Elsevier

Fisheries Research

Volume 252, August 2022, 106361
Fisheries Research

Lagrangian characteristics in the western North Pacific help to explain variability in Pacific saury fishery

https://doi.org/10.1016/j.fishres.2022.106361Get rights and content

Abstract

A new model for estimation of daily probability for the Pacific saury (Cololabis saira) encounter was proposed. The model performance was tested for the period of 2004–2018 (August–November) using the data from the Russian vessel monitoring system. The following physical oceanographic variables were used for encounter probability prediction: the absolute values and gradients (∇) of speed (V) of passive particles, imitating water parcels, and Lagrangian indicators. The positive effects on the encounter probability of saury were found for V, ∇V, and for the gradient of the finite-time Lyapunov exponent (∇Λ), while the effect of particle path length was negative. That means that saury preferred places close to the boundaries of the oceanographic features, where Lagrangian fronts are situated, but not inside the features themselves, because Λ is small in regular flows and large at Lagrangiam fronts. The model did not include information about years and volume of saury catches, but its monthly mean of catch probability in September had the highest correlation with Russian annual catches outside the national waters between Russia and Japan (r = 0.76, p = 0.001) and total annual catches there (r = 0.73, p = 0.002). Timeseries analysis of principle components (PC) from daily predictions of saury catch probabilities has also shown that the third PC correlated highly with the annual biomass of saury (r ≥ 0.8, p < 0.05). The model seems to be useful to manage Russian fishery and may help to explain the reasons for the saury biomass decline. The latter is very important to take into account for development of the stock assessment models.

Introduction

The Pacific saury (Cololabis saira) is a short-lived species living only 1–2 years (Suyama et al., 1992, Suyama et al., 1996). Therefore, it is very sensitive to environmental fluctuations, making forecasts of biomass dynamics uncertain. Spatial distribution of that fish may also change very much from year to year. Japanese scientific surveys have shown a shift in the high concentrations of saury to the east (Hashimoto et al., 2020). The fishing efforts have shifted to the east and even north (https://www.npfc.int/science/gis/catch-effort/saury), making saury fishing places inaccessible to many non-autonomous vessels due to the limit of distance for each trip by small vessels and the cost of fuel for larger vessels. The distribution of saury has been changed recently causing associated changes in the distribution of the fishing fleet. Earlier most of those vessels operated in the national waters between Russia and Japan and covered the main core of fish density distribution almost totally until 2019 (Fig. S1). So, a development of a species distribution model (SDM) is an urgent need because predictions of the most probable fishing places are important for the fishery.

SDMs have become a go-to technique for mapping and predicting habitat use by mobile species, but the utility of SDMs in generating near real-time predictions for fisheries applications has not yet been adequately explored (Scales et al., 2017). We are interested in short-term (daily to weekly) forecasts.

Members of the NPFC try to account for a spatiotemporal variability of saury concentration during standardization of catches per unit of effort (CPUE) following the protocol, adopted by the NPFC, which was developed with consideration of published overview (Maunder, Punt, 2004). The adopted protocol for CPUE standardization in the NPFC also allows one to try machine learning methods which we are going to test here. The Members estimated environmental effects on saury catch distribution, using the “joint CPUE” dataset and Vector-Autoregressive Spatiotemporal model (Thorson, 2019), which have shown that neither a single local variable, such as sea surface temperature (SST), sea surface height (SSH) and concentration of chlorophyll-a or regional environmental variable (SOI or PDO indices) nor a combination of them could explain the Pacific saury distribution shift. It was attributed mostly to a unmodelled spatiotemporal variability due to random effects (Hsu et al., 2021).

Necessity of a short-term prediction of fishing grounds for saury and other fish is undoubtful for Russian fishermen, because almost all of them can switch the target and choose what to catch: the Pacific saury, Walleye pollock, Chub mackerel, Japanese sardine etc., depending on which fish is closer to the shore or mothership. This social demand is regularly published in Russian mass media. Russian fishermen know approximate seasonal locations of fish, so monthly forecasts are useless. Fishing masters try to interpret spatial environmental information on their own each day. So, the urgent need for daily prediction of fishing places is obvious for us, moreover TINRO scientists are asked to make daily predictions every year. Empirical relations may be considered as a demerit by scientists, but the end users do not care about underlying mechanisms of formation of favorable fishing grounds in the same way as ordinary people do not care about empirical relations used for the weather forecast.

Here we would like to test the Lagrangian approach that uses history of fluid particles. It is able to distinguish water masses better than instantaneous information about water properties. The notion of Lagrangian front has been introduced by Prants et al. (2014). It is defined as a strong gradient of Lagrangian indicators which are specific functions of water parcel’s trajectories that contain information about the history of motion of water masses during a given period of time. These diagnostics allow one to characterize water masses in the fishery region simulating their transport and mixing in the altimetric velocity field. An overview of the Lagrangian approach can be found in the book (Prants et al., 2017). The connection between locations of saury catches and strong Lagrangian fronts has already been found (Prants et al., 2012). The recent works by Prants et al., 2020, Prants et al., 2021 on saury aggregation at Lagrangian fronts were mainly descriptive. Those authors did not try to predict places of saury catches as we plan to do in this paper. Mechanisms of formation of favorable conditions for marine life at Lagrangian fronts has been overviewed recently (Prants, 2022), but there is a gap in estimation of suitability of those structures for daily prediction of fishing places. The novelty of our work is in quantitative assessment of predictive ability of Lagrangian fronts in saury fishing.

