Water Supply Forecast using the Ensemble Streamflow Prediction
Model
Kevin Berghoff, Senior HydrologistNorthwest River Forecast Center
Portland, OR
Overview
Community Hydrologic Prediction System (CHPS)
3 Components to model• Calibration• Operational Forecast System• Ensemble Streamflow Prediction - ESP
Analysis
Historical Data
Calibration System (CS)
Real-Time Observed and Forecast Data
Operational Forecast System (OFS)
Ensemble Streamflow Prediction (ESP) System
Hydrologicand
Hydraulic Models
Analysis and Data Assimilation
Hydrologic andHydraulic Models
short term forecasts
current states
StatisticalAnalyses
Probabilistic
Short term to
Extended
time
Analysis
window
InteractiveAdjustments
flow
Community Hydrologic Prediction System - CHPS
Hydrologic and
Hydraulic Models
HEFS
CHPS Model Calibration
Historical precipitation and temperature used to generate Mean Areal Precipitation (MAP) and Mean Areal Temperature (MAT) for each basin
SAC-SMA and SNOW-17 model parameters are adjusted to match the simulated river flow to the observed flow data over the entire calibration period of record
Timing and attenuation of routed flows from upstream points
Precipitation
and
Temperature
Rain or
Snow
Accumulated
Snow Cover
(SWE)
Energy Exchange
At
Snow-Air
Interface
Areal Extent
of
Snow Cover
Snow Cover
Heat deficit
Liquid Water
Storage
Rain
plus
Melt
Deficit = 0
(to Soil Moisture Model)
Snow
Outflow
Snow-17 Model
Transmission
of
Excess Water
Ground
Melt
Snow Model
Soil Moisture/Runoff
Consumptive Use
River Routing
Reservoir Regulation
Flow and Stage
Forecasts
Sacramento Soil Moisture Accounting (SAC-SMA) Model
CHPS Model Calibration
Operational Forecast System
Observed and 10 Day Forecast Inputs Precip, Temperature
Operational Forecast System
ObservedDeterministic
Forecast
Ensemble Streamflow Prediction
ESP Trace Ensemble Plot
1966
1992
ESP Accumulated Flow VolumeApril – Sept Forecast Period
ESP Example: NF John Day at Monument
Median Forecast
622 KAF
101%
Initial model states: 01/18/2011
Analysis Period: 4/1/2011 – 9/30/2011
Each point represents possible outcome based on initial model states,
10 Day fx, historical precip and temperature scenarios
Example: Dillon Reservoir 2011 Forecast
Spaghetti Forecast
April – July forecast
Median forecast:
195 kaf
Calibration(parameter uncertainty)
©The COMET Program
Output of streamflow ensembles (cumulative
uncertainty)
Meteorological Input(precip and temp variability)
Current model states
User/forecaster(level of experience, personal bias)
Computer model(model structure uncertainty)
Real-time data(variability and uncertainty)
Sources of Uncertainty
Data Issues
ESP Uncertainty
ESP Uncertainty
Observed Streamflow
Simulated Streamflow
Summary
ESP Assumption: Past meteorological data is a reasonable representation of future scenarios
Ensemble of forecasts generated using past precip and temperature data, current model states (soil moisture, snowpack)
Flexibility – allows user to specify desired forecast period and statistical analysis
Allows users to incorporate probabilistic information into operational decisions
ESP forecasts useful when strong climatological signal present
ESP Cautions
Less than 30 day lead time (NWRFC specific) Unaccounted for sources of uncertainty Tend to under forecast high years, over forecast
low years
Questions?
Data Issues
ESP Uncertainty
Dec 20 Dec 27 Jan 3 Jan 11 Jan 190
100
200
300
400
500
600
700
800
900
NF John Day River at MonumentInitial Conditions/QPF effects
2011 Apr - Sep Volume Forecast
2010/2011 Forecast Date
Ap
r -
Sep
Vo
lum
e -
KA
F
ESP Sensitivity Study:Summer/Fall Soil Moisture
hours
days/weeks
months
seasonal
annual
Forecast time
Unc
erta
inty
Advances in Time Scales
Forecast Services provided for all time domains
NWRFC Forecast ProductsESP Concept
Time
Fl ow
Deterministic Probabilistic
1959
1989
1954
ESP Trace Ensemble Plot
1966
ESP VerificationDworschak Dam
ESP VerificationHungry Horse Dam
Recent Historical Perspective
JAN - JUL Observed Runoff Volume - MAF
0
20
40
60
80
100
120
140
160
180
Year
Ru
no
ff V
olu
me
- M
AF
Columbia R at Grand Coulee Columbia R at the Dalles Snake R at Lower Granite
Dashed line = 30 Yr Mean (1971-2000)
96.6 – 90%
63.6 - 101%
25.5 – 85%
Calibration(parameter uncertainty)
©The COMET Program
Output of “Raw” streamflow ensembles (cumulative uncertainty)
Meteorological Input(precip and temp variability)
Current model states (spatial and temporal scale dependent bias)
User/forecaster(level of experience, personal bias)
Computer model(model structure uncertainty)
Real-time data(variability and uncertainty)
ESP uncertainty
This slide from Kevin Werner