Physics-guided machine learning for saturation control across variable hydraulic conditions in heap leaching (dataset)

Simulated dataset supporting "Physics-guided machine learning for saturation control across variable hydraulic conditions in heap leaching" (arXiv:2610.15947). The file dataset.h5 contains 4096 numerical Richards equation trajectories for 1D heap leaching columns, generated with a modified Crank Nicolson solver using the van Genuchten Modified relations. Four hydraulic parameters and the initial saturation were sampled by scrambled Sobol sampling within the following ranges: saturated hydraulic conductivity Ks 130 to 232 cm/day, van Genuchten n 2.20 to 3.00, van Genuchten alpha 0.025 to 0.042 1/cm, saturated moisture content 0.270 to 0.388, and initial effective saturation 0.40 to 0.63. The runs are split into 2867 training, 614 validation and 615 evaluation cases. The simulator, data generation pipeline, training code and closed loop controller are available at https://github.com/Sultan-Hamdan/PGML-Copper-Heap-Leaching