Portal de Periódicos da CAPES

Artigo Produção Nacional Revisado por pares

BIBTEX RIS

Artificial intelligence techniques coupled with seasonality measures for hydrological regionalization of Q90 under Brazilian conditions

2016; Elsevier BV; Volume: 541; Linguagem: Inglês

10.1016/j.jhydrol.2016.08.046

1879-2707

Samuel Beskow, Carlos Rogério de Mello, Marcelle Martins Vargas, Leonardo de Lima Corrêa, Tamara Leitzke Caldeira, Matheus Fonseca Durães, Marílton Sanchotene de Aguiar,

Flood Risk Assessment and Management

Information on stream flows is essential for water resources management. The stream flow that is equaled or exceeded 90% of the time (Q90) is one the most used low stream flow indicators in many countries, and its determination is made from the frequency analysis of stream flows considering a historical series. However, stream flow gauging network is generally not spatially sufficient to meet the necessary demands of technicians, thus the most plausible alternative is the use of hydrological regionalization. The objective of this study was to couple the artificial intelligence techniques (AI) K-means, Partitioning Around Medoids (PAM), K-harmonic means (KHM), Fuzzy C-means (FCM) and Genetic K-means (GKA), with measures of low stream flow seasonality, for verification of its potential to delineate hydrologically homogeneous regions for the regionalization of Q90. For the performance analysis of the proposed methodology, location attributes from 108 watersheds situated in southern Brazil, and attributes associated with their seasonality of low stream flows were considered in this study. It was concluded that: (i) AI techniques have the potential to delineate hydrologically homogeneous regions in the context of Q90 in the study region, especially the FCM method based on fuzzy logic, and GKA, based on genetic algorithms; (ii) the attributes related to seasonality of low stream flows added important information that increased the accuracy of the grouping; and (iii) the adjusted mathematical models have excellent performance and can be used to estimate Q90 in locations lacking monitoring.