You are on page 1of 24

# REGIONAL FLOOD METHODS

## Australian Rainfall and Runoff

A guide to flow estimation

OVERVIEW

Project team RFFA methods in ARR 1987 Pilot study results for Victoria Status reports for other states

## A guide to flow estimation

PROJECT TEAM
Dr Ataur Rahman Mr Khaled Haddad Associate Professor James Ball Mr Erwin Weinmann Professor George Kuczera Mr Mark Babister Dr William Weeks Dr David Kemp WA Representatives

## Probabilistic Rational Method

Vic & Eastern NSW Frequency factors depend on elevation for six zones A-F for Eastern NSW.

## Multiple regression relationship

Western NSW

Quantile Regression
ACT
Australian Rainfall and Runoff
A guide to flow estimation

## METHODS IN AR&R 1987

Qld
No general method meets the primary criterion of being based on sufficient observed flood data. Main Roads Department Rational Method not easy to apply. Department of Primary Industries Rational Method - based on little/no observed streamflow records.
Australian Rainfall and Runoff
A guide to flow estimation

## SA - South East Region

Probabilistic Rational Method (Based on observed data from 20 catchments). Department of Agriculture Method - Index type approach (based on floods data from 6 small agricultural catchments).

## SA Northern and Western Region (Based on very little observed data)

Rational Method Bankfull frequency method Method of western NSW by McDermott and Pilgrim (1983). Regional Flood Frequency Method Based on records of 3 catchments.

## METHODS IN AR&R 1987

WA
Five regions. Rational method & Index flood method. Methods were based on data from 12-28 catchments.

NT
Methods were based on limited data. Rational method.
Australian Rainfall and Runoff
A guide to flow estimation

## METHODS IN AR&R 1987

Western Tasmania
Regional flood frequency method.

Eastern Tasmania
No design information based on observed data is available. Turners (1961) procedure is suggested.

METHODS TESTED

## Probabilistic Rational Method Quantile Regression Technique

Ordinary least squares (OLS); and Generalised least squares (GLS).

## CY is the central component of PRM

QY = 0.278 CY I t ,Y A
c

Limitations: Location based interpolation for CY Limited independent testing. Limited data on prediction uncertainty.
Australian Rainfall and Runoff
A guide to flow estimation

## PROBABILISTIC RATIONAL METHOD

Easy to apply Simple regional approach with meaningful independent variables CY integrates effects of other flood generation and attenuation factors in the equation.

## A guide to flow estimation

QUANTILE REGRESSION

## QRT can be expressed as:

QT = aB bC c D d
B, C, D, are variables (catchment area, main stream slope, rainfall intensity) and a, b, c, are regression coefficients to be estimated by:
Australian Rainfall and Runoff
A guide to flow estimation

## PRM was developed

FFY = CY/C10 C10 contour was mapped using MAPINFO

QRT - OLS method was developed using SPSS QRT - GLS method was developed using Statistical Package R. Independent testing was undertaken.
Australian Rainfall and Runoff
A guide to flow estimation

INTRODUCTION

Typical problems in streamflow data collation include: Unrepresentative sites; Missing records; and Outliers, trends and rating curve extrapolation errors.

## SELECTION OF CANDIDATE STATIONS

Initially 415 candidate catchments selected from Victoria based on the following criteria:
1. 2. 3. 4. 5. 6. Catchment Area (up to 1000 km2) Record Length (initially 10 years was cut-off) Regulation Urbanisation Landuse change Quality of data
Australian Rainfall and Runoff
A guide to flow estimation

LOCATIONS (Vic)

## INFILLING MISSING RECORDS

Method 1: Comparison of monthly instantaneous maximum (IM) data with monthly maximum mean daily (MMD) data for the year with data gap

## INFILLING MISSING RECORDS

Method 2: Regression of annual maximum mean daily flow series with annual instantaneous maximum series of the same station

14000 12000

IMM (ML/d)

Y Linear (Y)

MMD (ML/d)

4000

6000

8000

## INFILLING MISSING RECORDS

273 data points from 187 stations were filled by Method 1 60 data points from 44 stations were filled by Method 2 10% of stations did not have any missing records. Only 7.5 % of data points were infilled.

## A guide to flow estimation

10

TREND ANALYSIS

Two tests applied, Mann Kendall & Distribution Free CUSUM Test Mann Kendall test - checks for trends in time series; and CUSUM test - examines whether the mean values in two parts of a record are different.

## A guide to flow estimation

TREND ANALYSIS

Mann Kendall test showed 21% of stations have decreasing trend after 1990s
Vk - Station 230210 9 8 7 6 5 4 3 2 1 0
1970 1975 1980 1985
Year Vk

Vk

## Significant shift downwards, after the 1990s.

1990 1995 2000 2005
A guide to flow estimation

## Australian Rainfall and Runoff

11

TREND ANALYSIS It was also useful to plot time series data as below.
IMM - Station 230210
12000 10000

Flow (ML/d)

1995

2000

2005

Year

## This shows 10-15 years of significant downward trend

Australian Rainfall and Runoff
A guide to flow estimation

TREND ANALYSIS

Conclusions developed were Inclusion of sites with short records questionable Cut-off length introduced record 25 years minimum

A significant shift from ARR1987 (where minimum record length of 10 years was considered).
Australian Rainfall and Runoff
A guide to flow estimation

