Automatic detection of in-stream river wood from random forest machine learning and exogenous indices using very high-resolution aerial imagery

River wood (RW) plays a key role in shaping aquatic and riparian habitats while influencing sediment and water dynamics. This study presents the first automated RW detection model using Random Forest classification and near-infrared aerial imagery on the Meurthe River. By progressively incorporating exogenous indices, the model achieved recall, precision, and F1-scores between 12%–39%, 90%–94%, and 21%–54%, respectively. Validation on the Loire, Doubs, and Buëch rivers confirmed robust detection rates (75.41–86.57%) after filtering. The model also estimated RW characteristics, including length, diameter, area, and volume, with high accuracy post-calibration. These findings demonstrate the potential of remote sensing and AI for RW monitoring, providing an efficient decision-support tool for river management and habitat conservation. • The model is effective to detect and characterize automatically river wood. • The model allows to study the spatio-temporal dynamics of river wood between two surveys. • The model is generalizable on wide range of river systems. • The model is effective to detect vegetated, bars and water channel areas.