Forecasting Streamflow with the UW Hydrometeorological
Forecast System
Ed MaurerDepartment of Atmospheric Sciences, University of Washington
Pacific Northwest Weather WorkshopMarch 8, 2003
Photos from: www.metrokc.gov
UW Hydromet System and Water Resource Time Scales
Ref: A Plan for a New Science Initiativeon the Global Water Cycle, www.usgcrp.gov
Weather: floods, drainage, operations
Seasonal/Interannual: water supply planning, droughts
Climate: climate change, urbanization
Temporal and Spatial Scales of Hydrologic Variability
Ref: A Plan for a New Science Initiativeon the Global Water Cycle, www.usgcrp.gov
Accurate representation of spatial and temporal variability in:•Precipitation•Land surface hydrologyis essential for simulating hydrological response at this scale
MM5-DHSVM Streamflow Forecast System
Completely automatedIn use since WY 1998For details: Westrick, K.J., P. Storck, and C.F.
Mass, Description and Evaluation of a Hydrometeorological Forecast System for Mountainous Watersheds, Weather and Forecasting 17: 250-262, 2002.
Streamflow and Streamflow and other forecastsother forecasts
DHSVM
UW Real-time MM5
Distributed-Hydrology-Soil-Vegetation Model
Penn State/NCAR Mesoscale Model MM5
Used throughout the world for both research and operational forecasting
48-hour (and some 72-hour and longer) forecasts run twice daily at the University of Washington
High-resolution model (4-km) capable of capturing the complex orography of the region, including lee shading and windward precipitation enhancement
FOR MORE INFO...
http://www.atmos.washington.edu/mm5rt/http://www.atmos.washington.edu/mm5rt/
DHSVM land surface hydrology model
•Physically-based, distributed model
•Solves a water balance at each grid cell at each time step
•Horizontal scales typically 30m to 150m
•Designed for and extensively tested in complex terrain
Details on DHSVM at:Details on DHSVM at:http://www.hydro.washington.edu/http://www.hydro.washington.edu/
DHSVM CalibrationCalibration at 2 sites in Snohomish River
Basin• Used all available meteorological
observations (50sites), 1987-1991 • Used flow observations at two USGS
gauges:– Skykomish R. near Gold Bar– Snoqualmie R. at Carnation
Snoqualmie R. at Carnation
Peaks flows and average water balance are well simulated by DHSVM when forced by observed meteorology
UW Hydromet Domain - 2003
26 basins
~60 USGS Gauge Locations
48,896 km2
2,173,155 pixels
DHSVM @ 150 m resolution
MM5 @ 4 & 12 km
Web Site for Forecast Dissemination
•Automatically updated twice daily
•Graphic display indicates forecasted flood status
•Click through to:
‑Hydrographs‑Snow state (maps and points)
‑Point weather forecasts
Performance of Hydromet System
MM5-DHSVM
Observed
NWRFC
Sauk Snoqualmie
Using the Hydromet system for MM5 diagnosis
One exceptionally bad forecast for the Cedar R., events from January 25 to Feb 4, 2003Second peak:•Forecast:1200 cfs•Observed: 3700 cfs•Flood stages above bankfull occurred, and were not forecast
Of, course, not all forecasts were so bad…
Representative Meteorological Station – Mt. Gardner
Avg. Precipitation from 1/24 - 2/7:Observed: 1.0 mm/hSimulated: 0.7 mm/hTotal difference: ~100 mm
Precip
Average Temperature:Observed: +2.1CPredicted*: -0.1C
SWE:Observed: -50 mmPredicted*: +100 mm
MM5 biases in P and T combine to produce large underestimation in runoff
Temp SWE
Opportunity for Improving UW Hydromet Forecasts
1 – Precipitation/Temperature Bias Correction
Remove systematic biases in P, T, at land surface
2 – IMPROVE-2 Take advantage of the IMPROVE-2 experiment to examine the interplay between observation density and bias correction performance
3 – Initial State Updating Assimilation of snow and soil moisture information from an observationally constrained data set.
Use ground observations (SNOTEL sites) to adjust the basin snow stateChallenge:45-50 snow water observations for 48,000 km2 domain – low density places high dependence on interpolation assumptions
Snow State Updating with Observations
Expansion of Forecast Products
Source: Grimit and Eckel, 2003
•Probabilistic streamflow forecasts
‑Take advantage of ensemble MM5 simulations to estimate uncertainty in forecasts
•Forecasts of slope stability‑ DHSVM produces more than just streamflow
‑ Soil moistures, slopes in model provide additional forecasting capabilities
‑ Investigate landslide hazard forecasting
Probability of failure
Image courtesy of L. Bowling
Summary• UW Hydrometeorological Forecast System provides
accurate streamflow and snowpack predictions when forced with accurate meteorology and when properly initialized
• Improvements in both initialization and meteorological forecasts are ongoing, by analyzing current flood events and retrospective analysis
• The capabilities of the system are being expanded to include both probabilistic forecasts using ensembles, and to include landslide hazard evaluation
Acknowledgments
Current Streamflow Forecasts
•NWRFC provides river flow and stage forecasts at strategic points in Puget Sound region
•Use point forecasts of precipitation and temperature
•Streamflow produced by a lumped parameter hydrologic model (does not produce spatially distributed water
balance estimates) www.nwrfc.noaa.gov
Original Motivation for Developing UW Hydromet System
• Integrated modeling over a variety of spatial and temporal scales to examine:– Regionally consistent modeling of weather and land
surface hydrology, avoid site-specific calibration– Capture topographically-driven spatial variation in
precipitation, temperature, and wind fields– Produce experimental streamflow forecasts to
investigate skill in a coupled model setting
• Use hydrometeorological forecasts as a diagnostic tool for mesoscale atmospheric model