NWSRFSEnsemble Streamflow Prediction

Manual Calibration Program (MCP) Operational Forecast System (OFS)Ensemble Streamflow Prediction (ESP)

MCP

OFS ESP

NWS Workshop on Hydrologic ForecastingPrague Campus

Czech University of AgricultureJune 20-24, 2005

Ensemble Streamflow Prediction

NWSRFS models and current states.Historical MAPs and MATs from calibration.Flexible analysis window.Many forecast variables.Better performance under extreme conditions.Use of weather and climate forecasts.

ESP Uses

Long range seasonal water supply.Spring snowmelt volume forecasts.Spring snowmelt peaks.Minimum flows for navigation, irrigation, environmental, recreation, etc.Experimental Short Term EnsemblesHours to days

ESP Flexibility

Time WindowDays, Weeks, Months, Seasons

VariablesVolumeMean DischargePeak FlowLow FlowDays to Peak, Low, or Specified Rate

Key Model StatesInitial Conditions Are Important

Snow ModelLiquid and frozen water equivalentHeat contentAreal extent

Soils ModelUpper zone moisture content (tension/free)Lower zone moisture content (tension/free)Frozen ground

Accurate Snowpack States

Real time network consistent with calibration network.Quality control temperature and precipitation observations.Update snow water equivalent with snow course observations.

ESP Process

Trace GenerationStatistical Analysis

NWSRFSModel States

AnalysisWindow

ProbablisticForecast

NWSRFSModel

Parameters

(Pre-adjusted)Historical

Temp/Precip

Ensemble Streamflow Prediction

ClimateForecast

Adjustments

Daily RFCForecasting•Data Ingest•Data QC•Model Updating

Current Conditions•Soil•Reservoir Levels•Streamflow

HistoricalTime Series

All Years ofRecord

ForecastTime

Series

Mean ArealTime Series

PrecipitationTemperature

NWSRFSHydrologic

Models

Time St

ream

flow

April-July

Making an Ensemble Forecast Using ESP

-> Future Time

Past StagesSoil/SnowStates

19711972197319741975

Blend QPF/QTF

7172737475

Historical Precip/Temps for Past YearsCreates a Flow for Each Year

Flow Traces

Past <-Past Future Time - >

Defining Your Time For the Forecast

Past Future Time - >

Flow

Window

Make a frequency distribution using each ensemble value in the window…and then fit a probability function.

Frequency Diagram (PDF)

Cumulative Frequency Diagram (CDF)

100 90 80 70 60 50 40 30 20 10 0Exceedance Probability

Flood Stage 18000

6%

8690

8600

8350

8110

8040

8040

8040

8040

7780

7600

7420

7380

7190

7190

7130

6970

6930

6870

6750

6350

6240

6200

6100

5960

5590

5300

5250

5150

5140

4710

4680

4570

4110

4010

4010

3100

2990

2410

21500

19300

17400

17400

15200

14600

13700

13500

13300

13200

12900

11600

11100

10400

10400

10100

9640

9560

9310

8850

Raw Data(Ensemble Members)

Window

Elementary Probability Concept

Climate Variability In ESP

Pre -Adjustment Technique Weight/Modify on Input Side

7172737475

Post -Adjustment TechniqueWeight On Output Side

ESP Use of Weather and ClimateForecasts

HistoricalMAT and MAP

AdjustmentSystem

Adjusted HistoricalMAP and MAT

Weather Forecasts Climate Forecasts

ESP Product Generation

Generate conditional traces with ESPSelect product attributes and generate tables and graphics with ESP Analysis and Display Program (ESPADP).

ESPADP - ESP Analysis and Display Program

ESPADP Options

ESP Trace Ensembles

Mean Daily Flows

Mean Weekly Flows

ESPADP Web Interface“Allows Customers to Build Their Own Products”

Suggestions for Using ESP

Data quality control is a high priority.Be aware of biases or limitations in the model calibration.Avoid large changes to model states that cannot be explained by input data errors.Remember that ESP does not use runtime MODs.

Future use of ESP

Better use of weather/climate forecasts.Implementation of error models.Development of short-term techniques.Development of regulated forecasts.

Reservoir and diversion impacts.Interactive use by customers/partners.

Seamless Suite of MAT EnsemblesFrom Three Sources/Models After 'Shuffling'

-35

-30

-25

-20

-15

-10

-5

0

5

10

6-hour MATs

MA

T (C

)

ENS 1ENS 2ENS 3ENS 4ENS 5ENS 6ENS 7ENS 8ENS 9

Unconnected MAT EnsemblesFrom Three Sources/Models Before Connecting

-45

-40

-35

-30

-25

-20

-15

-10

-5

0

5

10

6-hour MATs

ENS 1ENS 2ENS 3ENS 4ENS 5ENS 6ENS 7ENS 8ENS 9

SHORT TERM DAY 1-3 HAS

MEDIUM RANGE DAY 4-14GFS(MRF)

LONG RANGE DAY 15 – 365CPC PRE-ADJUSTMENT/ OTHER

Example of Creating a Seamless Suite of MATs

Short Term5-day Precipitation Ensembles

QPF = 1.5 inches in two successive 6-hr periods, otherwise zero

Short Term 5-day Streamflow Ensembles

Short Term5-day Streamflow Forecast Distribution

Medium Range Forecasts

MRF is colder than normal in this case.

Input into ESP

Hourly instantaneous flow ensembles are created by ESP and saved. MRF shows higher flows than historical when it is warmer (during the first week). These may be converted into probabilistic forecasts…

An example of the skill in producingstreamflow runoff from using temperaturand precipitation downscaled fromthe MRF vs historicalprecipitation and temperature (ESP).

It shows by using temperatures from thedownscaled MRF in lieu of historical information that streamflow forecastscan be improved.

The End

Ensemble Streamflow Prediction Component

(ESP)