Hydrologic Applications of Ensemble Forecasts by h3p00p3d


									             Hydrologic Applications of
               Ensemble Forecasts

                            Julie Demargne

                           Hydrology Laboratory
                     Office of Hydrologic Development
                     NOAA/National Weather Service

              Ensemble User Workshop, NCEP, Oct. 31-Nov. 2, 2006
11/02/2006                                                         1

      •   Hydrologic ensemble products and services
      •   Ensemble Streamflow Prediction (ESP) system
      •   Requirements
      •   Leveraging and collaboration
      •   Conclusions

11/02/2006                                              2
             Hydrologic Products and Services

• Objectives:
      – Provide seamless and consistent probabilistic forecasts for all
        lead times
      – Reduce and account for both input and hydrologic uncertainties

• The methodology is currently tied to the lead times of available
  meteorological forecasts:
      – 1 to 5 days: short term
      – 6 to 14 days: medium range
      – Two weeks and beyond: long range

• The spatial scale ranges from a few km 2 to the continental

11/02/2006                                                                3
             Hydrologic Products and Services
• Seamless probabilistic forecasts for all lead times

11/02/2006                                              4
                      Hydrologic Products and Services
• Uncertainties in Hydrologic Forecast

                      Meteorological/Input uncertainty

                                 Reduced uncertainty due to pre-processing

                      Parametric uncertainty
                                          Reduced uncertainty due to calibration
                      Uncertainty in initial conditions

                                    Reduced uncertainty due to data assimilation

11/02/2006                           Lead Time                                     5
             Hydrologic Products and Services

  • Ensemble inputs needed:
       – Precipitation, temperature, potential evaporation, freezing level
       – For each River Forecast Center (RFC) basin and for all lead
         times (1 hr to 1 yr)

  • Ensemble outputs:
       – Streamflow, river stage, soil moisture, channel storage, etc.

  • Verification data and statistics:
       – For all ensemble forecasts
       – Requires retrospective forecasts (hindcasts)

11/02/2006                                                                   6
              Hydrologic Products and Services
  • Examples of ensemble graphical products
  NWS-NDFD High-Resolution Gridded Water                     NWS-NDFD High-Resolution Geospatial
     Resources Product Suite (WRPS)                          Water Resources Product Suite (WRPS)

 A   River Flood Outlook Products              B   Products for hydrometeorological inputs and
                                                               hydrologic forecasts

                                                                                Flood Stage

                                                                                                      Least Likely

                                                                                                      Most Likely

                                                                                                 _   Median Fcst

                                                                                                 x    Observed

11/02/2006                                                                                                       7
       Hydrologic Ensemble Prediction System

  • Currently under development and implementation

                           Atmospheric Ensemble

         Ensemble                                      Parametric
            Data            Hydrology and Water
                             Resources Models          Ensemble
         Assimilator                                   Processor

                           Hydrologic Ensemble

                       Hydrology and Water Resources
                        Ensemble Product Generator

11/02/2006                                                          8
                  ESP System Products
   From meteorological ensembles to probabilistic flood maps

  flood mapping

11/02/2006                                                     9
        Ensemble Streamflow Prediction (ESP)
  Long-term ESP for all RFCs & short- to medium term ESP for
    test basins
     + APRFC
     (Alaska)              Short- to medium-term ESP test basins

11/02/2006                                                         10
             ESP System Under Improvement

       Climatology & meteorological forecasts

                                Correct bias, account for meteorological
     ESP Pre-Processor

       Ensemble traces of precipitation, temperature

                                   Account for uncertainties in initial
      Hydrologic model              conditions & model parameters
                                         Data assimilator, parametric uncertainty
       Ensemble traces of streamflow         processor under development

                             Correct bias, account for residual hydrological
     ESP Post-Processor                        uncertainty

        Final ensemble traces of     Reflect meteorological & hydrological
               streamflow                        uncertainties
11/02/2006                                                                          11
             ESP: Ensemble Pre-Processor

 • Current ensemble forecasts from NWP have significant biases
   in the mean and in the spread
 • Single-value forecasts (HPC/RFC deterministic, atmospheric
   model ensemble mean) have additional skill
 • Goal of Ensemble Pre-Processor:
    – remove biases from single-value forecasts
    – capture skill and uncertainty in single-value forecasts
    – preserve space-time properties of hydrometeorological
       variables (Schaake Shuffle method)
    – account for temporal scale-dependency of
       hydrometerological variables

11/02/2006                                                   12
                  ESP: Ensemble Pre-Processor
  • Current Ensemble Pre-Processor:
     – Short-range ensembles conditioned on HPC/RFC deterministic
     – Medium-range ensembles conditioned on ensemble means from
       frozen version of NCEP Global Forecast System (GFS)
     – Climate adjustment of historical ensembles based on CPC outlook
       (i.e. probability shift) products
     – Climatology distribution re-sampling to better estimate true
       climatological distribution

   Ensemble Pre-Processor

       RFC QPF           RFC          Ensembles out to
        & QTF          Subsystem          Day 5
   GFS Ensemble          GFS          Ensembles out to
                                                         Blender    ensembles
      Means            Subsystem          Day 14
                                                                   out to a year

    CPC outlook         Statistical    Ensembles out
     forecast          Adjustment        to a year

11/02/2006                                                                     13
                   ESP: Ensemble Pre-Processor
     • Envisioned Ensemble Pre-Processor:
        – Integrate other ensemble forecasts: SREF, etc.
        – Integrate medium-term ensemble forecasts from frozen version of
          NCEP Global Forecast System (GFS)
        – Integrate additional climate forecasts: CFS forecasts
   Ensemble Pre-Processor
       HPC/RFC           RFC
       forecasts       Subsystem      Multi-model
                                    Ensembles out to
     SREF, others           ?           Day 5

