Prediction of low flows in ungauged basins - Runoff Prediction in Ungauged Basins: Synthesis across Processes, Places and Scales - page 182

to choose in particular environments. The assessment is
performed at two levels (see Section 2.4.3 ). The Level 1
assessment is a meta-analysis of studies reported in the
literature. The Level 2 assessment involves a more focused
and detailed analysis of individual basins from selected
studies of Level 1 in terms of how the performance
depends on climate and catchment characteristics as well
as on the method chosen. More details are reported in the
comparative study of Salinas et al.( 2013 ). In both Level 1
and Level 2 assessments, the performance was evaluated
by leave-one-out cross-validation, where each catchment
was treated as ungauged and the runoff predictions were
then compared to the observed runoff. The performances
obtained by the comparative assessment are estimates of
the total uncertainty of

Figure 8.15. Map indicating the countries included in the Level 1
assessment. After Salinas et al.( 2013 ).

runoff predictions in these comparison with the other runoff signatures in Chapter

12 , the R² of predicting low flow indices were calculated
for all methods in all studies with the exception of tree
rings. The 25% and 75% quantiles of these R² are 0.57 and
0.78, respectively.
Figure 8.15 shows the global coverage of the studies
listed in Table A8.1 . Most of the cross-validation assess-
ments were performed in Europe and North America and
only a few studies cover Australia and Asia ( Table A8.1 ).
Most available studies were for humid and cold climates,
and fewer studies in monsoon, semi-arid, arid and tropical
climates.

ungauged basins.

8.5.1 Level 1 assessment
Table A8.1 summarises the 19 studies that were used for
the Level 1 assessment. The number of catchments in each
study ranges from 40 to 1003, with a median of 150.
Several studies compare different approaches which results
in a total of 27 results for predictive performance. The
selection of studies was guided by the overall aim of
supporting geographical coverage, but preference was
given to studies that optimally served the benchmarking
aim by providing a comparative assessment of two or more
methods on a larger data set, by using performance meas-
ures obtained by cross-validation, and by focusing on
standardised low flow indices (specific runoff, or standard-
ised low flows) to factor out the dominance of catchment
area on the low flow estimates. The regionalisation
methods used are process-based methods, geostatistics,
global regression, regional regression and estimation from
short record methods (record augmentation methods).
Three performance measures were used in the assessment:
the coefficient of determination (R²) of estimated and
observed low flow indices, the root mean square error
(RMSE), and the relative root mean square error
(RRMSE), i.e., RMSE divided by the average low flow
index over the study area ( Table 2.2 ). In cases where not all
performance measures were available, they were back-
calculated from the available data where possible (indi-
cated in Table A8.1 ). For the comparisons, only those
studies were used where R² was available or could be
back-calculated. Even if the studies listed in Table A8.1
are not completely compatible, e.g., due to different low
flow characteristics or evaluation methods, the collection
of studies does provide an indication of the performance
range to be expected for different methods of predicting
low flows in ungauged basins around the world. For

How good are the predictions in different climates?
Figure 8.16 shows that the highest performance is obtained
for humid catchments, but there are also studies in humid
climates that report a significantly lower performance. In
arid climates, the performance is never very high, but more
studies are needed to clearly show this behaviour. The
most likely reason for this finding is that arid regions tend
to be very heterogeneous with a high variability of low
flow producing processes, and low flows generally tend to
be lower and more variable, and therefore harder to predict.
Cold environments exhibit the largest performance range.
This could be because this class contains sub-polar and
mountainous environments that may be hydrologically
very complex, with many different storage types that com-
plicate low flow behaviour (ice/groundwater).

Which method performs best?
The regionalisation methods represented in the assessment
included: one result from the process-based methods group
(continuous runoff models); four results from the geosta-
tistics group of methods where runoff at the target site was
estimated as a weighted mean of runoff at the surrounding
gauges; ten global regression and seven regional regression
results from the regression methods group; and five results
from the short records group that used various methods.
The assessments in each group are not based on exactly the