Bringing new insights into the Curve Number method based on a large catchment sample

• Ano de defesa: 2025
• Autor(a) principal: Brandão, Abderraman Róger de Amorim
• Tipo de documento: Dissertação
• Tipo de acesso: Acesso aberto
• Idioma: eng
• Instituição de defesa: Biblioteca Digitais de Teses e Dissertações da USP
• Palavras-chave em Português: aprendizado de máquina; chuva-escoamento; cobertura do solo; escoamento superficial; infiltração; infiltration; land cover; machine learning; rainfall-runoff; scs-cn; surface runoff
• Link de acesso: https://www.teses.usp.br/teses/disponiveis/18/18138/tde-22042025-095016/

In various fields, determining surface runoff is essential, and the Curve Number method from the Natural Resources Conservation Service (NRCS-CN) is the most widely used approach for estimating runoff from precipitation events. However, there are still unresolved issues in diverse aspects of the method. Data analysis and machine learning have become an integral part of modern scientific methodology, opening opportunities to explore these questions. The objective of this dissertation is to analyze the NRCS-CN method based on a large sample of watersheds from different countries. In the first part, CN values were calculated, and the effects of the standard initial abstraction value (λ= 0.2) were investigated in comparison with a proposed substitute (λ= 0.05), using three distinct methods. The precipitation threshold for CN calibration and the relationship between the method\'s parameters and watershed characteristics were also evaluated using machine learning techniques. The results indicated that the least squares method outperformed the tabulated CN method. The performance was more accurate using λ= 0.05 compared to the standard λ= 0.2, and the precipitation threshold of 25.4 mm showed the best performance for CN calculation. Land cover emerges as the most influential characteristic in determining the value λ. The spatial variability of CN revealed that high CN values were associated with slope, precipitation, and silt content in the soil, while low CN values align to forested lands, sandy soil regions, and the aridity index. In the second part, the analysis compared the ranking of precipitation and runoff event pairs with the use of natural data in calibrating the method. The analysis of ordered data resulted in higher CN values compared to natural data, varying from five to fifteen units. The estimation of CN using ordered data improved runoff performance across all methods compared to natural data. By employing a comprehensive sample of watersheds, this research provides new perspectives for improving the application of the method. The advancements resulting from this study have the potential to enhance the accuracy of runoff estimates, with practical implications that can benefit various fields of science, especially in unmonitored regions.