1. Development of an Ensemble Gridded Hydrometeorological Forcing Dataset over the Contiguous United States Andrew J. Newman 1, Martyn P. Clark 1, Jason Craig 1, Bart Nijssen 2, Andrew Wood 1, Ethan Gutmann 1, Naoki Mizukami 1, Levi Brekke 3, and Jeff R. Arnold 4 1 National Center for Atmospheric Research, Boulder CO, USA 2 University of Washington, Seattle WA, USA 3 U.S. Department of Interior, Bureau of Reclamation, Denver CO, USA 4 U. S. Army Corps of Engineers, Institute for Water Resources, Seattle WA, USA

2. Outline Motivation Methodology Example output Validation ▫Comparisons to deterministic ▫Statistical properties Summary

3. Opportunities for hydrologic prediction Hydrological Prediction: How well can we estimate the amount of water stored? Accuracy in precipitation estimates Fidelity of hydro model simulations Effectiveness of hydrologic data assimilation methods Water Cycle (from NASA) Meteorological predictability: How well can we forecast the weather and climate? Opportunity: Characterize historical forcing uncertainty -> improve water stored estimates Current uncertainties for land-surface modeling are generally order of magnitude Can we begin to quantify space-time uncertainty? hydrological predictability meteorological predictability MotivationMethodologyInput DataSummaryValidationExample Output

4. Precipitation Data Uncertainty MotivationMethodologyInput DataSummaryValidationExample Output Point to grid uncertainties: ▫Measurement errors  Gauge placement issues (e.g. roof)  Is a gauge working properly? ▫Measurement representativeness  Sub-pixel variability  Take closest measurement?  Interpolate nearby gauges? ▫Missing Data  Interpolate available obs?  Fill missing data?

5. Step 1: Locally weighted regression at each grid cell: Probability of Precipitation (PoP) via logistic regression, then amount and uncertainty (least squares mean & residuals) observations Example over the Colorado Headwaters Ensemble Generation Clark & Slater (2006), Newman et al. (2014, in prep) MotivationMethodologyInput DataSummaryValidationExample Output Example over the Colorado Headwaters

6. Step 2: Synthesize ensembles from PoP, amount & uncertainty using spatially correlated random fields (SCRFs) Example over the Colorado Headwaters observations Clark & Slater (2006), Newman et al. (2014, in prep) Ensemble Generation MotivationMethodologyInput DataSummaryValidationExample Output Other Methodological choices: Topographic lapse rates derived at each grid cell for each day vs. climatology (e.g. PRISM) Used serially complete (filled) station data rather than only available obs vs. using only available observations Included SNOTEL observations vs. excluding Final Product: 12 km, daily 1980-2012, 100 members, precipitation & temperature (1.5 TB)

7. 12,000+ stations with serially complete data (interpolation & CDF matching) Precipitation, temperature or both Development of Serially Complete Station Dataset MotivationMethodologyInput DataSummaryValidationExample Output

8. GHCN-Daily(Cooperative networks, Automated (ASOS/AWOS)) in US, Canada and Mexico (thanks to Kyoko Ikeda) Use QC from National Climatic Data Center (NCDC) SNOwpack TELemetry (SNOTEL) observations from National Resource Conservation Service (NRCS) have no QC Applied QC from Serreze et al. (1999) Removes gross temperature and precipitation errors Also corrects for gauge under-catch using large accumulation events -> Precip to SWE ratio Development of Serially Complete Station Dataset MotivationMethodologyInput DataSummaryValidationExample Output

9. Variable station data availability Stations come and go, have missing data Use all stations with > 10 years of data and > 10 years of overlap with ≥ 30 stations within 500 km (generally following Eischeid et al. 2000) Interpolate temperature gaps of 1-2 days Use CDF matching from best available rank correlated station to fill larger temperature gaps and all precipitation missing data Assumes stationary CDF in time Development of Serially Complete Station Dataset MotivationMethodologyInput DataSummaryValidationExample Output

10. MotivationMethodologyInput DataSummaryValidationExample Output Central US Flood of 1993 June 1993 total precipitation

11. Validation: Comparisons to other datasets (temperature) Minimal differences in mean temperature Except in intermountain west Maurer uses constant elevation lapse rate (-6.5 K km -1 ) Ensemble lapse rate derived from station data (including Snotel) Maurer shown to be too cold in higher elevations (e.g. Mizukami et al. 2014) Ensemble warmer than Maurer in high terrain MotivationMethodologyInput DataSummaryValidationExample Output

12. Validation: Comparisons to other datasets (PoP) Maurer et al. (2002) Interpolation between observations increases precip occurrence Ensemble: SCRFs applied to PoP field more realistic PoP Generally reduces PoP Data differences may be responsible for PoP increases MotivationMethodologyInput DataSummaryValidationExample Output

13. Validation: Comparisons to other datasets (amount) Ensemble - Maurer Maurer et al. (2002) NLDAS - Maurer Xia et al. (2012) Daymet - Maurer Thornton et al. (2014) Version 2 (regridded to 12km) MotivationMethodologyInput DataSummaryValidationExample Output

14. Validation: Comparisons to other datasets (amount) CONUS precip difference PDFs NLDAS & Maurer agree very closely (both use PRISM correction, similar input gauges) Daymet wettest Slightly larger spread in ensemble – most distinctly different MotivationMethodologyInput DataSummaryValidationExample Output

15. Validation: Discrimination & Reliability Reliability (blue & black lines) Want to be close to 1-1 line (predicted = observed probabilities) Generally good reliability MotivationMethodologyInput DataSummaryValidationExample Output >0 mm >13 mm >25 mm>50 mm >0 mm>13 mm >25 mm>50 mm

16. Validation: Discrimination & Reliability Discrimination (red & black lines) Want to maximize separation of PDFs Generally good separation MotivationMethodologyInput DataSummaryValidationExample Output >0 mm>13 mm >25 mm>50 mm >0 mm>13 mm >25 mm>50 mm

17. Example Application Snowmelt dominated basin in Colorado Rockies Example water year daily temperature (a) Snow water equivalent accumulation (b) Simple temperature index model (optimized for Daymet (green)) MotivationMethodologyInput DataSummaryValidationExample Output

18. Summary Developed ensemble hydrometeorolgical dataset Required development of serially complete station dataset Both will be available for download at: http://ral.ucar.edu/projects/hap/flowpredict/subpages/pqpe.php Methodology follows Clark and Slater (2006) but for CONUS Ensemble has: Realistic probability of precipitation (PoP) Estimate of forcing uncertainty Quantification of uncertainty for data assimilation Potential Applications: Propagation of forcing uncertainty into model states Useful for state uncertainty estimation (e.g. SWE) Ensemble calibrations can allow for estimates of parameter uncertainty (impact of noise on parameter estimation) MotivationMethodologyInput DataSummaryValidationExample Output

19. Next Steps Include downward shortwave and humidity in ensemble Improved downward shortwave radiation product (Slater et al., in prep) Stations across US at many elevations, climatic conditions Will allow for improved temperature/moisture – shortwave relationships Include ensemble into data assimilation system(s) Particle filter and ensemble Kalman filter Can we improve daily to seasonal streamflow prediction using automated systems? MotivationMethodologyInput DataSummaryValidationExample Output

20. Validation: Comparisons to other datasets (amount) Precip difference PDFs Nearly symmetric differences, except vs Daymet (left panel) NLDAS & Maurer agree very closely (both use PRISM correction) (right panel) Slightly larger spread in ensemble – most distinctly different MotivationMethodologyInput DataSummaryValidationExample Output