Estimating fishing effort in small-scale fisheries using high-resolution spatio-temporal tracking data (an implementation framework illustrated with case studies from Portugal)

• Published in: Ecological indicators Vol. 154; p. 110628
• Main Authors: Rufino, Marta M., Mendo, Tania, Samarão, João, Gaspar, Miguel B.
• Format: Journal Article
• Language: English
• Published: Elsevier Ltd 01.10.2023; Elsevier
• Subjects: Fishing effort estimation; Highly resolved boat tracks; Modelling track data; Small scale fisheries; Fishing effort estimation; Modelling track data; Highly resolved boat tracks; Small scale fisheries
• ISSN: 1470-160X; 1872-7034
• DOI: 10.1016/j.ecolind.2023.110628

•Series of procedures to develop a protocol to analyse high temporal resolution tracking data (small scale fisheries) are developed and illustrated.•Development of expert validation data, pre-processing data, and assement of the best temporal resolution.•Methods evaluation (statistical modelling and machine learning methods) based on error measures and fishing effort indicators.•New method using modes to classify boats behaviour is presented.•Portuguese SSF require less than 2 min temporal interval (ping rate) and the best method were mode and random forest applied to moving averaged speed. Small-scale fisheries (SSF, boats < 12 m) represent 90% of this sector at a worldwide scale and 84% of the EU fleet. Mapping the areas and intensity where the fishing operations occur is essential for spatial planning, safety, fisheries sustainability and biodiversity conservation. The EU is currently regulating position tracking of SSF fishing vessels requiring precision resolved geo-positional data (sec to min resolution). Here we developed a series of procedures aimed at categorizing fishing boats behaviour using high resolution data. Our integrated approach involve novel routines aimed at (i) produce an expert validated data set, (ii) pre-processing of positional data, (iii) establishing minimal required temporal resolution, and (iv) final assessment of an optimized classification model. Objective (iv) was implemented by using statistical and machine learning (ML) routines, using novel combinations of fixed thresholds estimates using regression trees and classification methods based on anti-mode, Gaussian Mixture Models (GMM), Expectation Maximisation (EM) algorithms, Hidden Markov Models (HMM) and Random Forest (RF). Of relevance, the final evaluation framework incorporates both error quantification and fishing effort indicators. We tested the method by running through four SSF fisheries from Portugal recorded every 30 sec, with 183 boat trips validated, and concluded that the more robust time interval for data acquisition in these metiers should be <2 min and that mode and random forest methods with pre-data treatment gave the best results. A special effort was concentrated in a visual support provided by the results produced by this new method, making its interpretation easier, thus facilitating transference and translation into other fishery levels. After the current validation in the Portuguese SSF fleet, we posit that our novel procedure has the potential to serve as an integrated quantitative approach to the EU SSF management.