Downscaling and Probability Applications in Climate Decision Aids May 11, 2011 Glenn Higgins Manager, Environmental Sciences and Engineering Department
Downscaling and Probability in Climate Modeling The Challenge: Operationalize Climate Decision Aids Accessible User-oriented Products Timely, Repeatable Process Specific to Scenario, Region, Locality, Decision Time-frames Characterized for Accuracy and Confidence Transparent Pedigree and Provenance Key Issues to Address IPCC Global Climate Models (GCM) coarse resolution Many diverse scenarios and outcomes represented Translation from Science to Useful Engineering Products Solutions: Downscaling using Regional Climate Models (RCM) Analysis of Probabilities to Estimate Confidence in Results Tailored Decision Products with Uncertainty
Climate Modeling and Decision Aids Global and Regional Climate Models Mean Annual Temperature at Washington, DC Global scale Regional scale Users Decision Aids Regional Climate Models Global Climate Models Global Observing System Provide consistent, higher resolution forecasts for regions Provide coarse Resolution Climate Forecasts at Global Scales Provides Data Needed to Run and Validate Climate Models 3
The Process starts with Global Climate Data and Ends with Actionable Products Actionable decision aid Mean "Oppressive" Days Per Year (Models have monthly biases with respect to GHCN removed) Change in Deaths due to Change in Oppressive Days Additional City Metro Pop Current Future Change Deaths per Million Deaths for Metro Richmond 1.2 M 17.47 47.22 29.8 26.78 32 Lynchburg 246 K 11.91 36.56 24.7 22.19 5 Roanoke 296 K 10.69 34.16 23.5 39.90 12 Wash. DC 5.3 M 16.31 35.56 19.3 17.33 92 Norfolk 1.8 M 13.28 38.31 25.0 22.53 40 'Change' is Future value - Current value Observations, Best Science, and Physical Models Users Decision Aids Regional Climate Models Global Climate Models Global Observing System Empirical and explicit decision domain modeling 4 Raw Global-scale model results Regional Downscaling Captures local effects and better physics High resolution regional model results
Regional climate modeling process Systems approach Model Initialization (initial and boundary conditions) IPCC AR4 or reanalysis data Dynamical Downscaling Regional Climate Model (WRF) Middle East Model Validation Southwest US 5 Global Historical Climatology Network (GHCN) OSU PRISM Gridded Observational Dataset WRF output Intercomparison and Bias Correction Uncertainty Reduction & Characterization Quantitative Climate Decision Aids Actionable Decision Aids with Confidence Intervals
Modeling Approach and choices Current Future Historical Reference Run Current Period Free Run Future Period Free Run (2000-2009) NCEP Reanalysis 2.5x2.5 deg (2001-2009) ECHAM5 IPCC 20 th century run1 1.9x1.9 deg (2030-2039) ECHAM5 IPCC A1B scenario 1.9x1.9 deg 12 km 36 km 108 km Dynamical downscaling Running WRF in "climate mode" Grid nudging Deep soil T update SST update Variable CO 2 WRF 3.1.1 domain setup LW radiation CAM SW radiation CAM PBL Surface layer YSU Monin-Obukhov Keep the modelled fields close to the input data Allow for a seasonal variation in the deeplayer soil temperature Diurnal variation of the skin SST; dynamic subskin SST variation Ensures more accurate representation of the radiative budget Land surface Noah LSM Microphysics Cumulus WSM 5-class Kain-Fritsch WRF output Downscaled climate decision aids 6 Model physics configuration Modeling experiments
Model validation Comparison of Model Outputs to Known High-Quality Climatologies Monthly mean temperature (2001-2009) OSU-Prism WRF-NCEP WRF-ECHAM5 C OSU-Prism Annual mean precipitation (2001-2009) WRF-NCEP WRF-ECHAM5 in. WRF-NCEP downscaled temperature and precipitation agrees well with the PRISM data, WRF-ECHAM shows a cold and wet bias. 7
Uncertainty Management Schema Ensembles and Statistical Bias Correction Inaccuracy Model Fails to Represent Totality of the Earth System Some Climate Modeling Uncertainty Sources Longer Period Runs and Perturbation Ensembles Limited Realizations Lengths of Runs Insufficient to Exactly Characterize Model Outcomes Time-Series and Extreme Events Analysis Inter-Annual Variability Natural Long-Period cycles in Weather Confound Specification of Climatological Statistics Uncertainty Reduction and Characterization Uncertainty Management Reduction: Establish Historical Reference Run Apply Best Trusted Climatology (PRISM) Perform Statistical Bias Corrections Characterization: Assess Distributional Forms Identify UMVUEs of expectations and standard errors of fitted parameters January Average Daily Temperature Change January Standard Error of Daily Temperature Change 8 Climate Change Signature Standard Error of Climate Change Signature
Uncertainty reduction through bias correction to reference run permits model inter-comparison Step 1 Step 2 Step 3 Step 4 Parameterize 2 nd order autoregressive processes representing reference, control, and future datasets Operate inverse transforms on each to recover underlying autoregressive deviates Apply reference period process to current dataset Apply reference period process to future dataset Means Standard Deviations Lagged Correlations Location, monthly and time-of-day Undoes Generating function Deviates are data-events Subtle frequency space analysis is enabled Empirically forced to match reference run s statistics Reference inverse + control forward = model bias Removes model bias signature from future runs Preserves climate change signature Establishes inter-comparability Annual mean precipitation (2001-2009) OSU-Prism WRF-ECHAM5 WRF-ECHAM5 bias corrected in. 9
Towards Decision Products-- Degree-Day Energy Products Derived from Downscaled Data Computed 65F Baseline HDDs with Standard Error Estimates Jan ECHAM/WRF 2000-2009 HDD Jan ECHAM/WRF Future Current HDD
From Heating/Cooling Degree Products to Energy Decision Aids We developed an energy usage regression model that predicts changes in energy use (electricity and natural gas) due solely from climate change signatures from uncertainty reduced model outputs. Determine parameterized empirical models of energy consumption from degree-days data output Use multivariate regression to fit coefficients for time trends and sensitivity Automatically identify and report outlier data points using crosscorrelation techniques Propagate climate model uncertainties through regression equations to combine model and regression fitting errors into a single composite error bound Monthly average cooling degree days 500 450 current 400 future 350 300 250 200 150 100 50 0 1200 1000 800 600 400 200 0 Monthly average heating degree days current future Annual Energy use Electricity [GWhr] Natural Gas [Millions CF] Annual Energy use Electricity [GWhr] Current 2000-2009 Peterson AFB Future 2030-2039 87.72 88.25 268.23 255.19 Current 2000-2009 Fort Carson Future 2030-2039 134.65 135.20 Natural Gas [Millions CF] 1192.16 1134.21 11 Monthly Mean Heat/Cooling Degree Data Traditional Energy Demand Products
Conclusion Operationalized Decision Aids Products Require: Regional Climate Modeling Probabilistic & Statistical Treatment of Uncertainty User Tailored End-Products Uncertainty Treatment though to End-Products Results: Regional and Local Specificity Confidence in Products for Application Actionable Products
Uncertainty Knowledge Framework Uncertainty is not error, but error contributes to uncertainty Many definitions and interpretations Taxonomy of uncertainty (Tannert el al. 2007 EMBO Rep. 8 (10): 892 6) Uncertainty ontology W3C Incubator Group Report 31 March 2008 Operationalization implies defining uncertainty as a limitation on product utility From a user s perspective, uncertainty is thus a measure of the confidence which may be placed in a product