Use of Climate Projections for Water Supply Planning Alison Adams, Ph.D., P.E. NCPP Workshop August 12-16, 2013

It takes a team FSU/COAPS Vasu Misra Lydia Stefanova UF/WATER INSTITUTE Wendy Graham Syewoon Hwang TAMPA BAY WATER Alison Adams Tirusew Asefa Jeff Geurink

3 Florida’s Largest Regional Public Water Supplier Wholesale drinking water to six governments 2.4 Million Residents mgd annual average Seasonal to multi-year variable climate

Tampa Bay Water’s Climate Change Assessment Project …. In this project we are using dynamically and statistically downscaled climate model output to drive hydrologic models and explore potential impacts of climate variability and climate change on water availability and water allocation decisions 4

Water Institute Research ReanalysisGCMs_ future GCMs_ retro. Bias-correction Application for Tampa Bay region Hydrologic model (IHM) Downscaling Raw GCMs or Reanalysis Bias-corrected GCMs Observation Downscaled GCM Observation Statistical method; BCSD, SDBC, BCCA, BCSA, etc. Dynamical downscaling MM5, RSM, etc. R1, R2, ERA40, 20CRCMIP3: CCSM, GFDL, HadCM3, etc.

1.Statistical downscaling –Comparative evaluation of 4 methods (BCSD_daily, BCCA, SDBC, BCSA) Hwang and Graham (2013) Hydro. Earth Syst. Sci –Hydrologic simulation Submitting to ASABE transaction 2.Evaluation of downscaled reanalysis data R1+MM5 (Hwang et al., 2011) R2+RSM (Stefanova et al., 2011) ERA40+RSM (Stefanova et al., 2011) 20CR+RSM (DiNapoli and Misra, 2012) –Hwang et al 2013 Reg. Environ Change What we have done so far

1.Uncertainty of Bias-correction in climate change impact assessment 2.Evaluated Future Projections through hydrologic model What we have done so far

Spatial variability (Variograms)

3 GCMs + Regional Spectral Model (RSM), CCSM, HadCM3, and GFDL Spatial resolution (10kmx10km) over southeastern US Variables: hourly Prec., humidity, wind speed, etc., daily Tmax/min data –Daily bias-corrected Prec. data are available Retrospective simulation period: Future simulation (AR4 A2 scenario): Data

Hydrologic implication

Integrated Hydrologic Model TBW and SWFWMD commissioned the development and application of an integrated surface water/groundwater model for the Tampa Bay Region. The Integrated Hydrologic Model (IHM) was developed which integrates the EPA Hydrologic Simulation Program-Fortran for surface-water modeling with the US Geological Survey MODFLOW96 for groundwater modeling. Ross et al., 2004 (IHM theory manual)

Assessment of the utility of dynamically-downscaled regional reanalysis data to predict streamflow in west central Florida –Reanalysis data – robust proxy of historic atmospheric observations –Verifying accurate prediction of historic climatic and hydrologic behavior using reanalysis data is an essential first step before using retrospective and future GCM projections to predict potential hydrologic impacts of future climate change Assessment of dynamically downscaled GCM future projections Dynamical Downscaling

Study period from 1989 to R1+MM5 (Hwang et al., 2011) R2+RSM (Stefanova et al., 2011) ERA40+RSM (Stefanova et al., 2011) CR+RSM (DiNapoli and Misra, 2012) IHM calibration/verification period Bias-corrected reanalysis data for hydrologic model Wet seasonDry season

Comparison of the mean annual cycles of (a) monthly mean and (b) standard deviation of daily precipitation. Raw results monthly mean precipitation standard deviation of daily precipitation

Comparison of time series of (a) annual total precipitation and (b) standard deviation of daily precipitation over the year Raw resultsBias-corrected results

Comparison of error statistics of monthly areal precipitation predictions Raw Bias- corrected RawBias-corrected

Comparison of observed vs. simulated mean monthly streamflow Raw results Bias-corrected results

Raw results Bias-corrected results Comparison of observed vs. simulated annual time series

Comparison of error statistics of monthly streamflow simulations for each target station; (a) PBIAS, (b) RSR, (c) R 2, and (d) NSE

Bias correction removed all errors in mean daily, monthly and annual ppt Errors in daily std dev were removed but not for monthly and annual totals Raw reanalysis data has errors in time series of daily rainfall were not corrected and these errors were aggregrated into monthly and annual timeseries Daily ppt timing errors propagated and were enhanced by the non-linear streamflow processes in IHM All reanalysis data underestimated streamflow In the unconfined aquifer region, rainfall errors lead to underpredicting groundwater levels. Bias corrected reanalysis data for hydrologic model results

Ppt errors propagated and enhanced by the non-linear hydrologic processes produced low hydrologic model skill and can have significance water supply planning implications Need to reproduce more detailed ppt characteristics than daily rainfall to accurately capture hydrologic behaviors for water supply planning Conclusion: Using daily CDF mapping for bias correction is not sufficient for predicting hydrologic behavior. Improvements in RCM physics and parameterization or development of more enhanced bias correction techniques Results of bias corrected reanalysis data for hydrologic model

