Suggestions for research to fill critical capability gaps to support short-term water management decisions Martyn Clark, David Gochis, Ethan Gutmann, and Roy Rasmussen NCAR Research Applications Laboratory

Outline The myriad of uncertainties in hydrologic monitoring and prediction products Critical capability gaps – and research that is needed to fill them Summary of research suggestions 2

Barriers to prediction hydrological uncertainty meteorological uncertainty Water Cycle (from NASA) Hydrological uncertainty: How well can we estimate the amount of water stored in the snow and soil? Accuracy of precipitation estimates Fidelity of hydro model simulations Effectiveness of hydrologic data assimilation methods Meteorological uncertainty: How well can we forecast the weather?

Historical Simulation Q SWE SM Historical Data PastFuture SNOW-17 / SAC 1.Run hydrologic model up to the start of the forecast period to estimate basin initial conditions; Typical streamflow forecasting method…

Historical Simulation Q SWE SM Historical DataForecasts PastFuture SNOW-17 / SAC 1.Run hydrologic model up to the start of the forecast period to estimate basin initial conditions; 2.Run hydrologic model into the future, using an ensemble of local-scale weather and climate forecasts. Typical streamflow forecasting method…

Historical Simulation Q SWE SM Historical DataForecasts PastFuture SNOW-17 / SAC Forecast Uncertainties

Historical Simulation Q SWE SM Historical DataForecasts PastFuture SNOW-17 / SAC Uncertainties in model inputs UncertaintiesMethod(s)Impact Model inputsProbabilistic quantitative precipitation estimationInitial Conditions Hydrological Model Perturb model states Multiple parameter sets Multiple model structures Initial conditions Forecasts Weather/Climate ForecastsEnsemble forecasts (the seamless suite)Forecasts

Historical Simulation Q SWE SM Historical DataForecasts PastFuture SNOW-17 / SAC Uncertainties in the hydrological model UncertaintiesMethod(s)Impact Model inputsProbabilistic quantitative precipitation estimationInitial Conditions Hydrological Model Perturb model states Multiple parameter sets Multiple model structures Initial conditions Forecasts Weather/Climate ForecastsEnsemble forecasts (the seamless suite)Forecasts

Historical Simulation Q SWE SM Historical DataForecasts PastFuture SNOW-17 / SAC Uncertainties in initial conditions UncertaintiesMethod(s)Impact Model inputsProbabilistic quantitative precipitation estimationInitial Conditions Hydrological Model Perturb model states Multiple parameter sets Multiple model structures Initial conditions Forecasts Weather/Climate ForecastsEnsemble forecasts (the seamless suite)Forecasts DA TO REDUCE UNCERTAINTIES

Historical Simulation Q SWE SM Historical DataForecasts PastFuture SNOW-17 / SAC Uncertainties in weather/climate forecasts UncertaintiesMethod(s)Impact Model inputsProbabilistic quantitative precipitation estimationInitial Conditions Hydrological Model Perturb model states Multiple parameter sets Multiple model structures Initial conditions Forecasts Weather/Climate ForecastsEnsemble forecasts (the seamless suite)Forecasts

Historical Simulation Q SWE SM Historical DataForecasts PastFuture SNOW-17 / SAC The whole enchilada UncertaintiesMethod(s)Impact Model inputsProbabilistic quantitative precipitation estimationInitial Conditions Hydrological Model Perturb model states Multiple parameter sets Multiple model structures Initial conditions Forecasts Weather/Climate ForecastsEnsemble forecasts (the seamless suite)Forecasts

Local-scale hydrologic analyses Hydrologic data assimilation Meteorological Analyses ensemble QPE, etc. Hydrologic model Integrated hydro-LSM (from WRF-Hydro) Local-scale probabilistic meteorological forecasts Forecast blending methods Hydrologic model Integrated Hydrologic model—LSM (from WRF-Hydro) Short-term forecasts merged radar & high-res NWP Local-scale probabilistic hydrologic forecasts Medium-range forecasts global NWP models (from outside NCAR) Probabilistic downscaling Seasonal forecasts statistical and dynamical (from outside NCAR) Conditioned weather generators Statistical post-processing methods

