Prediction of seasonal runoff in ungauged basins - Runoff Prediction in Ungauged Basins: Synthesis across Processes, Places and Scales - page 129

groups of regionalisation methods and four catchment
characteristics. The regionalisation methods are regres-
sion, spatial proximity, geostatistics and process-based
approaches. The catchment characteristics are aridity
(potential evaporation by mean annual precipitation),
mean annual air temperature, mean elevation and catch-
ment area.
To make the Level 2 assessment comparable with both
the Level 1 assessment of seasonal runoff and the assess-
ment in the other chapters of this topic, two performance
measures were used. In both instances, the performance is
estimated from the predicted Pardé coefficients rather than
on the monthly runoff, since annual runoff is dealt with in
Chapter 5 . The first performance measure is the NSE
calculated on the 12 Pardé coefficients for each site. The
second performance measure is the normalised error (NE)
of the range of the Pardé coefficients (max-min) as well as
the absolute value of the latter (ANE). The NE and ANE
identify how well the strength of the seasonality is pre-
dicted. The NE highlights biases in the methods while the
ANE is a measure of the overall performance. The predict-
ive performance of the timing of the maximum was also
analysed but is not presented here because it was always
very good. Note that the ANE is an error measure, 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 median NSE of monthly
Pardé coefficients were calculated for all methods in each
study separately. The 25% and 75% quantiles of these NSE
are 0.84 and 0.91, respectively.

Figure 6.27. Median Nash Sutcliffe efficiency (NSE) (circles),

median spatial per-month adjusted r 2 (pluses) and median Spearman
correlation coefficient (squares) of predicting seasonal runoff in
ungauged basins stratified by the number of catchments within each
study. Each symbol refers to a result from the studies shown in Table
A6.1. Lines indicate a study that compared the same methods in
different regions. Boxes show 25%

-

- 75% quantiles.

apparently, much can be gained by the availability of
runoff stations to identify the dominant processes at the
ungauged sites.

Main findings of Level 1 assessment
-

In humid and cold regions the performance of predic-
tions of seasonal runoff in ungauged basins is sig-
nificantly better than in arid climates.

- Geostatistical methods perform better than process-

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

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 characteris-
tics is presented in Figures 6.28

- The performance clearly increases with the number of

stream gauges in the region.

6.30. The top panels of
Figures 6.28 and 6.29 show a clear decrease of NSE per-
formance and increase of ANE error of the regression and
spatial proximity approaches with aridity for aridity indices
greater than 1. For the most arid catchments, predicting the
range of the runoff regime is difficult as they may depend on
the local soil moisture status and local precipitation effects.
These are catchments from the US data set. For geostatistics
and process-based methods, no clear dependence on the
aridity is apparent. All methods are unbiased with regards
to aridity ( Figure 6.30 ) except spatial proximity, which
shows a slight underestimation for humid places.
Figures 6.28 and 6.29 indicate that the dependence of
the performance on air temperature and elevation depends
on the region. In Austria the NSE performance of the
regression method increases with elevation and decreases
with temperature, due to the fact that the seasonality of

-

6.5.2 Level 2 assessment
The Level 1 synthesis of existing studies (Table A6.1)
clearly showed that many studies only report summary
statistics of regionalisation performance and/or catch-
ment characteristics, which hampers detailed attribution
of the performance and inter-study comparison of results.
The objective of the Level 2 synthesis is to examine and
explain the performance of the regionalisation methods
in greater detail. Two study authors from the Level 1
assessment studies, plus two other authors provided
detailed information about climate and catchment char-
acteristics in a consistent way and reported the regional-
isation performance for each catchment (Table A6.2).
This data set combines data from 1641 catchments, four