code that
reads the NetCDF output files and computes verification statistics.
- Plotting routines
6
Recent activities
Statistical Downscaling Testbed Update
2006.
|
