Parameter Estimation of Rainfall-Runoff Models using Fuzzy Logic Intelligent Systems

Investigator

Stavros Anastasiadis (PhD candidate, University of the Aegean)

 

Supervisor

Dr Demetris F. Lekkas (University of tha Aegean)

Dr Ilias G. Pechlivanidis (co-advisor)

 

Overview

The hydrological behaviour of a catchment can be conceptualised using spatially distributed and highly interrelated water, energy and vegetation processes. Several attempts have been made to mathematically describe the different hydrological processes based on physical laws. Due to the high heterogeneity and temporal variability of physical properties, which consequently influence the individual hydrological components, assumptions and simplifications are usually required.

Uncertainty is present in model development due to the high spatiotemporal variability of the hydrological processes and the imprecise knowledge of the system. Estimating the total uncertainty inherent to a hydrological model involves the identification and quantification of four sources: natural uncertainties, data uncertainties, model parameter uncertainties, and model structure uncertainties.

Fuzzy rule based models offer an alternative for modelling the different hydrological processes. The advantage of fuzzy rules models lies on the use of linguistic terms to represent the relations between system variables resulting to reduced data requirements when compared to conceptual or data-based models.

A Simple Fuzzy Rainfall-Runoff model (sFRR)

sFFR has been developed to simulate the rainfall-runoff relationship using fuzzy logic concepts. Input (i.e. rainfall) and output (i.e. streamflow) variables are divided into 5 linguistic categories such as “very low, low, medium, high, and very high”. Five fuzzy rules are created such as “If the rainfall is very low then runoff is very low”.

sFRR reveals the simple structure of fuzzy based models and the need for suitable cluster methods in unlabeled data. Stavros is currently developing a fuzzy hydrological model of higher complexity introducing data uncertainty using fuzzy concepts.

Preliminary Results