An Integrated Geostatistical–Machine Learning Framework for Predicting Resurgence and Canaliculi in the Belo Monte Hydropower Dikes Using Borehole Data

The Belo Monte Hydroelectric Complex, located on the Xingu River in Altamira, Pará, Brazil, includes two powerplants: Pimental, built directly on the riverbed, and Belo Monte, which uses water diverted from an intermediate reservoir. This reservoir is contained by 28 dikes, mostly underlain by migmatite residual soils. During construction and early operation, tubular cavities known as canaliculi and associated resurgence processes were identified downstream of several dikes. However, the spatial controls governing these phenomena remain poorly understood in tropical lateritic environments. This study proposes an integrated geostatistical–machine learning framework for the spatial prediction of resurgence and canaliculi. Ordinary Kriging was applied to model the elevation of the Young Residual Soil (YRS) contact surface using approximately 450 borehole records. The model was subsequently extrapolated to downstream areas lacking direct subsurface data through Support Vector Machine (SVM) regression. A digital terrain model was combined with the modeled YRS contact to identify potential outcrop zones associated with resurgence. Spatial statistical indicators based on minimum-distance metrics were developed to quantify the correspondence between predicted outcrop areas and fieldmapped occurrences. The results show strong spatial agreement between predicted YRS outcrop zones and mapped resurgence and canaliculi points across five representative dikes, with average shortest distance ranging from 4.4 to 50.6 m. Cross-validation indicated satisfactory predictive performance of the YRS digital model, with RMSE values between 4.16 and 9.56 m for kriging and between 3.36 and 10.65 m for SVM. The proposed framework provides a replicable and cost-effective predictive tool for dam safety management, supporting the definition of priority inspection corridors and early detection of anomalous resurgence and canaliculi behavior in tropical dam foundations.