Section snippets

Overview

The modern SDMs for saury, known to us, are on a monthly scale of observation (Chang et al., 2019, Hua et al., 2020). That time scale is unacceptable for a short-term prediction of the fishing places in our case. It is worth to remember that the optimum ranges of predictors are not stationary. They may change in time. This, in turn, may be associated with the changes in physiological states of fish, making it practically impossible to select a single global and reliable indicator (Karedin, 1984

Results

Random Forest (RF) model, tuned on a full training set of uncorrelated features, had the highest AUC (0.86) value in the test after optimization of the hyperparameters. AUCs of other machine learning technics that we tried, BRTs, had the values less than 0.82 (Table S2). Importance of features was exceptionally low and six of them had the Altmann’s p value greater than 0.05 in the selected method, the RF (Table 2). We optimized RF hyperparameters again using only coordinates and month as

Discussion

Importance of spatiotemporal (j_lon2, j_lat2, month) predictors was the highest. The Lagrangian indicators could not explain the spatiotemporal variability on its own, but they could do it much better if common pattern of migration was captured by the explicit spatiotemporal components. Probably, the importance of spatial components is associated with topographic features of the seafloor, such as the continental shelf break. In the work like this, with the spatiotemporal scales equal to

Conclusion

The SDM for saury catches, tuned with the same method (RF) and spatiotemporal resolution of the data, performs better (AUC=0.85) using Lagrangian indicators than other water properties (AUC=0.71) used in our previous work (Kulik et al., 2019). The developed SDM here may be used for one day ahead forecasting of the most probable places for saury fishing, but the best performance is achieved when Lagrangian indicators are provided without lags. Therefore, a development of forecasting system for

CRediT authorship contribution statement

Vladimir V. Kulik: Formal analysis, Investigation, Methodology, Software, Data curation, Writing – original draft. Sergey V. Prants: Conceptualization, Investigation, Supervision, Project administration, Funding acquisition, Methodology, Resources, Writing – review & editing. Michael Yu. Uleysky: Software, Data curation, Visualization, Validation, Writing – review & editing. Maxim V. Budyansky: Software, Data curation, Resources, Writing – review & editing.

Declaration of Competing Interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Acknowledgements

The work of SVP, MVB and MYuU on Lagrangian simulation was supported by the POI FEBRAS Program (State Task No. 121021700341–2). The work of VVK was supported by the VNIRO State Task (No. 076–00005-20–02).

References (56)

  • N. Usui et al.

    Meteorological Research Institute multivariate ocean variational estimation (MOVE) system: some early results

    Adv. Space Res.

    (2006)
  • Q. Xing et al.

    Application of a fish habitat model considering mesoscale oceanographic features in evaluating climatic impact on distribution and abundance of Pacific saury (Cololabis saira)

    Progr. Oceano

    (2022)
  • Z. Zhang et al.

    Empirical Orthogonal Functions

  • A. Altmann et al.

    Permutation importance: a corrected feature importance measure

    Bioinform

    (2010)
  • D.W. Apley et al.

    Visualizing the effects of predictor variables in black box supervised learning models

    J. R. Stat. Soc. Ser. B Stat. Methodol.

    (2020)
  • A.A. Baitaluk et al.

    Usloviya formirovaniya promyslovyh skoplenij sajry u yuzhnyh Kuril’skih ostrovov [Condition of saury fishing grounds formation in different water masses in the area off south Kuril Islands]

    Izv. Tinro.

    (1998)
  • A. Bakun

    Fronts and eddies as key structures in the habitat of marine fish larvae: opportunity, adaptive response and competitive advantage

    Sci. Mar.

    (2006)
  • Barnes, R., 2020. dggridR: Discrete Global Grids....
  • L. Breiman

    Random forests

    Mach. Learn

    (2001)
  • Caruana, R., Karampatziakis, N., Yessenalina, A., 2008. An empirical evaluation of supervised learning in high...
  • Y. Chang et al.

    Modelling the impacts of environmental variation on habitat suitability for Pacific saury in the Northwestern Pacific Ocean

    Fish. Oceanogr.

    (2019)
  • C. Chen et al.

    Using random forest to learn imbalanced data (Report No. 666). Department of Statistics

    (2004)
  • R.B. Cleveland et al.

    STL: a seasonal-trend decomposition procedure based on loess

    J. . Stat.

    (1990)
  • G. De’ath

    Boosted trees for ecological modeling and prediction

    Ecol

    (2007)
  • M. Fernandez-Delgado et al.

    Do we need hundreds of classifiers to solve real world classification problems?

    J. Mach. Learn. Res.

    (2014)
  • M. Hashimoto et al.

    Comparison of biomass estimates from multiple stratification approaches in a swept area method for Pacific saury Cololabis saira in the western North Pacific

    Fish. Sci.

    (2020)
  • Jassby, A.D., Cloern, J.E., 2020. wql: Exploring water quality monitoring data....
  • Y.P. Karedin

    Kratkosrochnoe prognozirovanie na promyslah sajry i sardiny [Short-term forecasting during the fishery of saury and sardine]

    Izv. Tinro.

    (1984)
  • View full text