A new modeling strategy for the predictive model of chub mackerel (Scomber japonicus) central fishing grounds in the Northwest Pacific Ocean based on machine learning and operational characteristics of the light fishing vessels
Haibin Han,Chen Shang,Bohui Jiang,Yuhan Wang,Yang Li,Delong Xiang,Heng Zhang,Yongchuang Shi,Keji Jiang
DOI: https://doi.org/10.3389/fmars.2024.1451104
IF: 5.247
2024-10-15
Frontiers in Marine Science
Abstract:The chub mackerel ( Scomber japonicus ) is one of the most influential small pelagic fish in the Northwest Pacific Ocean, and accurate modeling approaches and model selection are critical points in predicting the Scomber japonicus fishing grounds. This study investigated the changes in catches and fishing days on no moonlight and bright moonlight days (2014-2022) and compared the differences in predictive performance between the LightGBM and RF models on three datasets under the two modeling approaches [those based on the light fishing vessels operational characteristics (Approach one) and those not (Approach Two)]. The results were as follows: 1) Stronger moonlight intensity (e.g., full moon) can limit the fishing efficiency of light fishing vessels, with most years showing a trend of a higher percentage of fishing days on bright moonlight days than catches percentage, i.e., no moonlight days resulted in higher catches with lower fishing days; 2) Compared to Modeling Approach Two, under Modeling Approach one, RF model achieved better predictive performance on dataset B, while the LightGBM model achieved better predictive performance on both datasets A and B; 3) Overall, the Approach One achieved more satisfactory prediction performance, with the optimal prediction performance on the complete dataset C improved from 65.02% (F1-score of the RF model, Approach Two) to 66.52% (F1-score of the LightGBM model, Approach Two); 4) Under the optimal modeling approach (Approach One) and the optimal model (LightGBM model), the differences in the importance of the variables on dataset A (no moonlight days) and dataset B (bright moonlight days) were mainly centered on the environmental variables, with CV, SLA, and SSS being the most important in dataset A, and CV, DO, and SLA being the most important in dataset B. This study provides a more scientific and reasonable modeling undertaking for the research of light purse seine fishing vessels, which is conducive to guiding fishermen to select the operating area and operating time of the Scomber japonicus fishery more accurately and comprehensively and realizing the balanced development of fisheries in terms of ecology and economy.
marine & freshwater biology