Implementing Probabilistic Climate Outlooks within a Seasonal Hydrologic Forecast System Andy Wood and Dennis P. Lettenmaier Department of Civil and Environmental Engineering for Session H21H Annual Fall Meeting, AGU San Francisco December 14
MAGIC (or to some, ALCHEMY) at least: spatial disaggregation temporal disaggregation if not also: bias correction variable transformation Hydrologic Prediction using Climate Forecasts OUTLINE western U.S. forecasting system use of NCEP CPC seasonal outlooks CPC downscaling tests Conclusions climate forecasts at a coarse scale (in space and sometimes time) hydrologic forecasts at a scale suitable for water management application (e.g., relatively finer) Needs Verification
Background: W. US Forecast System
ESP ENSO/PDO ENSO CPC Official Outlooks Seasonal Forecast Model (SFM) CAS OCN SMLR CCA CA NSIPP-1 dynamical model VIC Hydrolog y Model NOAA NASA UW Seasonal Climate Forecast Data Sources
Background: W. US Forecast System Soil Moisture Initial Condition Snowpack Initial Condition
Background: W. US Forecast System targeted statisticse.g., runoff volumes monthly hydrographs
Background: CPC Seasonal Outlooks e.g., precipitation
Background: CPC Seasonal Outlook Use spatial unit for raw forecasts is the Climate Division (102 for U.S.) CDFs defined by 13 percentile values ( ) for P and T are given
Background: CPC Seasonal Outlook Use probabilities => anomalies precipitation
Approach: CPC Seasonal Outlook Use climate division anomalies => model forcing ensembles we want to test (1) and (2): testing (2) is easy, using CPC retrospective climate division dataset testing (1) is more labor-intensive, less straightforward (2) “downscaling” (1) “Shaake Shuffle” CPC monthly climate division anomaly CDFs spatial / temporal disaggregation ensemble formation monthly climate division T & P ensembles daily 1/8 degree Prcp, Tmax and Tmin ensemble timeseries
Approach: CPC Seasonal Outlook Use Downscaling Evaluation Question 1: Does hydrologic simulation driven by the downscaled forcings reproduce expected* streamflow mean and variability? *expected = simulated from 1/8 degree observed forcings (Maurer et al.) Spatial Disaggregation transform CPC climate division retrospective timeseries ( ) into monthly anomaly timeseries (%P, delta T) apply anomalies to 1/8 degree monthly P and T means (from UW COOP- based observed dataset of Maurer et al., 2001) yields: 1/8 degree monthly P and T timeseries Temporal Disaggregation daily weather generator creates daily P and T sequences for 1/8 degree grid scale and shift sequences by month to reproduce monthly 1/8 degree P and T timeseries values
Results: CPC-based flow w.r.t. UW obs dataset Answer to Question 1: (will the resulting forcings reproduce expected* streamflow mean and variability?) SURPRISINGLY WELL, BUT NOT QUITE mean std dev
Results: CPC-based flow w.r.t. UW obs dataset Additional examples show similar results Mean pretty well reproduced; variability biased mean std dev
Results: CPC temp/precip w.r.t. UW obs dataset based on
Results: CPC temp/precip w.r.t. UW obs dataset based on
Results: CPC Seasonal Outlook Use Downscaling Evaluation Question 2: Will bias-correction of CPC climate variables correct errors in the derived streamflow simulation? specific to calendar month and climate model grid cell cpc distrib
Results: CPC-based flow w.r.t. UW obs dataset Answer to Question 2: (does bias correction help?) YES (but) mean std dev
Results: CPC-based flow w.r.t. UW obs dataset Additional examples show similar results Mean pretty well reproduced; variability improved mean std dev
For more information: Conclusions Our current approach for downscaling CPC seasonal outlooks is adequate from hydrologic perspective, but would be improved by a bias-correction step. A systematic temperature bias appears to exist in the CPC climate division dataset, relative to the UW 1/8 degree observed climate dataset (aggregated to the CPC C.D. unit). Still needed… Assessment of ensemble formation step in CPC forecast use.
Questions?
Framework: Downscaling CPC outlooks downscaling uses Shaake Shuffle (Clark et al., J. of Hydrometeorology, Feb. 2004) to assemble monthly forecast timeseries from CPC percentile values
Results: CPC-based flow w.r.t. UW obs dataset Additional examples show similar results Mean pretty well reproduced; variability biased mean std dev
Results: CPC temp/precip w.r.t. UW obs dataset based on
Results: CPC-based flow w.r.t. UW obs dataset Additional examples show similar results Mean pretty well reproduced; variability improved mean std dev