Monitoring and forecast products reviewed in the ST-doc 13 Also: NRCS NWCC water supply forecasts Monitoring products: USGS stream gauging NRCS SNOTEL NRCS snowcourse NWS COOP observer RFC Precipitation analysis

Outline The myriad of uncertainties in hydrologic monitoring and prediction products Critical capability gaps – and research that is needed to fill them Summary of research suggestions 14

Historical Simulation Q SWE SM Historical DataForecasts PastFuture SNOW-17 / SAC Uncertainties in model inputs UncertaintiesMethod(s)Impact Model inputsProbabilistic quantitative precipitation estimationInitial Conditions Hydrological Model Perturb model states Multiple parameter sets Multiple model structures Initial conditions Forecasts Weather/Climate ForecastsEnsemble forecasts (the seamless suite)Forecasts

Clark & Slater, 2006 – JHM Step 1: Estimate precipitation CDF at each grid cell Step 2: Synthesize ensembles from the CDF Uncertainties in model inputs corresponding observations Case study over the Colorado Headwaters Gap: CONUS-domain ensemble hyper-resolution forcing data (Precip, Temp, RH, wind, SW, LW) Use all available data networks, remote sensing products and NWP reanalyses Ensure consistency among variables and realistic space-time variability

Historical Simulation Q SWE SM Historical DataForecasts PastFuture SNOW-17 / SAC Uncertainties in the hydrological model UncertaintiesMethod(s)Impact Model inputsProbabilistic quantitative precipitation estimationInitial Conditions Hydrological Model Perturb model states Multiple parameter sets Multiple model structures Initial conditions Forecasts Weather/Climate ForecastsEnsemble forecasts (the seamless suite)Forecasts

The quest for physically realistic streamflow simulations Current assessments: DMIP –A well-calibrated conceptual model can perform just as well as a calibrated physics-based model (when evaluating streamflow at the calibration point) –Kirchner’s “mathematical marionettes”? Contemplating stationarity… –Parameters in conceptual models are often assigned unrealistic values to compensate for structural weaknesses –Conceptual models subject to the same stationarity predicament that plagues statistical streamflow forecasting systems –Treat the symptom or the disease? Can develop increasingly elegant methods for data assimilation and statistical post-processing, but does not address the root cause of forecast errors –(Go as far as you can with physics and do the rest with statistics) 18

Multi-scale WRF-Hydro Driver/Coupler Atmospheric (Re)analyses Global Earth System Models HRU-based 2-d surface and sub-surface flow Lumped Regional Weather/ Climate models process complexity horizontal complexity Model library: Multi-model Multi-physics Multi-scale Research under the WRF-Hydro umbrella: Improve simulations of hydrology from local-continental scales Gap: Research needs similar to those required to support long-term planning Improved “fit-for-purpose” hydrologic simulations and analyses using multi-model / multi-physics / multi- scale hydrologic modeling capabilities Improved estimation of hydrologic model parameters from local- continental scales Improved estimates of hydrologic uncertainty Gap: Research needs similar to those required to support long-term planning Improved “fit-for-purpose” hydrologic simulations and analyses using multi-model / multi-physics / multi- scale hydrologic modeling capabilities Improved estimation of hydrologic model parameters from local- continental scales Improved estimates of hydrologic uncertainty

Historical Simulation Q SWE SM Historical DataForecasts PastFuture SNOW-17 / SAC Uncertainties in initial conditions UncertaintiesMethod(s)Impact Model inputsProbabilistic quantitative precipitation estimationInitial Conditions Hydrological Model Perturb model states Multiple parameter sets Multiple model structures Initial conditions Forecasts Weather/Climate ForecastsEnsemble forecasts (the seamless suite)Forecasts DA TO REDUCE UNCERTAINTIES

EnKF Sample Results: snow data assimilation Interpolated SWE Mean & Std. Dev Model Truth Slater & Clark, JHM 2006 Data withholding experiments at 53 stations in Colorado -- stations represent high-resolution model grid cells without any data

