Small-scale fisheries (SSF) play a crucial role in the Mediterranean Sea, contributing significantly to coastal livelihoods, employment, food security, and local economies. These fisheries are highly diverse and often operate with multiple passive gears within a single trip, targeting different species based on season, market demand, and fisher preference. This gear diversity, combined with the absence of trip-level gear reporting, poses a challenge for accurate monitoring, gear-specific effort estimation, and sustainable management. This study presents a Machine Learning-based approach to predict the type of fishing gear used during individual hauling events from high frequency vessel tracking data. Tracking data were collected from 10 SSF multi-gear vessels based in Ancona (Italy) between January 2023 and March 2024, and over 7000 hauling events were detected from a total of 1634 trips. Each event was labelled through fisher validation and expert-informed spatial analysis. Predictive models – Ridge Classifier, Logistic Regression, Decision Tree, Random Forest, and Extreme Gradient Boosting – were trained and tested using various sets of predictors. Two classification levels were explored: i) gear categories (nets vs. pots) and ii) specific gear types (i.e., gillnets, trammel nets, and three types of pots). With fewer predictors and optimized tuning, Random Forest reached 95% test accuracy for gear category and Extreme Gradient Boosting achieved 86% for specific gear type classification, successfully maintaining low levels of overfitting. The shared, reproducible hauling event-level approach offers a scalable tool for automated gear classification in multi-gear fisheries and contributes to more precise monitoring, management, and traceability in small-scale coastal systems. • ML algorithms were used to classify SSF passive fishing gears in multi-gear trips. • Haul-based analysis improved resolution over traditional trip-level approaches. • RaFo and XGBo achieved the best performance on test set, with minor overfitting. • The method supports SSF monitoring, aiding sustainable management and policy design. • The reproducible workflow facilitates transferability to other multi-gear contexts.
Lattanzi et al. (Sun,) studied this question.