3 GCMs + Regional Spectral Model (RSM) –CCSM, HadCM3, and GFDL (not available yet) Spatial resolution (10kmx10km) over southeastern US Variables –hourly Prec., humidity, wind speed, roughness, etc. –daily Tmax/min data –Daily bias-corrected Prec. data are available Retrospective simulation period – Future simulation (AR4 A2 scenario) – CLAREnCE10 data

 Future Bias Correction methods: CDF mapping CDF: 1 Precipitation Sim_retro. BC_retro + Sim_future raw 2 raw1 obs Example 1 Bias-corrected Sim_future BC-Sim_future

Raw results Bias-corrected results 1. Mean daily precipitation

1 CDFm 3 CDFm_%bias 4 EDCDFm_%bias 2 EDCDFm Raw results Bias-corrected results 1. Mean daily precipitation

1 CDFm 3 CDFm_%bias Raw results Bias-corrected results 2. Std. of daily precipitation 4 EDCDFm_%bias 2 EDCDFm

1 CDFm 3 CDFm_%bias Raw results Bias-corrected results 2. Std. of daily precipitation 4 EDCDFm_%bias 2 EDCDFm

 Spatial distribution of mean temperature (map comparison)  Annual cycle of  Monthly mean Tmax and Tmin  Differences between the simulations for 1969~1999 & 2039~2069

Observation 1969~1999 Approx. +2`C = 2039~2069 CCSM HadCM3 GFDL 14.5˚C 18.5˚C

Observation 1969~1999 Approx. +3`C = Approx. +2`C = 2039~2069 CCSM HadCM3 GFDL 27˚C 30 ˚C

Raw results 1.1 Mean daily T max & T min T max T min T max T min T max T min Bias-corrected results T max T min T max T min T max T min

Raw results Bias-corrected results 1.2 Mean temperature change: 2039~2069 – 1969~1999 T max T min

 Spatial distribution of mean precipitation (map comparison)  Annual cycle of  Monthly mean precipitiation  Differences between the simulations for 1969~1999 & 2039~2069

Raw CCSM results significantly underestimate the mean precp. by 2.5mm over the region Raw HadCM3 and GFDL results overestimate by 1~2mm Based on the future scenario, precipitation may decrease or increase Observation 1969~1999 CCSM HadCM3 GFDL 3.9mm 3.3mm 1.8mm 0.8mm Way off!! underestimated 2039~2069 Even lower

Raw results 2.2 Mean daily precipitation Bias-corrected results

2.3 Mean precipitation change: 2039~2069 – 1969~1999 Raw results Bias-corrected results

 Annual ET, ET fraction (ET/Precip.)  Mean streamflow  Design flow estimations

3.1 ET estimations Annual average ET (mm/year)ET fraction (ET/Precp.) CCSMHadCM3GFDL CCSMHadCM3GFDL

Retrospective simulation results 3.2 mean streamflow (Alafia River station) Future simulations Streamflow Change (Future-retro.) CCSM HadCM3 GFDL

3.2 mean streamflow (Alafia River station) Streamflow Change (Future-retro.) Precipitation Change (Future-retro.)

Retrospective simulation results 3.3 Std. of streamflow (Alafia River station) Future simulations Streamflow Change (Future-retro.)

CCSM HadCM3GFDL CCSM HadCM3 GFDL 3.3 Design flow estimation 7Qxx high (low) flow means the average maximum (minimum) flow for seven consecutive days that has probable recurrence interval of once in xx years, respectively. 7Q2 low flow

Used 3 dynamically downscaled GCMs (i.e., CCSM, Had3CM, GFDL), 3 CDF construction strategies for CDF mapping bias-correction, and monthly delta method for future scenarios  Differences among GCM projections overwhelmed differences among bias correction techniques.  Temperature Results  All GCMs successfully reproduced spatial distribution and mean climatology of retrospective daily temperature  All consistently estimated 2-3 o C increase of mean temperature for future (2039~2069) under future A2 scenario.  Precipitation Results  Dynamically downscaled retrospective CCSM predictions are way off!  Retrospective HadCM3 and GFDL reproduce seasonal cycle of precipitation.(e.g., wet summer)  Different GCMs produced conflicting precipitation change estimates for future A2 scenarios (some higher, some lower)

 Hydrologic implications  Even with consistent increased temperature estimates, differences among future precipitation estimates propagate into significant differences in future hydrologic predictions ( i.e. ET, mean streamflow predictions, and 7Q10 estimates).  Precipitation signal overwhelms temperature signal in predicting hydrologic implications of projected future changes. Q. How many GCMs are required to get an accurate representation of range of possible future precipitation projections and thus range of possible hydrologic change? Q. Should we continue to use CCSM in our analysis?

 Consider other climate model products & GHG scenarios…  NARCCAP, CMIP5, COAPS products, etc.?  Other methodologies to downscale/bias-correct climate model results?  Statistical downscaling methods in order to increase number of GCMs considered?