From pixels to peaks: integrating LiDAR and RGB drone imagery to map mussel spat on intertidal rocky shores

Remote monitoring of the spatial cover of sessile intertidal species can provide important information for the conservation and management of commercial species. However, the use of high-resolution RGB imagery combined with topographic data acquired by Unmanned Aerial Vehicles (UAVs) to map rocky intertidal sessile invertebrates, particularly wild spat of Mytilus galloprovincialis , remains largely unexplored. Galicia (northwestern Spain) is the leading producer of farmed mussels in Europe, and most of mussel rafts are seeded with wild spat collected from intertidal rocks. Using a UAV equipped with an RGB camera and LiDAR sensor, we surveyed a rocky intertidal area in Galicia where mussel spat are harvested. Adult mussels and spat, the goose barnacle Pollicipes pollicipes , other barnacles, and red, brown and green macroalgae were classified using a model stack composed of random forest, gradient tree boosting, support vector machine and k-nearest neighbours. Predictors included the RGB, digital surface model (DSM) and topographic indicators (e.g. orientation, slope, topographic position index, topographic ruggedness index) from a raster composite with a pixel size of 10 mm. Model comparison revealed that gradient tree boosting performed best, with an overall accuracy of 0.84, a kappa value of 0.82 and user accuracies of 0.92 for adult mussels, 0.81 for mussel spat and 0.96, 0.92 and 0.87 for red, brown and green macroalgae, respectively. The prediction map revealed the spatial distribution of mussel spat expressed as percentage cover in 15 cm x 15 cm cells. Twenty percent of the cells in which mussel spat were present exhibited ≥50% cover, which is lower than the values reported from in situ surveys in the region. Elevation (derived from the DSM) and the RGB channels were the most important features for model-based discrimination of the classes. Our findings demonstrate the usefulness of elevation data for remote detection of intertidal populations, highlighting the potential value of UAV surveys for managing small-scale fisheries. • Wild mussel spat harvesting for aquaculture needs to be monitored and managed • A rocky intertidal zone was surveyed with a drone equipped with RGB camera and LiDAR • A machine learning model stack was fitted with the RGB, elevation and topography data • XGBoost model yielded OA of 0.84, κ of 0.82 and user accuracy of 0.81 for mussel spat • The mean cover observed in 15 x 15 cm cells containing mussel spat was 27%. • In 20% of the cells containing mussel spat, cover was 50% or higher.