Decoupling Geometric and Compaction Effects on Pervious Concrete Properties: A Gaussian Process Regression and Sensitivity Analysis

ABSTRACT Specimen geometry and compaction methodology vary considerably across pervious concrete studies, yet their independent contributions to measured porosity and compressive strength are not well established. This study systematically investigates how mould aspect ratio (AR), surface-area-to-volume (SAV) ratio, and compaction layer height jointly influence these properties through a controlled experimental programme paired with Gaussian process regression and a variance-based Sobol sensitivity analysis. Twelve molds–compaction configurations spanning six molds geometries and two compaction strategies were tested, with total compaction energy per unit volume held constant to isolate the effect of energy distribution. Results reveal a clear decoupling between the two properties: porosity is predominantly controlled by geometric parameters (a combined first-order Sobol contribution of 73.7 %), whereas the compressive strength is primarily governed by the SAV ratio and layer height (a combined contribution of 88.2 %). Predictive uncertainty is highest at extreme ARs and large layer heights, identifying these as priority regions for further experimental work. These findings suggest that a single cube-to-cylinder conversion factor is unlikely to preserve both porosity and strength simultaneously, and that separate characterization strategies may be warranted for each property.