Forecasting Albacore (Thunnus alalunga) Fishing Grounds in the South Pacific Based on Machine Learning Algorithms and Ensemble Learning Model

Abstract

To achieve high-precision forecasting of different grades of albacore fishing grounds in the South Pacific Ocean, we used albacore fishing data and marine environmental factors data from 2009 to 2019 as data sources. An ensemble learning model (ELM) for albacore fishing grounds forecasting was constructed based on six machine learning algorithms. The overall accuracy (ACC), fishing ground forecast precision (P) and recall (R) were used as model accuracy evaluation metrics, to compare and analyze the accuracy of different machine learning algorithms for fishing grounds forecasting. We also explored the forecasting capability of the ELM for different grades of fishing grounds. A quantitative evaluation of the effects of different marine environmental factors on the forecast accuracy of albacore tuna fisheries was conducted. The results of this study showed the following: (1) The ELM achieved high accuracy forecasts of albacore fishing grounds (ACC = 86.92%), with an overall improvement of 4.39~19.48% over the machine learning models. (2) A better forecast accuracy (R2 of 81.82–98%) for high-yield albacore fishing grounds and a poorer forecast accuracy (R1 of 47.37–96.15%) for low-yield fishing grounds were obtained for different months based on the ELM; the high-yield fishing grounds were distributed in the sea south of 10° S. (3) A feature importance analysis based on RF found that latitude (Lat) had the greatest influence on the forecast accuracy of albacore tuna fishing grounds of different grades from February to December (0.377), and Chl-a had the greatest influence on the forecast accuracy of albacore tuna fishing grounds of different grades in January (0.295), while longitude (Lon) had the smallest effect on the forecast of different grades of fishing grounds (0.037).

Feature Selection for Explaining Yellowfin Tuna Catch per Unit Effort Using Least Absolute Shrinkage and Selection Operator Regression

“…Additionally, Zhang et al [37] found that the seawater temperature at depths of 300 m and 150 m, the chlorophyll concentration, and the Southern Oscillation Index are also key environmental features influencing yellowfin tuna catch rates. This study further discovered that the temperature of the water layer at depths of 300 to 450 m significantly affects yellowfin tuna CPUE, possibly due to the close correlation between yellowfin tuna's vertical activity range and its concentration primarily at a depth of 300 m [1,38,39].…”