Aplicação de métodos estatísticos avançados na modelagem numérica tridimensional na análise do fluxo de percolação em barragens de terra

Abstract

The elements of an embankment dam’s internal drainage system are commonly dimensioned based on two-dimensional (2D) analyses under steady-state flow conditions. However, since the dam is a complex three-dimensional (3D) structure with transient flow conditions, this simplification may lead to the underestimation of drainage components. In this context, this study aims to develop a methodology based on the numerical analysis of seepage flow through the dam body and foundation, using three-dimensional transient modeling. For the validation of the proposed methodology, data from the Olho D’Água Dam, located in Várzea Alegre - CE, were used, obtained from the executive project, management reports from the Water Resources Management Company (COGERH), and academic studies conducted by Araújo (2013) and Vidal (2022). Monitoring periods were selected to analyze hydraulic heads observed in piezometers and water level recorders installed in the dam body and foundation, based on the decomposition of historical data series, verification of missing data, and outlier detection using the Minimum Volume Ellipsoid (MVE) method. Based on topographic data and borehole investigations, the conceptual 3D model domain was developed using MODFLOW, with data processed in Surfer software. The correlation analysis showed a direct relationship between water accumulation in the reservoir and the increase in hydraulic heads within the dam. The decomposition of the time series related to the hydraulic head measurements identified atypical residuals that did not follow seasonal trends or patterns, which were subsequently classified as outliers. The model calibration, performed through sensitivity analysis, was divided into five stages in which hydrogeological parameters and medium anisotropy under transient flow conditions were evaluated. The statistical parameters used in the analyses included Mean Error (ME), Root Mean Square Error (RMSE), correlation coefficient (r), coefficient of determination (R2), Nash-Sutcliffe Efficiency (NSE), and Mean Absolute Error (MAE), comparing observed and simulated hydraulic heads. At the end of each stage, the evaluated hydrogeological parameters were fixed based on the most favorable statistical conditions. In the final stage, the results showed ME of 0.587 m, RMSE of 0.123, r of 0.892, R2 of 0.796, NSE of 0.645, and MAE of 2.67 m, demonstrating progressive improvements in model accuracy. For model validation, R2 of 0.786 and NSE of 0.645 were obtained, indicating satisfactory calibration. Scenario simulations allowed for assessing the efficiency of the internal drainage system, identifying issues not observed in the 2D modeling, such as seepage emergence on the downstream slope near the abutments. In steady-state 3D modeling, seepage flow through the dam body was 3.7 times greater than that obtained in the 2D modeling. Furthermore, MODFLOW, commonly used for groundwater flow modeling in large areas, proved to be an effective tool not only for analyzing seepage flow in embankment dams but also for supporting the monitoring of these structures. The application provides quick and accurate information for decision-making and allows for anticipating preventive measures through scenario simulations of potential critical conditions.