IAHR, founded in 1935, is a worldwide independent member-based organisation of engineers and water specialists working in fields related to the hydro-environmental sciences and their practical application. Activities range from river and maritime hydraulics to water resources development and eco-hydraulics, through to ice engineering, hydroinformatics, and hydraulic machinery.
Log On
About IAHRDirectoryCommitteesMy IAHRNews & JournalseLibraryeShopEventsJoin IAHRWorld CongressDonate
spacer.gif
spacer.gif eLibrary
spacer.gif eLibrary
You are here : eLibrary : IAHR World Congress Proceedings : 32nd Congress - Venice (2007) : THEME B: Data Acquisition and Processing For Scientific Knowledge and Public Awareness. : A sugeno fuzzy model for daily rainfall series prediction based on local meteorological data
A sugeno fuzzy model for daily rainfall series prediction based on local meteorological data
Author : Alberto Cavallo , Roberto Greco
A fuzzy model for the prediction of daily rainfall height series has been developed. Twenty elementary fuzzy rules linking input variables (mean atmospheric pressure, mean wind direction and intensity, mean air temperature and relative humidity, rainfall height) observed at day i, with the output variable (rainfall height of day i+1) have been defined. For each of the elementary rules, a membership function for the input variable have been introduced, with shape depending on three parameters. An incomplete rule base of the model has been defined, formed by the elementary rules, i.e. rules with a single premise, and by all the possible pairs of rules (in total 130 pairs). Model output has been calculated as the weighted average of crisp values of rainfall height corresponding to each rule of the rule base. The model has been tested against daily meteorological data collected at Ponza, in the Tirrenian Sea, between years 1982 and 2002. Three model calibration strategies have been tested: yearly calibration, dividing the data set in two parts, for model training and validation, respectively; monthly calibration, using separately data belonging to the same month; real time calibration, using the data observed during the previous five years in the 30 days before the date at which the prediction is to be done. The first results of model identification show that with the first two calibration strategies the model has a tendency to overspecialize on the calibration (training) time series, while validation results are worse, although the generated rainfall height series resemble the observed ones in terms of total monthly precipitation, number of rainy days, peak precipitation, thus implying that the model has learnt the statistics of the phenomenon. The real time calibrated model looks more promising as a predictive tool. Further research will be devoted to the selection of the most effective calibration strategy as well as to improve model structure by modifying some of the fuzzy rules.
File Size : 199,011 bytes
File Type : Adobe Acrobat Document
Chapter : IAHR World Congress Proceedings
Category : 32nd Congress - Venice (2007)
Article : THEME B: Data Acquisition and Processing For Scientific Knowledge and Public Awareness.
Date Published : 01/07/2007
Download Now