Refining image-based approaches for river monitoring: a focus on graphical enhancement and software parameter sensitivity

Image-based techniques for estimating river surface velocity include traditional Large-Scale Particle Image Velocimetry (LSPIV) methods based on cross-correlation of tracer displacements and alternative optical flow approaches that infer motion from image intensity gradients. The accuracy of both is contingent on the quality of the input images and the configuration of the processing algorithms. This study presents a sensitivity analysis evaluating the impact of different combinations of graphical enhancement filters and software parameterizations on performance under varying environmental conditions. Image-based measurements were performed in two field case studies in Sicily (Italy): the Oreto and Platani rivers. Acquired videos were pre-processed applying enhancement filters and testing multiple parameter configurations. Analyses were carried out using PIVlab and SSIMS-Flow, each tested with distinct processing setups. Overall, 1296 surface velocity fields were derived and compared to ADCP reference data. To objectively characterize water clarity and the background conditions, the Normalized Difference Turbidity Index (NDTI) was used for the acquired sequences. The findings of the study demonstrate that both filter selection and software settings significantly influence image velocimetry accuracy. PIVlab, relying on cross-correlation, showed greater robustness in visually complex scenarios, while SSIMS-Flow, based on intensity gradients, achieved higher accuracy in uniform and turbid conditions. Poor performance was generally associated with unsuitable filter or parameter combinations. These findings underscore the necessity of adapting image processing workflows to specific environmental contexts. The proposed evaluation framework offers a replicable approach for optimizing LSPIV and optical flow accuracy and supports the development of adaptive monitoring solutions for hydrological applications.