A Photogrammetric Simulation Framework for Rockfall Change Detection with Statistically Validated Measurement Noise

Rockfalls are natural slope-instability phenomena that pose a significant hazard to infrastructure and human activity. In recent years, the increasing availability of high-resolution three-dimensional (3D) models acquired through photogrammetric techniques has enabled detailed pre-/post-event analyses of rock slopes. However, in this domain, the availability of accurate ground truth for the quantitative evaluation of 3D change detection methods and for training machine-learning approaches aimed at recognising and volumetrically quantifying detachments on rock faces remains very limited. This work presents a simulator that, starting from a 3D model of a rock face, generates pre-/post-failure scenarios through controlled removal of rock blocks and produces photogrammetric acquisitions affected by realistic measurement noise. The pipeline emulates the main processing stages of the reconstruction workflow. Noise realism is validated and calibrated by comparing real and simulated data through a multi-indicator framework (marginal distribution, variogram, power spectrum, and multiscale roughness), integrated into a Mahalanobis-distance-based acceptance test with empirical thresholds derived from real measurements. Results from two pilot sites show that, after site-specific tuning of the simulator noise levels, the calibrated configurations reproduce the main magnitude and spatial-structure characteristics of the real noise, with stronger agreement for the fixed stereo-pair configuration and partial but still informative agreement for the more complex UAV-based case. Moreover, the simulator provides a controlled environment for benchmarking and sensitivity analyses of change detection methods.