1 Introduction
Wiki Markup
The migration of the prototype R&D to the current version of the NFFS adaptor consists of several steps:
- Increase robustness module for operational purpose. This includes adding error handling and creation of log files, adding flags to module output/result files, module under subversion (SVN), simplified configuration (removing unnecessary items);.
- Updating test configurations SC080030 for case studies developed under the R&D project;.
- Documentation of configuration of the adapter in NFFS;.
Wiki Markup
2 Role in NFFS
The role of the Historic Forecast Performance Tool is to provide a probability distribution of the water level forecasts (or flow) conditioned on the deterministic water level forecast (or flow forecast). This can one, a few or many or percentiles or quantiles (including median or any other percentile/quantile like 0.05, 0.10, 0.25, 0.50, 0.75, 0.95).
The Historic Forecast Performance Tool adapter is linked to NFFS by means of the general adapter (see Figure 2.1).
Figure 2.1 Schematic Interaction Interction between between Delft-FEWS and Historic Forecast Performance Tool adapter (see Werner et al., 2004, Weerts et al., 2010 and references therein).
3 Method description
...
The Historic Forecast Performance Tool estimates the uncertainty due to all uncertainty sources affecting the forecast error. In NFFS (i.e. Delft-FEWS), the Historic Forecast Performance Tool is implemented as a post-processor on a deterministic forecast (see Figure 3.1).
Figure 3.1 Example Historic Forecast Performance Tool as postprocesser in NFFS.
3.2 Calibration Quantile Regression relationships
...
The following folder structure is necessary and contained in the ModuleDataSet file
config
QR_models
| Wiki Markup |
|---|
• locationId\[1\] |
| Wiki Markup |
|---|
• locationId\[2\] |
| Wiki Markup |
|---|
• locationId\[3\] |
• locationId[1]
• locationId[2]
• locationId[3]
• locationId[nWiki Markup
Work
4.2.4 Location and file naming convention
...
Several flags can be added to the timeseries in output.xml:
Flag="0" ”0” value is equal to original value (t < t0)
Flag="1" ”1” value is corrected and reliable
Flag="2" ”2” value is corrected, reliable but interpolated in between lead times
Flag="5" ”5” value is unreliable, extrapolated beyond the domain Quantile Regression relationships calibration
...
<timeStep unit="hour" multiplier="1"/>
<readWriteMode>add originals</readWriteMode>
...
<moduleInstanceId>ImportTelemetry</moduleInstanceId>
<valueType>scalar</valueType>
...
<valueType>scalar</valueType>
<parameterId>H.simulated.forecast.95</parameterId>
...
</timeSeriesSet>
<timeSeriesSet>
<moduleInstanceId>QR_2072_H_Forecast</moduleInstanceId>
...
</area> area>
<timeSeriesSet>
...
Koenker, R.: Quantile Regression, Cambridge University Press., 2005.
unmigrated-wiki-markup
Koenker, R.: Quantile regression in R: A vignette, \ [online\] Available from: [http://cran.r-project.org/web/packages/quantreg/vignettes/rq.pdf], 2010. \\
Koenker, R. and Basset, G.: Regression Quantiles, Econometrica, 46(1), 33-50, 1978.
Koenker, R. and Hallock, K. F.: Quantile Regression, The Journal of Economic Perspectives, 15(4), 143-156, 2001.
Weerts, A.H., J. Schellekens, F. Sperna Weiland, 2010. Real-time geospatial data handling and forecasting: Examples from DELFT-FEWS forecasting platform/system, IEEE J. of Selected Topics in Appied Earth Observations and Remote Sensing, 3, 386-394, doi: 10.1109/JSTARS.2010.2046882.
Weerts, A.H., H.C. Winsemius, J.S. Verkade, 2011. Estimation of predictive hydrological uncertainty using quantile regression: Examples from the National Flood Forecasting System (England and Wales), Hydrol. Earth Syst. Sci., 15, 255--265, doi:10.5194/hess-15-255-2011. Available from: http://www.hydrol-earth-syst-sci.net/15/255/2011/hess-15-255-2011.html.
Werner, M.G.F., Van Dijk, M. and Schellekens, J., 2004, DELFT-FEWS: An open shell flood forecasting system, In 6th international conference on Hydroinformatics, Liong, Phoon and Babovic (Eds.), World Scientific Publishing Company, Singapore, 1205-1212.
...
A Calibration of Quantile Regression Relationships
In order to derive the error models, a long enough hindcast needs to be performed. For each lead time considered, an error model (i.e. relation between forecast value, forecast error and leadtime) can be derived from such a hindcast. Background information can be found in Weerts et al. (2011) also at http://www.hydrol-earth-syst-sci.net/15/255/2011/hess-15-255-2011.html or directly via http://www.hydrol-earth-syst-sci.net/15/255/2011/hess-15-255-2011.pdf.
A procedure has been written in R to support this (https://repos.deltares.nl/repos/QuantileRegression/trunk/QR_derive). To use this procedure, first produce a long enough hindcast and make sure that observed values are written to a CSV file as follows:
...
QR_derive "CSV/VIKING1_obs_01.csv" "CSV/VIKING1_mod_01.csv" "VIKING" "1" "0.05, 0.25, 0.5, 0.75, 0.95" "-990" |
...