Geography Reference
In-Depth Information
Figure 7.18. Map indicating the countries included in the Level 1
assessment.
Figure 7.19. Absolute normalised error in the centre of the FDC
(full circles), the proportion of sites with Nash
Sutcliffe efficiency
calculated over quantiles lower than 0.75 (pluses), the proportion of
sites with absolute normalised error lower than 1 (empty squares),
and the mean relative root mean square error (empty circles) of
predicting FDCs in ungauged basins stratified by climate. Each
symbol refers to a result from the studies indicated in Table A7.1 .
Boxes show 25%
-
-
75% quantiles.
of long-term FDCs based upon 1, 2 and 5 years of
observed runoff all outperform any of the other methods.
More detailed comparisons (Castellarin et al., 2004a ) with
other error measures suggest that 5 years of observed run-
off give better estimates of FDC in all respects but with just
1 and 2 years the estimate depends on the error measure
examined. The study that compared the index method with
two quantile regression methods found that the two regres-
sion methods resulted in the same performance, which was
better than that of the quantile regression. However, this
comparison was only over two catchments. In both com-
parative studies, the regressions methods perform better
than the index method, but when one considers all studies
the index method performs best.
Figure 7.20. Absolute normalised error in the centre of the FDC (full
circles), the proportion of sites with Nash
Sutcliffe efficiency
calculated over quantiles lower than 0.75 (pluses), the proportion of
sites with absolute normalised error lower than 1 (empty squares),
and the mean relative root mean square error (empty circles) of
predicting FDCs in ungauged basins stratified by regionalisation
method. Each symbol refers to a result from the studies listed in Table
A7.1. Lines indicate studies that compared different methods for the
same set of catchments. Boxes show 25% - 75% quantiles.
-
to perform slightly better than in tropical regions and
slightly lower than in cold regions.
How does data availability impact performance?
Figure 7.21 shows that performance increases with the
number of catchments used in the analysis. The trend is
particularly clear for the comparison for the intermedi-
ate classes (shown as pluses) of catchment number
(from 20 to 250 catchments per study). Also, the com-
parison between the smallest and largest classes (full
circles) is clear. Statistical regularity would suggest that
the smaller sample sizes would account for some of this
decrease in performance. This is consistent with other
studies that have evaluated the effect of sample size on
regionalisation (Spence et al., 2007 ). This trend is due
to the higher stream gauge density in the larger studies.
These results suggest that even if one is interested in
Which method performs best?
The regionalisation methods represented in the assessment
included 13 results for index methods, 11 results for
regression approaches and 3 results for estimated FDC
from short records of 1, 2 and 5 years. The assessments
in each group are not based on exactly the same regional-
isation approach, but the methodology is similar. Figure
7.20 indicates that methods using short records seem to
have the best performance, even though there are few
studies. The study that compared all three methods for
the same catchments using the same performance measure
(shown in Figure 7.20 as grey lines) shows that predictions
Search WWH ::




Custom Search