use of 4 km, 1 hr, precipitation forecasts to drive a statistical-distributed hydrologic model for...

Use of 4 km, 1 hr, Precipitation Forecasts to Drive a Statistical-Distributed Hydrologic Model for

Flash Flood Prediction

J8.4 20th Conference on Hydrology, AMS Annual Meeting 2006

Seann Reed1, Richard Fulton1, Ziya Zhang1,2, Shucai Guan1,3

Presented by John Schaake4

1Hydrology Laboratory, Office of Hydrologic DevelopmentNational Weather Service, NOAA, Silver Spring, Maryland

 2University Corporation for Atmospheric Research

 3RS Information Systems, Inc.

4Consultant to Office of Hydrologic Development, Annapolis, MD

Forecastedfrequencies

A Statistical-Distributed Model for Flash Flood Forecasting at Ungauged Locations

HistoricalReal-time

simulated historical

peaks (Qsp)

Simulated peaks distribution (Qsp) (unique for each

cell)

Archived

QPE

Initial hydro model states

StatisticalPost-processor

Distributed hydrologic model (HL-

RDHM)

Distributed hydrologic model (HL-

RDHM)

Real-time

QPE/QPF

Max forecasted

peaks

• The statistical-distributed model outputs gridded flood frequency forecasts.

• Here we express flood frequencies in terms of the return period (T) associated with the annual maximum flood. T = 1/p where p is the probability of occurrence in a given year (e.g. If p = 0.5, T = 2 years).

Why a frequency-based approach?

• Frequency grids provide a well-understood historical context for characterizing flood severity; values relate to engineering design criteria for culverts, detention ponds, etc.

• Computation of frequencies using model-based statistical distributions can inherently correct for model biases.

• A companion study has shown that the statistical-distributed approach can improve (on average) upon current, lumped model based procedures at least down to 40 km2 basin sizes.

Hydrology Laboratory Research Distributed Hydrologic Model (HL-RDHM)

• This implementation of RDHM uses:– 2 km grid cell resolution – Hourly 4 km QPE and QPF grids are resampled to 2 km

(nearest neighbor resampling)– Gridded SAC-SMA– Hillslope routing within each model cell– Cell-to-cell channel routing – Uncalibrated, a-priori parameters for Sacramento (SAC-

SMA) and channel routing models (Koren et al., 2004)

• RDHM showed good performance in the Distributed Model Inter-comparison Project (DMIP) (Reed et al, 2004)

• An operational prototype version of RDHM is running at two NWS River Forecast Centers

Multisensor Precipitation Nowcaster (MPN)

• Radar-based extrapolative nowcaster with the capability to:– Produce one-hour rainfall forecasts on a 4 km grid with a 5

minute update frequency (1 hour updates used here)– Use mosaicked WSR-88D radar data (used single radar here)– Use real time rain gauge data for radar bias adjustment (not

used here)– Optionally use:

• Storm growth/decay accounting (not used)• Progressive smoothing with forecast lead time (used)• Local (20 km) or area-averaged storm motion vectors (used local)

• MPN is currently running at HL in real-time for a 5 radar test region in the mid-Atlantic states

Study Basins

OK

AR

INX Radar

SRX Radar

N

No Short Station Name Area Period of record Time toName (km2) (hourly flow) peak (hrs)

1 SPRINGT Flint Ck at Springtown AR 36.8 6/1993-9/2004 3

2 SSILOAM Sager Ck nr W. Siloam Springs OK 48.9 9/1996-9/2004 3

3 CHRISTI Peacheater Ck at Christie OK 64.7 5/1993 - 9/2003 6

4 CAVESP Osage Ck near Cave Springs AR 89.9 4/2000-9/2004 4

5 DUTCH Baron Fork at Dutch Mills AR 105.1 10/1992-9/2004 2

6 KNSO2 Flint Ck near Kansas OK 284.9 6/1993- 9/2004 6

7 ELMSP Osage Ck near Elm Springs AR 336.7 10/1995-9/2004 78 ELDO2 Baron Fork at Eldon OK 795.1 10/1992 - 9/2004 139 ISILOAM Illinois R. South of Siloam Springs AR 1489.2 7/1995 - 9/2004 1710 TALO2 Illinois R. near Tahlequah OK 2483.7 6/1993-9/2004 37

Interior,Flash floodbasins

Basins are well covered by either the INX or SRX radar

• Forecast scenarios

– Zero QPF : Assumes zero future rainfall

– Persistence (Eulerian): Assumes the last hour of rain is repeated in future 1 hr period with no advection, followed by zero rain beyond 1 hour

– QPF: Uses 1 hour QPF grids from MPN followed by zero rainfall beyond 1 hour

• We have run experiments for 9 events. Example results are presented on the next several slides.

Forecast Experiments

Rainfall (mm)

QPE 7/3/2004 5-6 UTC QPF 7/3/2004 5-6 UTC

QPE 7/3/2004 6-7 UTC QPF 7/3/2004 6-7 UTC

N

Rainfall (mm)

QPE 7/3/2004 5-6 UTC QPF 7/3/2004 5-6 UTC

QPE 7/3/2004 6-7 UTC QPF 7/3/2004 6-7 UTC

N

Example MPN 1 hr Forecast Grids Compared to Multisensor QPE(similar patterns observed with some smoothing)

Observed Forecast for same hour

14 UTC 15 UTC

16 UTC 17 UTC

Example Forecasted Frequencies Available at 4 Times on 1/4/1998

In these examples, frequencies are derived from routed flows, demonstrating the capability to forecast floods in locations downstream of where the rainfall occurred.

