The interplay of citizen-sourced, conventionally surveyed, and meteorological data in recreational fisheries

Abstract

Traditional methods of collecting data on angler activity involve conventional surveys, such as creel surveys and aerial surveys, which are often costly. A modern, cost-effective alternative is utilizing online platforms and smartphone applications (apps) designed for anglers. Previous studies identified correlations between data reported by citizens via these apps and data gathered from conventional surveys. However, it is still unclear if the activities recorded by the two sources are directly related, or if other “intermediate” variables are primarily related to the conventionally surveyed data. In my first study, I employed Bayesian networks (BNs) to explore this question, focusing on two metrics: daily catch rate and daily fishing effort. These metrics were sourced from creel surveys, aerial surveys, and Angler’s Atlas website with related MyCatch app in Alberta and Ontario, Canada. I included additional factors, e.g., weather conditions, as possible “intermediate” variables in the network. To study the uncertainty of the results, I measured the strength of connections between variables using Bayesian model averaging. Waterbody webpage views were directly related to daily and weekly-aggregated boat counts in Ontario (51% and 100% probability) and to weekly-aggregated creel survey-reported fishing duration in Alberta (100%). This highlights the value of citizen-sourced data in providing unique insights beyond meteorological factors, with online interest serv ing as a potentially reliable proxy for angler pressure and effort. In my second study, I aimed to evaluate three BN structure learning approaches: (i) expert knowledge, (ii) ChatGPT, and (iii) data-driven models, in predicting angler activity as reported through aerial surveys on the Ontario dataset. The Friedman test indicated no significant difference in prediction accuracy between the three models. These findings underscore the potential of AI-driven approaches, as the ChatGPT-assisted model performed on par with expert-based and data-driven models, demonstrating its viability for ecological predictions.