12

## RATING CURVE ISSUES

QE = annual maximum flood series data point (reported value) QM= maximum measured flow
Rating Ratio ( RR ) = QE QM

RR 1 - considered to be free of rating curve extrapolation error. a RR value >> 1 indicates potential rating curve error
Australian Rainfall and Runoff
A guide to flow estimation

## Graph shows typical example

Likely Rating Curve Error - 222202
6 5 4

## Data points subject to possible rating curve errors

QE/QM

3 2 1 0 0 10 20 Data Point 30 40 50

13

## RATING CURVE ISSUES

A rating ratio of 20 may be adopted? Any station with RR>20 would be excluded.
Histogram of Rating Ratio Values
10000
4387
Frequency

1000

Frequency

384 111

## 90% of rating ratios lie between 1 & 20

61 19 18 18

100

Rating Ratio - RR
9 10 10 4 5 4 2 1 1 1 4 2 5 3 2 1 2 0 0

10

1 2 4 6 8 10 12 14 16 18 20 22 24 26 28 30 32 34 36 38 40 42 44 46 48 50 More

Rating Ratio

## AT-SITE FLOOD FREQUENCY

A LP3 distribution was fitted to each stations annual maximum flood data Bayesian parameter fitting procedure was adopted. FLIKE was used.

14

## Locations of the study 133 catchments

Australian Rainfall and Runoff
A guide to flow estimation

CATCHMENT AREAS
60 50 40 Frequency 30 20 10 0 0 to 50 51 to 100 101 to 200 201 to 500 501 to 750 Catchment area (km2) 751 to 1000

## Range: 3 997 km2 Mean = 319 km2

12% catchments < 50 km2, 20% catchments < 100 km2 and 80% catchments < 500 km2.
Australian Rainfall and Runoff
A guide to flow estimation

15

STREAMFLOW RECORDS

## A guide to flow estimation

CATCHMENT CHARACTERISTICS
Variable area forest Evap Rain S1085 sden I QSA Description Catchment area (km2) Fraction of catchment area under forest Mean annual areal potential evapotranspiration Mean annual rainfall Slope of central 75% of main stream Stream density Design rainfall intensity of 2 and 50 years ARI and 1 and 12 hours duration Fraction quaternary sediment area

16

## C10 REGRESSION EQUATION

C10 = 2.631 0.125 log(rain) + 1.02 log(evap) + 0.072 log(S1085) + 0.128 log(sden)
rain = mean annual rainfall (mm) evap = mean annual evaporation (mm) S1085 = Slope of main stream (m/km) sden = stream density (km/km2)

17

## TESTING Q10 PREDICTION

50 45 40 35 30 25 20 15 10 5 0 T14 3 T8 18 T16 24 T10 36 T9 52

Q (m3/s)

Q10_PRM (C10 from contour map) Q10_PRM (C10 from regression) 90% CI-LL Q10_FFA 90% CI-UL

## TESTING Q50 PREDICTION

80 70 60 Q (m3/s) 50 40 30 20 10 0 T14 3 T8 18 T16 24 T10 36 T9 52 90% CI-UL Q50_PRM (C10 from contour map) Q50_PRM (C10 from regression) 90% CI-LL Q50_FFA

18

RELATIVE ERRORS

10

20

50

100

28

17

26

30

35

38

31

19

23

36

41

44

19

## FLOOD FREQUENCY PLOT

Station 227210
100000

10000

Discharge(ML/day)

1000

100

Gauged flow

10 1.25 2

10

20

50

100

200

## QRT PREDICTION EQUATIONS

log( Q ) = b 0 + b1 log( area ) + b 2 log( rain ) + b 3 log( sden ) for T = 2 years log( Q ) = b 0 + b1 log( area ) + b 2 log( I 2 _ 12 ) + b 3 log( sden ) for T = 5 years log( Q ) = b 0 + b1 log( area ) + b 2 log( I 2 _ 12 ) + b 3 log( rain ) + b 4 log( sden ) for T = 10 , 20 , 50 , 100 years

## Only 4 variables are required:

area in all equations sden in all equations I2_12 for T = 5, 10, 20, 50 and 100 years rain for T = 2, 10, 20, 50 and 100 years.
Australian Rainfall and Runoff
A guide to flow estimation

20

## TESTING Q10 PREDICTION

50 45 40 35 Q (m3/s) 30 25 20 15 10 5 0 T14 3 T8 18 T16 24 T10 36 T9 52 Q10_OLS Q10_GLS 90% CI-LL Q10_FFA 90% CI-UL

## TESTING Q50 PREDICTION

100 90 80 70 Q (m3/s) 60 50 40 30 20 10 0 T14 3 T8 18 T16 24 T10 36 T9 52 Q50_OLS Q50_GLS 90% CI-LL Q50_FFA 90% CI-UL

21

RELATIVE ERRORS

22

## SUMMARY PILOT STUDY

The updated PRM can provide design flood estimates with reasonable accuracy. C10 values can be obtained from
C10 contour map or developed prediction equation (provides less biased results).

## SUMMARY PILOT STUDY

Median relative error values for the new PRM ranged from 17% to 44%. For 7% cases relative error values > 100%. The PRM tends to underestimate flows compared to flood frequency estimates, with 60% of the cases being underestimated.

23

## SUMMARY PILOT STUDY

Prediction equations have been developed for Victoria using OLS and GLS methods. Little differences in prediction equations between OLS and GLS. GLS procedure provides better estimation reliability.

24