A Regionalized Flow Duration Curve Method to Predict Streamflow for Ungauged Basins: a

Case Study of the Rappahannock Watershed in Virginia, USA

A Regionalized Flow Duration Curve Method to Predict Streamflow for Ungauged Basins: a

Case Study of the Rappahannock Watershed in Virginia, USA

Yusuf Mohamoud and Rajbir ParmarUS EPA – Ecosystems Research Division

Athens, Georgia

Yusuf Mohamoud and Rajbir ParmarUS EPA – Ecosystems Research Division

Athens, Georgia

88thth IAHS Scientific Assembly and 37IAHS Scientific Assembly and 37thth

IAH Congress IAH Congress Hyderabad, IndiaHyderabad, India

Outline Outline

Flow duration curve: historical backgroundFlow duration curve: historical backgroundDevelopment of regionalized flow duration Development of regionalized flow duration curve (RFDC)curve (RFDC)Separation of flow magnitude and time Separation of flow magnitude and time sequence: new paradigm sequence: new paradigm Modeling streamflow for Modeling streamflow for ungaugedungauged sites sites using regionalized flow duration curve using regionalized flow duration curve (RFDC) and HSPF(RFDC) and HSPFComparison of RFDC and HSPF predictive Comparison of RFDC and HSPF predictive performanceperformance

Flow duration curve applicationsFlow duration curve applicationsExtending short period data records and Extending short period data records and filling missing data points filling missing data points

Predicting flow and water quality time series data Predicting flow and water quality time series data for for ungaugedungauged sites sites

Forecasting flow and water quality time seriesForecasting flow and water quality time series

FDC can be used for FDC can be used for baseflowbaseflow separationseparation

FDC can be used for calibrating rainfallFDC can be used for calibrating rainfall--runoff runoff models particularly for models particularly for ungaugedungauged basins basins

Effect of geology on low flows Effect of geology on low flows (Searcy, 1963)(Searcy, 1963)

1

32

54

6

Information content in flow duration curves (after Information content in flow duration curves (after Searcy, 1963)Searcy, 1963)

Prediction of flow duration curve (FDC) and Prediction of flow duration curve (FDC) and

streamflowstreamflow

Step 1. Develop regional regression equations (QStep 1. Develop regional regression equations (Q.1.1 to Qto Q9999) for ) for watersheds in the Appalachian, Ridge and Valley, and watersheds in the Appalachian, Ridge and Valley, and Piedmont physiographic provinces Piedmont physiographic provinces Step 2. Predict FDC for gauged and Step 2. Predict FDC for gauged and ungaugedungauged sites of the Midsites of the Mid--Atlantic regionAtlantic regionStep 3. Convert FDC to streamflow time series dataStep 3. Convert FDC to streamflow time series dataStep 4. Test FDC methodStep 4. Test FDC method’’s predictive performance s predictive performance

Step 5. Compare FDC and HSPFStep 5. Compare FDC and HSPF

Regionalization approaches: MidRegionalization approaches: Mid--Atlantic RegionAtlantic Region

A new paradigm: flow, duration curve, percentile flowsA new paradigm: flow, duration curve, percentile flows

Map of the study watershed showing gauged and ungauged

sites

FDC prediction and streamflow conversion toolFDC prediction and streamflow conversion tool

Soils, climate, geology, geomorphology, land use (watershed descriptors)

R2 = 0.99n = 45

0

50

100

150

200

250

300

350

0 100 200 300 400

RFDC predicted percentile flows (l/sec/km 2)

Obs

erve

d pe

rcen

tile

flow

s (l/

sec/

km2 ) R2 = 0.92

n = 45

0

50

100

150

200

250

300

350

400

0 50 100 150 200 250 300

HSPF predicted percentile flows (l/sec/km2)

Obs

erve

d pe

rcen

tile

flow

s (l/

sec/

km2 )

Comparisons of observed percentile flows and percentile flows predicted by HSPF and RFDC methods For Site 9, 13, and 22 (FDC predicts only the magnitude component of streamflow)

HSPF

RFDC

Rapdian River GaugeR2 = 0.8033

0

20

40

60

80

100

120

0 20 40 60 80HSPF Simulated Monthly Streamflow

(m3/sec)

Obs

erve

d M

onth

ly

Stre

amflo

w (m

3 /sec

)