         GFS             GFS
                                      Multi-model                   Blended
       forecasts       Subsystem
                                    Ensembles out to   Blender     ensembles
        Others              ?           Day 14                    out to a year

          CFS            CFS
        forecast       Subsystem      Multi-model
                                     Ensembles out
        Others              ?          to a year
11/02/2006                                                                    14
             ESP: Envisioned Multi-model
                Ensemble Forecasts
                                           Flood Stage

                                                                Least Likely

                                      GFS                       H
                                     HPC                        Most Likely
                                                           _   Median Fcst
                                                           x    Observed

             Mean of multi-   ROC
             model ensemble            Multi-model
                                       ensembles         Performance measures
                                                         for individual and multi-
                                                         model ensemble
11/02/2006                                                                      15
                                ESP: Verification
   • GFS subsystem: precipitation ensembles from RFC QPF (days 1
     & 2) and GFS single-value forecasts (days 3 to 14)
   • Forecast flows compared to 2 references:
        – observed flow (show all errors – hydrologic and input uncertainties)
        – simulated flow (show only errors from input – input uncertainty)

      Brier Score measures mean squared probability error
             Reference flow: observed                         Reference flow: simulated

                                        Perfect score: BS=0

11/02/2006    Verification of 24-hr flow from Mar. 2003 to Dec. 2004 for 5 ABRFC basins   16
                              ESP: Verification
 • RFC subsystem: precipitation ensembles from RFC QPF (days 1
   & 2) and re-sampled climatology (days 3 to 14)
 • Forecast flows compared to simulated flow (input uncertainty only)

      Reliability Diagram measures agreement between forecast probability
      and mean observed frequency
      10% Prob. threshold             50% Prob. threshold         85% Prob. threshold

                          Deviation from diagonal gives conditional bias
11/02/2006   Verification of 24-hr flow from Mar. 2003 to Dec. 2004 for 5 ABRFC basins   17
                              ESP: Verification
 • RFC subsystem: precipitation ensembles from RFC QPF (days 1
   & 2) and re-sampled climatology (days 3 to 14)
 • Forecast flows compared to simulated flow (input uncertainty only)

      Relative Operating Characteristic (ROC) measures resolution (ability of
      forecast to discriminate between events & non-events)
      10% Prob. threshold             50% Prob. threshold         85% Prob. threshold

                               Perfect scores: HR = 1 and FAR = 0
11/02/2006   Verification of 24-hr flow from Mar. 2003 to Dec. 2004 for 5 ABRFC basins   18

      • Reliable and skillful ensemble forecast inputs for
        Ensemble Streamflow Prediction (ESP) system
             – For all RFC basins
             – For all lead times from 1 hr to 1 yr

      • Weather and Climate ensemble re-forecasts (> 20 yrs)
        with “recent” models for EPP calibration and hydrologic
        forecast verification

11/02/2006                                                        19
             Leveraging NCPO/Climate Prediction
              Program for the Americas (CPPA)
              Climate Testbed

        Research                    Advanced
        Projects          CPPA      Hydrologic      Water
                          Core      Prediction    Customers
          CPPA           Project     Service
        Ensemble                     (AHPS)

               Hydro-Testbed / Hydroclimatic Component

11/02/2006                                                    20
             Collaboration: CPPA Project Areas
  • Forecast System Development:
     – Develop prototype software components that enable new procedures
       to be developed and tested in an RFC environment; facilitate HOSIP
       process so that proven science can be used in RFC operations
  • Testbed and RFC Pilot Projects:
     – Evaluate new procedures for possible future use by the RFCs; these
       activities are organized around proposed GE2/HEPEX testbed
       projects to ensure that they are open to all scientists
  • Support for OHD involvement in NLDAS project
  • Development of Supporting Data Sets:
     – Produce retrospective data sets (including forecasts, observations,
       and analyses of precipitation, temperature, and streamflow) for
       testbed activities; make them available to the scientific community via
       an anonymous ftp site maintained by OHD
  • HEPEX Support:
     – OHD leadership of the HEPEX project and OHD contributions to
       GAPP and GEWEX

11/02/2006                                                                       21
              Collaboration with NCPO PIs
  • NCEP (Mitchell, Toth et al.)
  • Clark (verification, ensemble prediction, data assimilation, pre-
  • East-Wide and West-Wide forecast system and multi-model
    applications (Wood and Lettenmeier)
  • K & A Georgakakos (INFORM, California)
  • Alternative algorithms: Clark/Hay, Werner, Cong, Princeton, NCEP,
  • Conditional uncertainty confidence: NCEP, Princeton, Washington
  • Verification statistics: Bradley, UCI, Arizona, Weber
  • Future unified NWS Ensemble Pre-Processor:
         Community Ensemble Pre-Processor (CEPP)

11/02/2006                                                              22
   • Reliable and skillful hydrometeorological ensemble
     forecasts are a prerequisite for producing reliable and
     skillful hydrologic ensemble forecasts

   • Significant science challenges:
        – Preprocessing of atmospheric and climate prediction
        – Hydrologic data assimilation to reduce uncertainty in initial
        – Modeling and accounting of parametric and structural
        – Verification of hydrometeorological and hydrologic ensembles

   • Community effort is needed toward multi-model ensemble

11/02/2006                                                                23
             Thank you

11/02/2006               24

To top