Geoscience Reference
In-Depth Information
Figure 4.14
WS2 catchment, H J Andrews Forest, Oregon: (a) depth 5 regression tree and (b) discharge pre-
diction using a regression tree with 64 terminal nodes (after Iorgulescu and Beven, 2004, with kind permission
of the American Geophysical Union).
response of two small catchments in the H J Andrews Research Forest in Oregon, one of which had been
harvested of trees and one of which had been left as a control. This method of assessing the impacts of
catchment change of time is presented as a case study in Chapter 8.
4.6.4 Fuzzy Inference
When the first edition of this topic appeared, a decade ago, models based on fuzzy inference were only
in the initial stages of being applied to hydrological models but I thought that it might be a promising
technique. There have been studies that extend early work on modelling soil water flows (Bardossy
et al.
, 1995; Schulz and Huwe, 1997) into rainfall-runoff modelling ( Ozelkan and Duckstein, 2001;
Hundecha
et al.
, 2001; Vernieuwe
et al.
2005), flow forecasting (Nayak
et al.
, 2005; Alvisi
et al.
, 2006)
and estimation of missing rainfall records (Abede
et al.
, 2004), but it cannot be said that fuzzy inference
has had a major impact on rainfall-runoff modelling practice. It may still prove to be promising in the
future, as the hydrological modelling community starts to address issues of epistemic uncertainties in
simulating catchments and the limitations of available hydrological data. The related methodologies of
fuzzy rules and fuzzy trees can both be used to represent rainfall-runoff information, with the results
expressed as either defuzzified (crisp) numbers or fuzzy possibilistic quantities. The latter, in particular,
could be useful as a way of representing some difficult uncertainties. Alvisi
et al.
(2006) raise an issue
in prediction, that combinations of circumstances that have not been seen in training the fuzzy inference
