Regionalization of hydrological variables for the Paraná state, Brazil

• Ano de defesa: 2021
• Autor(a) principal: Nascimento, Jéssica Garcia
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: eng
• Instituição de defesa: Biblioteca Digitais de Teses e Dissertações da USP
• Palavras-chave em Português: Machine learning; Balanço hídrico; Hydrological models; Machine Learning; Modelos hidrológicos; Multiple regressions; Regressões lineares múltiplas; Water balance
• Link de acesso: https://www.teses.usp.br/teses/disponiveis/11/11152/tde-07042021-153935/

The knowledge of the flows and discharge in the rivers is essential for water resources management since it represents the availability of water in the watersheds. Generally, the long-term average flows (Qm), flows exceeded or equaled in 90% and 95% of the time (Q90 and Q95, respectively), which represents minimum flows, and annual discharge (Q) are frequently used in the water resources management. The information about those parameters can be considered a challenge, especially in developing countries where monitoring by gauges is limited in terms of density and frequency of observations. Thus, hydrological models can be applied to predicted flows in unmonitored watersheds, as the Multiple Linear Regressions (MLR) method, the oldest and widely used method for regionalization problems, and new techniques as machine learning approaches, for example, the Random Forest (RF). Additionally, the water balance can be used to estimate the annual discharge of rivers in watersheds. In this context, remote sensing products offer precipitation (PPT) and evapotranspiration (ET) products with great spatial and temporal coverage, which can be used in hydrological models, improving its performance. In a general scenario, this study aimed to analyze the performance of Integrated Multi-satellitE Retrievals for GPM (IMERG) products to estimate the PPT over Paraná state, Brazil. Subsequently, the IMERG monthly products were used with watershed morphological descriptors to build hydrological models for predicting the flows (Q90, Q95, and Qm) in 81 watersheds in Paraná state, Brazil. Lastly, we predicted the Q using the IMERG monthly products and the ALEXI (Atmosphere-Land Exchange Inverse) ET over 28 watersheds in Paraná state, by water balance equation. The models performed very well to predict the hydrological variables, which demonstrated the importance of remote sensing and hydrological models in water resources management.