EnKF Sample Results: streamflow assimilation Use of the ensemble Kalman filter to use mis-match between observed and simulated streamflow to update hydrologic states (Clark et al., AdWR 2008) with DA no DA Barnett’s Bank Gap: Integrated hydrologic data assimilation system Multiple ground-based and remotely sensed observations Evaluate different assimilation strategies (esp. the role of the particle filter)

Historical Simulation Q SWE SM Historical DataForecasts PastFuture SNOW-17 / SAC Uncertainties in weather/climate forecasts UncertaintiesMethod(s)Impact Model inputsProbabilistic quantitative precipitation estimationInitial Conditions Hydrological Model Perturb model states Multiple parameter sets Multiple model structures Initial conditions Forecasts Weather/Climate ForecastsEnsemble forecasts (the seamless suite)Forecasts

Different models for different forecast lead times –0-6 hours  Radar extrapolation –6-72 hours  High-resolution regional NWP model –3-14 days  Global-scale NWP model –Seasonal  Disaggregated seasonal climate outlooks Requirements –An ensemble of sub-daily weather sequences –Preserve inter-site correlations, temporal persistence, and correlations between variables –Minimize abrupt changes when a new model is introduced Methods: The “Schaake Shuffle” (Clark et al., 2004) –Estimate CDF at a given forecast lead time (using the most appropriate forecasting method at that lead time) –Sample ensemble members from the CDF in a way that preserves observed space-time correlation structure (e.g., use rank of observed weather patterns to resample). Integrated set of forecast inputs from minutes to seasons

Example Results: Cle Elum River Basin (Central Washington) Forecasts based on NWP model output Forecasts based on historical data Clark and Hay (2004) – Journal of Hydrometeorology

Hourly instantaneous flow ensembles are created by ESP and saved. MRF shows higher flows than historical when it is warmer (during the first week). These may be converted into probabilistic forecasts… Operational Implementation: NWSRFS Werner et al. (2005) – Journal of Hydrometeorology

Improvements in forecast skill is most pronounced during the rising limb of the hydrograph. (20-year hindcast) Operational Implementation: Probabilistic skill Ranked Probability Skill Score (RPSS) Werner et al. (2005) – Journal of Hydrometeorology

New products 28 Gap: Integrate best-available forecast products to construct conditioned weather sequences suitable for hydrologic models Post-processing methods to improve statistical reliability of model ensembles Stochastic methods to reproduce observed space-time variability in local- scale climate Seamlessly merge forecasts of different type and resolution.

Outline The myriad of uncertainties in hydrologic monitoring and prediction products Critical capability gaps – and research that is needed to fill them Summary of research suggestions 29

Summary of research suggestions CONUS-domain ensemble hyper-resolution forcing data (Precip, Temp, RH, wind, SW, LW)  Use all available data networks, remote sensing products and NWP reanalyses  Ensure consistency among variables and realistic space-time variability Hydrologic modeling  Improved “fit-for-purpose” hydrologic simulations and analyses using multi-model / multi- physics / multi-scale hydrologic modeling capabilities  Improved estimation of hydrologic model parameters from local-continental scales  Improved estimates of hydrologic uncertainty Integrated hydrologic data assimilation system  Multiple ground-based and remotely sensed observations  Evaluate different assimilation strategies (esp. the role of the particle filter) Integrate best-available forecast products to construct conditioned weather sequences at forecast lead times from minutes to seasons, suitable for hydrologic models  Post-processing methods to improve statistical reliability of model ensembles  Stochastic methods to reproduce observed space-time variability in local-scale climate  Seamlessly merge forecasts of different type and resolution. 30

Reclamation-USACE-NCAR Streamflow predictability project Assess performance of current hydrologic models used by the NWS, and assess dependence of model performance on  Physical characteristics of the basins (climate, vegetation, soils, topography)  Reliability of quantitative precipitation estimates (e.g., station density, radar) Assess the relative importance of hydrologic and meteorological/ climatological information in determining forecast skill Conduct research to improve estimates of uncertainty  During model spin-up  During the forecast period Conduct research to reduce forecast uncertainty  Better hydrologic models  Better weather forecasts and climate outlooks  Adoption of hydrologic data assimilation methods and statistical post- processing methods Examine impact of different sources of uncertainty in water management decisions