Example Forecast Grid and Corresponding Forecast Hydrographs for 1/4/1998 15z

Eldon (795 km2)

Dutch (105 km2)

Implicitstatistical adjustment

0

200

400

600

800

1000

1/4/98 0:00 1/4/98 12:00 1/5/98 0:00 1/5/98 12:00 1/6/98 0:00 1/6/98 12:00 1/7/98 0:00 1/7/98 12:00

Date

Flow

(CM

S)

0

10

20

30

40

50

Simulated flow

Observed flow

QPF - 1/4/1998 3:00:00 PM UTC

Adjusted fcst peak

Fcst Time

0

100

200

300

400

500

600

1/4/98 0:00 1/4/98 12:00 1/5/98 0:00 1/5/98 12:00 1/6/98 0:00 1/6/98 12:00

Date

Flow

(CM

S)

0

20

40

60

80

100

Simulated flow

Observed flow

QPF - 1/4/1998 3:00:00 PM UTC

Adjusted fcst peak

Forecast time

~11 hr lead time

~1 hr lead time

Method to Calculate “Adjusted” PeaksAt Validation Points

• Probability matching was used to compute adjusted flows at validation points.

• For implementation we can only assume a similar implicit correction if we are considering frequency-based flood thresholds at ungauged locations.

DUTCH

0

0.2

0.4

0.6

0.8

1

1 10 100 1000

Flow (cms)

Pro

b of

Occ

urre

nce

SimulatedObserved

157 cms(simulated)

247 cms(adjusted)

• Lead times are computed relative to the simulated peak time.• All results shown are for CAVESP (90 km2) and single Event (7/2004)

CAVESP

0

20

40

60

80

100

120

0 2 4 6 8 10

Lead Time (hrs)

Abs.

% P

eak

Erro

r

QPF 0 QPF Pers

Lead Time = 4 hrs

Peak errors of different forecasts relative to simulated flows as a function of lead time

Lead Time = 3 hrs

Lead Time = 2 hrs Forecast Accuracy at Different Lead Times

0

50

100

150

200

250

7/2/04 12:00 7/3/04 0:00 7/3/04 12:00 7/4/04 0:00 7/4/04 12:00

Date

Flow

(CM

S)

0

20

40

60

80

Rai

nfal

l (m

m)

Simulated flow Observed flow 2 year floodFcst Time QPF - 7/3/2004 6 UTC QPF0 - 7/3/2004 6 UTCPers - 7/3/2004 6 UTC Precipitation

0

20

40

60

80

100

120

140

160

180

200

7/2/0412:00

7/2/0416:48

7/2/0421:36

7/3/042:24

7/3/047:12

7/3/0412:00

7/3/0416:48

7/3/0421:36

7/4/042:24

7/4/047:12

7/4/0412:00

Date

Flow

(CM

S)

0

20

40

60

80

Rai

nfal

l (m

m)

Simulated flow Observed flow2 year flood Fcst TimeQPF - 7/3/2004 5:00:00 AM UTC QPF0 - 7/3/2004 5:00:00 AM UTCPers - 7/3/2004 5:00:00 AM UTC Precipitation

0

20

40

60

80

100

120

140

160

180

200

7/2/0412:00

7/2/0416:48

7/2/0421:36

7/3/042:24

7/3/047:12

7/3/0412:00

7/3/0416:48

7/3/0421:36

7/4/042:24

7/4/047:12

7/4/0412:00

Date

Flow

(CM

S)

0

20

40

60

80

Rai

nfal

l (m

m)

Simulated flow Observed flow2 year flood Fcst TimeQPF - 7/3/2004 4:00:00 AM UTC QPF0 - 7/3/2004 4:00:00 AM UTCPers - 7/3/2004 4:00:00 AM UTC Precipitation

CAVESP

0

20

40

60

80

100

120

0 2 4 6 8 10

Lead Time (hrs)

Abs.

% P

eak

Erro

r

QPF 0 QPF Pers

CAVESP

0

20

40

60

80

100

120

0 2 4 6

Lead Time (hrs)

Abs.

% P

eak

Erro

r

QPF 0 QPF Pers

CAVESP

0

20

40

60

80

100

120

0 2 4 6 8 10

Lead Time (hrs)

Abs.

% P

eak

Erro

r

QPF 0 QPF Pers

CAVESP

0

20

40

60

80

100

120

0 2 4 6 8 10

Lead Time (hrs)

Abs.

% P

eak

Erro

r

QPF 0 QPF Pers

1/4/19987/3/2004

6/21/20006/17/2000

Example Results for One Basin and Several Events

• Lead time is a function of both basin response and QPF

• QPF outperforms 0 QPF in these cases.

• Persistence performance varies; can be better or worse than either 0 QPF or QPF.

Conclusions• Gridded output examples highlight two intrinsic

benefits of the distributed hydrologic model – high resolution information and cell-to-cell routing.

• Initial examination of basin-event peak error vs. lead time cases (only a few examples shown here) indicate that: – Use of 1 hour MPN-based QPF consistently

improves upon assuming Zero QPF (MPN QPF rarely introduces more error than the Zero QPF case)

– Use of 1 hour Eulerian persistence often improves upon Zero QPF, but also shows worse results in a number of situations. Persistence can produce better or worse results than QPF.