Hydrological modelling using multiple Gaussians: application to Negro River Basin

We model water level data from hydrometric stations in the Negro River basin using multiple Gaussian distributions. To characterize hydrological seasonality, we apply statistical and unsupervised learning techniques to analyse the time series. Gaussian fitting reproduces key hydrological patterns, capturing seasonal variability and long-term trends. A Gaussian Mixture Model (GMM) is implemented, with parameters estimated via the Expectation Maximization algorithm and model selection guided by the Bayesian Information Criterion (BIC) and Akaike Information Criterion (AIC). Model performance is evaluated using Quantile-Quantile (QQ) analysis with bootstrap confidence envelopes, showing good agreement with observations. Results highlight variability and similarities in water level dynamics at Serrinha, Santa Maria do Boiaçú, and Moura stations. The approach is computationally efficient and adaptable to large-scale datasets, and can be extended to other Amazon subbasins, supporting improved understanding of hydrological dynamics and water resource management.