Successfully reported this slideshow.
We use your LinkedIn profile and activity data to personalize ads and to show you more relevant ads. You can change your ad preferences anytime.
Streamflow prediction in River Rhine: Exploring combinations of bias-correcting  forcing  and bias-correcting  flow Jan Ve...
Motivation and research questions <ul><li>Biases/uncertainty in predicted  forcing  used for streamflow prediction: </li><...
Research design Raw forcing (T,P) Hydrologic model Raw  flow Ensemble verification B-C forcing (T,P) Hydrologic model Raw ...
 
Data kindly provided by Florian Pappenberger @ ECMWF
Observed forcing data: E-OBS dataset Downloadable from KNMI @  http://eca.knmi.nl/
Bias-correction of temperature, precipitation and flow <ul><li>The random variables (one time/location): </li></ul><ul><ul...
Bias-correction of temperature, precipitation and flow (2/2) <ul><li>Temperature </li></ul><ul><li>normal regression: line...
B-C: preservation of space-time dependencies <ul><li>How to parameterize dependence? </li></ul><ul><li>Space-time patterns...
<ul><li>Skill of T correction </li></ul><ul><li>CRPSS = “% gain” over raw forecast  </li></ul><ul><li>~20-60% gain </li></...
<ul><li>Skill of P correction </li></ul><ul><li>~20-30% gain </li></ul><ul><li>Faster drop after 24 hour lead time </li></ul>
<ul><li>Skill of P correction for > “1-in-10  </li></ul><ul><li>day” observed P amount </li></ul><ul><li>~small gain or lo...
<ul><li>Skill of S for T and P correction. </li></ul><ul><li>~-10% to +10% </li></ul>
<ul><li>Skill of S for > “1-in-10” day observed S, with T and P correction. </li></ul><ul><li>~-40% to +20% </li></ul><ul>...
Next steps <ul><li>Q1: “What is the signal from bias-correction of  forcing  in streamflow?”: </li></ul><ul><li>Some way t...
Questions? <ul><li>(slides available from slideshare.net/janverkade) </li></ul><ul><li>Contact: </li></ul><ul><li>jan.verk...
Upcoming SlideShare
Loading in …5
×

"Streamflow prediction in River Rhine: Exploring combinations of bias-correcting forcing and bias-correcting flow

687 views

Published on

Published in: Technology, Business
  • Be the first to comment

  • Be the first to like this

"Streamflow prediction in River Rhine: Exploring combinations of bias-correcting forcing and bias-correcting flow

  1. 1. Streamflow prediction in River Rhine: Exploring combinations of bias-correcting forcing and bias-correcting flow Jan Verkade (Deltares and Delft University of Technology) James Brown (NOAA-NWS-OHD and UCAR)
  2. 2. Motivation and research questions <ul><li>Biases/uncertainty in predicted forcing used for streamflow prediction: </li></ul><ul><li>NWP models are skillful, but biased (mean, spread,..) </li></ul><ul><li>This bias/uncertainty propagates from forcing to flow </li></ul><ul><li>Bias-correction of precipitation is complex </li></ul><ul><li>Ultimately, flow bias-correction is always needed </li></ul><ul><li>Key research questions: </li></ul><ul><li>What is the signal from bias-correction of forcing in streamflow? </li></ul><ul><li>Is this signal maintained after bias-correction of flow , i.e. is forcing correction needed? </li></ul>
  3. 3. Research design Raw forcing (T,P) Hydrologic model Raw flow Ensemble verification B-C forcing (T,P) Hydrologic model Raw forcing (T,P) Hydrologic model Scenario 1 Scenario 2 Baseline B-C streamflow B-C streamflow
  4. 5. Data kindly provided by Florian Pappenberger @ ECMWF
  5. 6. Observed forcing data: E-OBS dataset Downloadable from KNMI @ http://eca.knmi.nl/
  6. 7. Bias-correction of temperature, precipitation and flow <ul><li>The random variables (one time/location): </li></ul><ul><ul><li>Predictand Y = observed temp/precip/flow. Assumed unbiased! </li></ul></ul><ul><ul><li>Potential predictors X = {X1,…,X5,…, Xm}; biased. </li></ul></ul><ul><li>The bias-corrected forecast: </li></ul><ul><li>How to parameterize for T and P? </li></ul><ul><ul><li>Parsimonious model (subject to skill!) </li></ul></ul><ul><ul><li>Model the statistical dependence (“traces”) </li></ul></ul>
  7. 8. Bias-correction of temperature, precipitation and flow (2/2) <ul><li>Temperature </li></ul><ul><li>normal regression: linear regression in normal space </li></ul><ul><li>Precipitation </li></ul><ul><li>logistic regression: linear regression in logistic space </li></ul><ul><li>Streamflow: </li></ul><ul><li>Krzysztofowicz approach: Hydrologic Uncertainty Processor </li></ul><ul><li>Prior: unconditional climatology </li></ul><ul><li>Posterior: distribution of flow conditional on ensemble mean </li></ul>
  8. 9. B-C: preservation of space-time dependencies <ul><li>How to parameterize dependence? </li></ul><ul><li>Space-time patterns of T and P </li></ul><ul><li>Cross-variable dependence in T and P </li></ul><ul><li>Critical for streamflow prediction </li></ul><ul><li>Empirical approach </li></ul><ul><li>Based on “Schaake shuffle” (Clark et al.) </li></ul><ul><li>Shuffle the bias-corrected ensemble members to preserve rank-ordering of the raw ensemble members </li></ul>
  9. 10. <ul><li>Skill of T correction </li></ul><ul><li>CRPSS = “% gain” over raw forecast </li></ul><ul><li>~20-60% gain </li></ul><ul><li>Gradual decline </li></ul><ul><li>with lead time </li></ul>
  10. 11. <ul><li>Skill of P correction </li></ul><ul><li>~20-30% gain </li></ul><ul><li>Faster drop after 24 hour lead time </li></ul>
  11. 12. <ul><li>Skill of P correction for > “1-in-10 </li></ul><ul><li>day” observed P amount </li></ul><ul><li>~small gain or loss </li></ul><ul><li>Failure of logistic regression to remove conditional bias (under-prediction of large P) </li></ul>
  12. 13. <ul><li>Skill of S for T and P correction. </li></ul><ul><li>~-10% to +10% </li></ul>
  13. 14. <ul><li>Skill of S for > “1-in-10” day observed S, with T and P correction. </li></ul><ul><li>~-40% to +20% </li></ul><ul><li>Loss of skill at long lead times. </li></ul><ul><li>Caution when “correcting” high P at long lead times! </li></ul>
  14. 15. Next steps <ul><li>Q1: “What is the signal from bias-correction of forcing in streamflow?”: </li></ul><ul><li>Some way towards answering that question </li></ul><ul><li>Need to establish why skilful forcing correction is not consistently translating into flow skill. </li></ul><ul><ul><li>Could it be due to the space-time and cross-variable dependence (“Schaake Shuffle”)? </li></ul></ul><ul><ul><li>Try Brown and Seo (2011) approach to conditional bias (bias-penalized kriging) </li></ul></ul><ul><li>Next, we’ll focus on Q2: </li></ul><ul><li>Is the signal from forcing bias-correction lost following flow bias-correction? </li></ul>
  15. 16. Questions? <ul><li>(slides available from slideshare.net/janverkade) </li></ul><ul><li>Contact: </li></ul><ul><li>jan.verkade@deltares.nl, twitter.com/janverkade </li></ul><ul><li>[email_address] </li></ul>

×