Data-Driven Bedload Inference from RFID Pebble Tracing in a Pre-Alpine Stream

We analyse pebble RFID tracing observations to investigate sediment transport dynamics in gravel-bed rivers using statistical modelling. This study examines a dataset of nearly 3500 tracer displacement measurements collected during 27 sediment-mobilizing events in a pre-Alpine reach in Italy. Our analysis follows three main steps, addressing tracer mobility patterns, event-scale transport dynamics, and reach-scale bedload inference. First, using Markov Chain analysis of state transitions on typical and high-magnitude transport events, we demonstrate that pebbles tend to maintain their mobility state between events, characterizing the between-event intermittency of bedload transport. A subsequent analysis of flow characteristics reveals that consecutive floods of similar magnitude exhibit increasing movement probability while maintaining similar virtual velocities. Finally, we train Gradient Boosting regression models to estimate distributions of pebble displacements and virtual velocities (defined, following common usage, as the ratio between the distance a tracer travels during a mobilising event and the duration of that event). Together with Monte Carlo propagation, these models are used to derive reach-scale volume estimates. The models identify flow rate and event duration as primary controls, while grain size has minimal influence within the sampled range of tracer dimensions. To strengthen our approach, we implement an extensive multi-stage validation process aimed at both single-tracer predictions and overall basin-scale movement estimates. The results indicate that high-magnitude transport events (12% of observations) contribute similar bedload volumes as typical events (88% of observations), highlighting the significant role of extreme events in total sediment transport. Model predictions yield bedload volume estimates that align well with independent measurements from a downstream sediment retention basin.