A linguistic decision tree approach to predicting storm surge

Royston, S.; Lawry, J.; Horsburgh, K. ORCID: https://orcid.org/0000-0003-4803-9919. 2013 A linguistic decision tree approach to predicting storm surge. Fuzzy Sets and Systems, 215. 90-111. https://doi.org/10.1016/j.fss.2012.10.001

Full text not available from this repository.

Official URL: http://dx.doi.org/10.1016/j.fss.2012.10.001

Abstract/Summary

A linguistic decision tree algorithm (LID3) is applied to the problem of predicting storm surge. Of particular interest is the prediction of large positive storm surge for flood warning purposes. The application site is the North Sea which has a well-understood physical system for the generation and progression of storm surge, which lends itself to testing of the LID3 algorithm on a real-world prediction problem. Using available water level and meteorological data, the decision tree provides predictions of surge on the Thames Estuary up to View the MathML source in advance, accurate to the order of View the MathML source, which is comparable to alternative data driven methods. However, the success of the data driven approaches applied here are all limited by the sparsity of training data for extreme events (which by their nature are rare). A major benefit of the decision tree approach is the ability to make inference from the resulting IF–THEN rules of the tree structure. In this application of the LID3 algorithm, clear and plausible model rules can be deduced from the tree structure that are consistent with our understanding of the physical drivers of storm surge at this location. The label semantic framework is interpreted probabilistically, allowing the user to employ standard statistical approaches to identify statistically significant rules. It is demonstrated that the rules can successfully discriminate between surges that may pose a threat and those that should not, based on tide gauge measurements available up to View the MathML source prior to the surge signal reaching the Thames Estuary. This is promising for the potential application of such computationally efficient and easy to implement rule learning algorithms for the further investigation of complex environmental systems.

Item Type:	Publication - Article
Digital Object Identifier (DOI):	https://doi.org/10.1016/j.fss.2012.10.001
ISSN:	01650114
Date made live:	26 Mar 2013 16:22 +0 (UTC)
URI:	https://nora.nerc.ac.uk/id/eprint/500750

Item Type:

Publication - Article

Digital Object Identifier (DOI):

https://doi.org/10.1016/j.fss.2012.10.001

ISSN:

01650114

Date made live:

26 Mar 2013 16:22 +0 (UTC)

URI:

https://nora.nerc.ac.uk/id/eprint/500750