984 3. To structure the modelling work as a sequential test of two competing hydrogeological hypotheses of increasing complexity, identifying the minimum model complexity that observed pressure data at the site can support, and that decisions about dewatering design require. The two competing hypotheses are formulated as follows: H1: Hydraulic anisotropy at Mingomba, represented as a directionally variable permeability tensor in a single-porosity continuum model, is sufficient to reproduce observed pressure data and predict pore pressure evolution under dewatering. Fracture effects are captured implicitly through bulk directional permeability without explicit fracture geometry. H2: Where H1 is insufficient, pressure evolution is governed by dual porosity behaviour, in which the rock matrix and fracture system constitute two interacting domains that exchange fluid through a transfer coefficient, producing delayed drainage responses that a single-porosity model cannot reproduce. Explicit representation of discrete fracture networks and individual faults is excluded from the present scope due to limited structural data at the site; their effects are represented through bulk anisotropy in H1 and through matrix-fracture exchange in H2. This is a deliberate parsimony choice consistent with the principle of building only the model complexity that available data can constrain. 3. METHODOLOGY 3.1 Uncertainty quantification framework The methodology follows the Popper-Bayes protocol for uncertainty quantification under decisionmaking (Caers, 2018; Fang et al., 2022). The protocol differs from conventional history matching in that it does not begin from a single calibrated base case. Instead, it begins from a stated prior probability distribution over all uncertain parameters, generates an ensemble of model realisations through Monte Carlo simulation, and explicitly tests whether the prior is consistent with observed data through a falsification step. If the prior is falsified, sensitivity analysis is used to diagnose which assumptions are problematic; if it is not falsified, Bayesian inversion is used to produce posterior probabilistic forecasts. Six stages structure the protocol: formulation of the decision problem and reward variables; statement of the prior uncertainty model; Monte Carlo simulation; global sensitivity analysis on data and reward variables; falsification testing; and Bayesian inversion of the non-falsified prior. The present study has completed stages one through three; stage four (sensitivity analysis) is the immediate next step, with stages five and six planned thereafter. 3.2 Geological model and prior uncertainty A three-dimensional structural model of the Mingomba deposit was constructed covering twelve stratigraphic units from the Petit Conglomerate at the surface to the Mindola Clastics Formation at depth. Prior probability distributions were assigned to matrix permeability for each stratigraphic unit (Table 1), to the vertical permeability ratio in the Bancroft Dolostone and Nchanga Member confining units, and to a scalar fracture permeability multiplier representing bulk fracture-induced enhancement of matrix permeability. All distributions were specified as uniform across the site-relevant range, reflecting limited site-specific data. Three regional aquifer recharge boundary conditions were defined based on documented inflow records from analogue Copperbelt mines (Table 2): inflow from the northwest representing the Lubambe analogue, inflow from the southeast representing the Konkola analogue, and a distributed component representing other lateral and surface recharge sources. The base values were informed by the historical inflow ranges reported for these adjacent mines, with uniform distributions spanning plausible Mingomba conditions given the documented hydraulic connections between the sites.
RkJQdWJsaXNoZXIy MTM0Mzk2