Robinson River GaugeR2 = 0.8138

0

10

20

30

40

50

0 10 20 30 40 50HSPF Simulated Monthly Streamflow

(m3/sec)

Obs

erve

d M

onth

ly

Stre

amflo

w (m

3 /sec

)Robinson River GaugeR2 = 0.9956

020

4060

80100

0 20 40 60 80RFDC Simulated Monthly Streamflow

(m 3/sec)

Obs

erve

d M

onth

ly

Stre

amflo

w (m

3 /sec

)Rapidan River Gauge

R2 = 0.9555

0

10

20

30

40

50

0 10 20 30 40 50RFDC Simulated Monthly Streamflow

(m 3/sec)

Obs

erve

d M

onth

ly

Stre

amflo

w (m

3 /sec

)

Ruckersville River GaugeR2 = 0.9449

05

1015202530

0 10 20 30RFDC Simulated Monthly Streamflow

(m3/sec)

Obs

erve

d M

onth

ly

Stre

amflo

w (m

3 /sec

) Ruckersville River GaugeR2 = 0.7783

0

5

10

15

20

25

30

0 10 20 30

HSPF Simulated Monthly Streamflow (m3/sec)

Obs

erve

d M

onth

ly

Stre

amflo

w (m

3 /sec

)

Comparison of observed mean monthly streamflow vs. mean monthlystreamflow simulated by RFDC and HSPF for three gauged sites

Comparisons of observed hydrograph and hydrographs predicted by Comparisons of observed hydrograph and hydrographs predicted by HSPF and RFDC methods for the Robinson Site (Site 13)HSPF and RFDC methods for the Robinson Site (Site 13)

0

10

20

30

40

50

60

1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45

Time (days)

Stre

amflo

w (m

3 /sec

)

RFDC Predicted (Rapidan Time Sequence)ObservedRFDC Predicted (Robinson Time Sequence)HSPF Predicted (No Correction)HSPF_Predicted (Magnitude Correction)HSPF Predicted (Time Sequence Correction)

`

0

5

10

15

20

25

30

35

40

45

50

1 3 5 7 9 11 13 15 17 19 21 23 25 27 29 31 33 35 37 39 41 43 45

Time (days)

Stre

amflo

w (m

3 /sec

)

RFDC Predicted (Rapidan Time Sequence)ObservedRFDC Predicted (Ruckersville Time Sequence)HSPF Predicted (No Correction)HSPF Prediced (Magnitude Correction)HSPF Predcited (Time Sequence Correction)

Comparison of observed hydrographs and HSPF and RFDCpredicted hydrographs for the Ruckersville evaluation sites (Site 22)

Comparisons of observed streamflow and streamflow predicted Comparisons of observed streamflow and streamflow predicted byby HSPF for the period between (01/01/1980 through 12/31/1990)HSPF for the period between (01/01/1980 through 12/31/1990)

Destination Site Source Site Model Calibration _____R2 N-S Daily (monthly) Daily (monthly)

Site 9-Rapidan River Site 9 0.65 (0.80) 0.65 (0.76)

Model Evaluation

Site 13-Robinson River Site 9 0.68 (0.66) 0.66 (0.55)

Site 22-Ruckersville Site 9 0.68 (0.64) 0.78 (0.74)

Sequence adjusted-HSPFSite 13- Robinson Site 13 0.92 (0.97) 0.89 (0.91)Site 22-Ruckersville Site 22 0.95 (0.99) 0.93 (0.92)

_______________________________________________________________________

Comparisons of observed streamflow and streamflow predicted by Comparisons of observed streamflow and streamflow predicted by RFDC for the period between (01/01/1980 through 12/31/1990)RFDC for the period between (01/01/1980 through 12/31/1990)

Destination Site Source Site Model Calibration _____________R2 N-S Daily (monthly) Daily (monthly)

Site 9-Rapidan River Site 9 0.96 (1.0) 0.92 (0.95)

Model Evaluation

Site 13-Robinson River Site 9 0.93 (0.98) 0.93 (0.95)

Site 22-Ruckersville Site 9 0.80 (0.95) 0.76 (0.91)

Sequence adjusted-HSPFSite 13- Robinson Site 13 0.93 (0.99) 0.93 (0.95)Site 22-Ruckersville Site 22 0.95 (0.99) 0.94 (0.97)____________________________________________________________________

