Time-lag Effect Of Skipjack Tuna(Katsuwonus Pelami) Habitat Distribution In The Western And Central Pacific Ocean

Skipjack tuna(Katsuwonus pelami)stocks are abundant and widely distributed in the tropical pelagic waters of the Pacific,Atlantic and Indian Oceans,and are the main target of purse-seine tuna fisheries in the western and central Pacific Ocean,with high commercial value.The spatial distribution and abundance dynamics of skipjack tuna populations are closely related to the marine environment,and traditional studies of skipjack tuna habitat have generally assumed that the influence of the marine environment on the abundance of tuna populations is immediate.The study of time lag effects on the distribution of bonito habitats provides a scientific basis for understanding the mechanisms driving spatial and temporal changes in bonito habitats.In this paper,a two-part study was conducted on the basis of seine fishery data and marine environmental data in the western and central Pacific Ocean from 2013 to 2020:in the first part,a distributed lag linear model(DLL)was constructed by selecting catch(CATCH),fishing activity(ACT)and effort per unit of catch(CPUE)as the abundance indicators of bonito and chlorophyll concentration a(CHL-a),which characterizes primary productivity.Distributed Lag Linear Models(DLLM)were constructed to discuss the time-lagged effects of primary productivity on the abundance dynamics of bonito populations at multiple spatial and temporal scales;the second part of the study firstly used the Elastic Net(EN)model to select environmental features that are important in explaining spatial and temporal variation and abundance of bonito habitats,and then constructed the models with lags of 0-1 day,0-5 day,0-1 day,0-1 day,0-5 day,0-1 day and 0-1 day.The study then constructed a Long-Short Term Memory(LSTM)model with four kinds of lag time,including 0-1 day,0-5 days,0-10 days and 0-15 days,combined with a random grid search to determine the optimal hyperparameters,and used the evaluation index to determine the optimal input step for the LSTM,completed the construction of the optimal habitat distribution prediction model and predicted the habitat distribution of the western and central Pacific booby in 2020.The prediction results were compared with the actual fishery distribution to verify the prediction effect of the LSTM model,and the results of the study are as follows.(1)The results of the DLLM model show that there are significant lagged relationships between CHL-a and the three abundance indicators,and there is no significant pattern in the effect of the choice of different temporal scales on the lag time of population abundance in response to changes in the biological environment;the choice of different spatial scales affects the lagged relationship between abundance and the biological environment,and it is hypothesized that when a spatial scale of 0.25°×0.25° is chosen in the habitat study,the lagged effect of primary productivity on abundance was not significant and the lagged effect time was about 0 days when the spatial scale of0.5°×0.5°,1°×1°,2°×2°,3°×3°,4°×4° and 5°×5° was chosen,and the lagged time interval of the effect of primary productivity on the abundance of bonito population was 0-2 days,1-4 days,2-6 days,3-7 days,5-10 days and 7-11 days respectively.Therefore,the study concluded that there is indeed a lagged effect of primary productivity on the change in abundance of bonito populations and the extent of the lagged effect is closely related to the spatial size of the area.(2)The results of the EN model showed that 14 environmental variables were important for the spatial and temporal distribution of bonito habitat abundance,including chlorophyll-a concentration(CHL-a),primary productivity(NPP),(SST)ST in the surface layer,NPP,salinity(SS)in the 100 m layer,dissolved oxygen concentration(DO),ST,SS in the 150 m layer,DO,ST,SS in the 200 m layer,DO,ST and ground-transfer flow velocity(GCV),as well as mixed layer thickness(MLD)and sea surface height(SSH).(3)The results of the grid-based random search for hyperparameter tuning showed that with a spatial scale of 0.25°×0.25°,two hidden layers set,an input time step of lag 0-5 days,the number of neurons in the input and hidden layers set to(32,32,32),the learning rate adjusted to 0.001,and the L1 and L2 regularization parametrization set to0.0001,the LSTM model The best fit was achieved with model MSE value of 0.055,MAE value of 0.179 and model R2 of 33%,indicating that there is a lagged effect of marine environment on habitat distribution at a small spatial scale,and the lagged effect has a short duration.(4)The LSTM model prediction results show that the predicted distribution of bonito habitat in 2020 ranges from 5°S-0°N,165°E-180°E,which is basically consistent with the distribution of the actual fishing grounds from 7.5°S-5°N,150°E-180°E;and the latter half of 2020 is the period of La Ni?a event,the predicted distribution of habitat in the third and fourth quarters compared with the first and second quarters The predicted distribution in the third and fourth quarters shows an upward westward trend compared to the first and second quarters,which is consistent with the spatial variability of the bonefish under the La Ni?a climate anomaly.