Prediction of low flows in ungauged basins - Runoff Prediction in Ungauged Basins: Synthesis across Processes, Places and Scales - page 184

8.5.2 Level 2 assessment
The Level 1 synthesis of existing studies ( Table A8.1 )
clearly showed that many studies only report summary
statistics of regionalisation performance and/or catchment
characteristics, which hampers detailed attribution of the
performance and comparison of results between studies.
The objective of the Level 2 synthesis is to examine and
explain the performance of the regionalisation methods in
greater detail. Six study authors from the Level 1 assess-
ment provided detailed information about climate and
catchment characteristics in a consistent way and reported
the regionalisation performance for each catchment ( Table
A8.2 ). This data set combines data from 2455 catchments,
four groups of regionalisation methods and four catchment
characteristics. The regionalisation methods are geostatis-
tics, global regression, regional regression and estimation
from short records (see Salinas et al., 2013 ). The catch-
ment characteristics are aridity (potential evaporation by
mean annual precipitation), mean annual air temperature,
mean elevation and catchment area. As performance indi-
cators the normalised error (NE) and the absolute normal-
ised error (ANE) were used (Table 2.2). The NE highlights
biases in the methods while the ANE is a measure of the
overall performance. Note that the ANE is an error meas-
ure, so it has been plotted downwards on the vertical axis
to make it comparable with the performance measures, i.e.,
higher up in the plot is better. For comparison with the
other runoff signatures in Chapter 12 , the R² of predicting
low flow indices were calculated for all methods in each
study separately. The 25% and 75% quantiles of these R²
are 0.57 and 0.73, respectively.

1.0

0.8

0.6

0.4

0.2

100 250
Number of catchments

Figure 8.18. Coefficient of determination (R²) of predicting low flows
in ungauged basins stratified by the number of catchments within each
study. Each circle refers to a result from the studies in Table A8.1.
Boxes show 25% - 75% quantiles. After Salinas et al. (2013) .

potential to explore the impact of environmental change
than statistical methods.

How does data availability impact performance?
Figure 8.18 shows the predictive performance (R²) as a
function of the number of catchments analysed in each study.
It is clear that the studies with less than 100 catchments have,
on average, the lowest performance and performance
increases with the number of catchments used in analysis.
This is because of the higher stream gauge density in the
larger studies. The performance decreases for very large data
sets (

250 catchments). This decrease is related to the higher
heterogeneity of larger study areas and to the fact that a
number of the studies used global regression methods that
did not perform very well in these regions.

>

To what extent does runoff prediction performance depend
on climate and catchment characteristics?
The assessment of the predictive performance of the
models with respect to the four climate and catchment
characteristics is presented in Figures 8.19 and 8.20 . Over-
all, the errors, ANE and NE, clearly increase with increas-
ing aridity and mean annual temperature T A . This means
that the performance is consistently lower in warmer, drier
and more arid environments. These are regions that tend to
be particularly heterogeneous and low flows may be small,
which makes them particularly hard to predict.
Figures 8.19 and 8.20 indicate that there is a tendency for
performance to increase with catchment elevation. The
average of all methods shows that errors decrease from
0.37 for low land catchments (mean elevation

Main findings of Level 1 assessment
- In humid regions the performance of predicting low

flows in ungauged basins tends to be better than in
other climates.

- Methods that use short runoff records at the site of

interest perform significantly better than any regional-
isation method provided 3

-

5 years of data are available.

- Regional regressions that divide a domain into sub-

regions and apply regression models separately
always perform much better than global regressions.

- Geostatistical methods can perform better than

regional regressions in regions with medium to high
stream gauge density if the stream network structure
is taken into account.

200 m
a.s.l.) to 0.16 for high mountain catchments. This may be
partially due to the higher specific discharges of mountain-
ous catchments compared to lowland catchments, which
may increase predictability. Also, in the high mountains,
low flows may be of a winter low flow type, so low flows
may depend on frost strength, which is closely related to

<

- The performance tends to increase with number of

stations in a region but may decrease if global regres-
sions are used in the large regions.