PUB in practice: case studies - Runoff Prediction in Ungauged Basins: Synthesis across Processes, Places and Scales - page 295

Figure 11.25. Range of catchment
characteristics for the study stream
gauges.

800

160

700 140

600 120

500 100

400 80

Explanation

300

60 Upper limit

(75th percentile + 1.5 *(75th percentile - 25th percentile))

200

40

100 20 75th percentile

0 0

Median

600 100

25th percentile

500

80

400

Lower limit
(25th percentile + 1.5 *(75th percentile - 25th percentile))
Values above the upper limit or below the lower limit

60

300

40

200

20

100

0 0

To estimate daily runoff at ungauged locations, this study
utilised the methods described in Archfield and Vogel
( 2010 ) and Archfield et al.( 2010 ), which are summarised
here (for a contextual description see Section 10.3.2 ). First,
a continuous daily flow duration curve was constructed by
regional regression equations obtained by relating the phys-
ical and climate characteristics of 48 gauged basins ( Figure
11.24 , black triangles) to selected flow duration curve stat-
istics (see Section 7.3.1 ). To ensure the estimated flow
duration curves were representative of all flow conditions,
only stream gauges with long records (greater than 20 years)
that contained the drought-of-record were used to develop
the regional flow duration curve regression equations.
A continuous flow duration curve was then interpolated
between the selected runoff statistics to obtain a continuous,
daily flow duration curve. A reference stream gauge is then
used to translate the flow duration curve into a hydrograph.
Instead of using the nearest reference stream gauge, the
study applied the map correlation method (see Archfield
and Vogel, 2010 ). The map correlation method provides a
geostatistical approach to select the reference stream gauge
whose runoff time series is estimated to be most correlated
with the ungauged location. Archfield and Vogel ( 2010 )
show that, for methods that transfer the timing of runoff
from one basin to another, correlation between runoff can be
an effective method to identify the reference stream gauge
when compared to the selection of the nearest reference
stream gauge. The full set of 66 study stream gauges ( Figure
11.24 ) were used as possible reference stream gauges and
included in the development of the map-correlation method.

Therefore, daily runoff was estimated from 1 September
1960 through 30 August 2004 for an ungauged location.
To validate the method, 18 stream gauges with observed
runoff records for this period ( Figure 11.24 ) were used in a
leave-one-out cross-validation procedure. One-by-one,
each stream gauge was removed and the location treated
as ungauged.
Observed and estimated runoff were then compared at
each of the removed stream gauges, and a NSE value and
root mean square error (RMSE) values were computed for
each of the 18 stream gauges using the natural-log values
of the observed and estimated daily runoff. Unregulated
daily runoff was able to be reliably estimated for ungauged
locations across Massachusetts, with NSE values ranging
from 0.98 to 0.69, with a median value of 0.86
( Figure 11.26 ); RMSE values ranged from 19% to 284%,
of the mean with a median value of 55% ( Figure 11.26 ).
A comparison of observed and estimated hydrographs for
stream gauges with the best (Hubbard River near West
Harland, CT (HUBB)) and worst (Burlington Brook near
Burlington, CT (BURL)) agreement over the period 1
October 1960, through 30 September 1962

the period
of time at the start of the worst drought of record

-

show
good agreement in both real and log space ( Figure 11.27 ;
Archfield et al., 2010 ).

-

Discussion
By providing technically defensible estimates of runoff
time series in ungauged basins, stakeholders in the state
of Massachusetts have now agreed upon the methods and
decision-support tool as the framework for negotiation
between environmental groups, water managers and
water suppliers. The methods and goodness-of-fit results
were disseminated to stakeholders in several formats.
The map correlation method represented a new approach

Results
Stakeholders in the study area required estimated daily
runoff that extended back to the drought-of-record and
through to present-day (2004) runoff conditions.