Analytic Hierarchy Process (AHP) in Ranking Non-Parametric Stochastic Rainfall and Streamflow Models

Abstract

Analytic Hierarchy Process (AHP) is used in the selection of categories of non-parametric stochastic models for hydrological data generation and its formulation is based on pairwise comparisons of models. These models or techniques are obtained from a recent study initiated by the Water Research Commission of South Africa (WRC) and were compared predominantly based on their capability to extrapolate data beyond the range of historic hydrological data. The different categories of models involved in the selection process were: wavelet (A), reordering (B), K-nearest neighbor (C), kernel density (D) and bootstrap (E). In the AHP formulation, criteria for the selection of techniques are: "ability for data to preserve historic characteristics", "ability to generate new hydrological data", "scope of applicability", "presence of negative data generated" and "user friendliness". The pairwise comparisons performed through AHP showed that the overall order of selection (ranking) of models was D, C, A, B and C. The weights of these techniques were found to be 27.21%, 24.3 %, 22.15 %, 13.89 % and 11.80 % respectively. Hence, bootstrap category received the highest preference while nearest neighbor received the lowest preference when all selection criteria are taken into consideration.

Keywords

• Analytic hierarchical process
• non-parametric stochastic
• hydrological data