HSPF Simulated versus RFDC Simulated Streamflow R2 N-S RMSE

Site Daily (monthly) Daily (monthly) (m3/sec/kmSite 25 0.59 (0.78) 0.53 (0.65) 0.017Site 24 0.58 (0.78) 0.51 (0.64) 0.017Site 23 0.53 (0.75) 0.52 (0.68) 0.022Site 22 0.56 (0.77) 0.49 (0.68) 0.017Site 21 0.59 (0.78) 0.51 (0.64) 0.017Site 20 0.57 (0.78) 0.54 (0.65) 0.018Site 19 0.57 (0.77) 0.47 (0.68) 0.019Site 18 0.57 (0.78) 0.53 (0.65) 0.018Site 17 0.58 (0.78) 0.54 (0.66) 0.017Site 16 0.58 (0.78) 0.52 (0.65) 0.017Site 15 0.59 (0.78) 0.52 (0.66) 0.017Site 14 0.58 (0.78) 0.49 (0.65) 0.017Site 13 0.57 (0.77) 0.50 (0.71) 0.016Site 12 0.59 (0.78) 0.47 (0.73) 0.017Site 11 0.59 (0.78) 0.51 (0.72) 0.016Site 10 0.58 (0.78) 0.52 (0.72) 0.016Site 09 0.57 (0.78) 0.51 (0.72) 0.016Site 08 0.58 (0.78) 0.47 (0.72) 0.017Site 07 0.58 (0.78) 0.47 (0.71) 0.017Site 06 0.58 (0.78) 0.48 (0.71) 0.017Site 05 0.56 (0.78) 0.50 (0.65) 0.017Site 04 0.57 (0.78) 0.47 (0.70) 0.017Site 03 0.53 (0.78) 0.34 (0.61) 0.017Site 02 0.58 (0.78) 0.45 (0.65) 0.018Site 01 0.51 (0.78) 0.31 (0.64) 0.017

Comparison of HSPF and RFDC simulated streamflow for 22 ungauged sites and 3 gauged sites of the Rapidan Watershed

Future ResearchFuture Research

Separation of streamflow magnitude and sequence componentsSeparation of streamflow magnitude and sequence componentsWhich variables and parameters are related to magnitude ?Which variables and parameters are related to magnitude ?

Which variables and parameters are related to time sequence?Which variables and parameters are related to time sequence?

Can streamflow prediction be improved through improved magnitudeCan streamflow prediction be improved through improved magnitudeprediction?prediction?

Can streamflow prediction be improved through improved sequence Can streamflow prediction be improved through improved sequence prediction?prediction?

Extend the RFDC method to predicting nutrient, sediment, and patExtend the RFDC method to predicting nutrient, sediment, and pathogen hogen concentration and load duration curvesconcentration and load duration curves

Other Applications: Prediction of total suspended solids using tOther Applications: Prediction of total suspended solids using the RFDC he RFDC methodmethod

Rapidan River

0100200300400

500600700800900

1 26 51 76 101 126 151 176 201 226 251 276 301 326 351Day

Tota

l Sus

pend

ed S

olid

s (m

g/l) Observed

Simulated

ConclusionConclusionFDC only captures the magnitude component of streamflow (FDC FDC only captures the magnitude component of streamflow (FDC has no time sequence)has no time sequence)

Time sequence is obtained from a nearby gauged site (no magnitudTime sequence is obtained from a nearby gauged site (no magnitude e is required)is required)

RFDC method had higher predictive performance than HSPFRFDC method had higher predictive performance than HSPF

RFDC can be useful to improving rainfallRFDC can be useful to improving rainfall--runoff modelsrunoff models

Regionalization methods are suitable for FDC prediction hence flRegionalization methods are suitable for FDC prediction hence flow ow magnitudemagnitude

Predicting magnitude and time sequence components of streamflow Predicting magnitude and time sequence components of streamflow together is a major weakness of rainfalltogether is a major weakness of rainfall--runoff models (e.g., HSPF)runoff models (e.g., HSPF)

Predicting magnitude and time sequence components of streamflowPredicting magnitude and time sequence components of streamflowseparately makes RFDC highly suitable for predictions of separately makes RFDC highly suitable for predictions of ungaugedungauged basinsbasins

QuestionsQuestions