We can't find the internet
Attempting to reconnect
Something went wrong!
Hang in there while we get back on track
High-efficiency clean production of yellowfin tuna (Thunnus albacares) based on three-dimensional marine environmental variables and interpretable machine learning
Summary
This paper is not primarily about microplastics — it uses machine-learning tree models to predict fishing grounds for yellowfin tuna in the Indian Ocean based on oceanographic and catch data, with only a passing mention of microplastics as one of many marine pollutants.
• Extra Trees is the optimal model for predict yellowfin tuna fishing ground • Latitude, longitude, year, and water temperature are key drivers • The yellowfin tuna annual catch showed significant inter-annual fluctuations Fishing grounds forecasting is an important tool to enhance fleet productivity, reduce greenhouse gas emissions and marine pollution (microplastics, etc.), and thus achieve sustainable development of yellowfin tuna ( Thunnus albacares ) fisheries in the Indian Ocean. In this paper, based on the annual catch data of yellowfin tuna in the Indian Ocean from 2000 to 2023, we analyzed the changing law and compared the differences in the fishing grounds prediction performance of the six tree models with different training and test set division ratios, and finally based on the model built-in method, SHAP and LIME, the optimal model was visualization interpretation. The main results were as follows: 1) the annual catch showed significant inter-annual fluctuations and reached a historical peak in 2004-2005; 2) the Extra Trees model consistently showed the optimal prediction performance among all the training/test set partitioning schemes, while the AdaBoost model had the worst. As the number of training samples increased, the prediction accuracy and stability of all six models improved; 3) global importance indicated that latitude, longitude, year, and water temperature were the key drivers of the distribution of yellowfin tuna. In the local interpretation analysis, the SHAP dependence plot further revealed significant interaction effects among different variables, while the SHAP force plot and LIME interpretation results indicated that the main variables driving the model decision were basically the same in the positive and negative samples, but the specific degree of contribution varied depending on the sample area and the values of the variables.