Related Experiment Video
Updated: Aug 5, 2026
Development of an Individual-Tree Basal Area Increment Model using a Linear Mixed-Effects Approach
Published on: July 3, 2020
A hybrid machine learning framework boosts length-based stock assessments for data-limited fisheries under
Richard Kindong1,2,3,4, Ousmane Sarr1, Njomoue A Pandong5
1College of Marine Living Resource Sciences and Management, Shanghai Ocean University, Shanghai 201306, China.
Abstract:
Over 80% of global fisheries catches lack data for traditional stock assessments. Using simulated populations of three life-history species under equilibrium and non-equilibrium scenarios, we evaluated two length-based methods: the length-based spawning potential ratio (LBSPR) and the length-based Bayesian (LBB) method. LBSPR outperformed LBB across most scenarios, particularly for medium- and long-lived species, while both methods were sensitive to uncertainties in natural mortality and growth coefficient. We then integrated LBSPR with a support vector machine (SVM) to refine estimates of fishing mortality (F) and spawning potential ratio (SPR). The SVM-LBSPR hybrid improved accuracy, reducing root-mean-square error by up to 48% for F and 18% for SPR on independent test data. By bridging mechanistic models with machine learning, this hybrid framework enhances the robustness of data-limited stock assessments, offering a pragmatic tool for fisheries managers navigating parameter uncertainties and dynamic population changes.