Lagrangian characteristics in the western North Pacific help to explain variability in Pacific saury fishery
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)
- et al.
An algorithm for oceanic front detection in chlorophyll and SST satellite imagery
J. Mar. Syst.
(2009) - et al.
A multi-scale high-resolution analysis of global sea surface temperature
Remote Sens. Environ.
(2017) - et al.
Evaluating the spatiotemporal dynamics of Pacific saury in the Northwestern Pacific Ocean by using a geostatistical modelling approach
Fish. Res.
(2021) - et al.
Habitat suitability of Pacific saury (Cololabis saira) based on a yield-density model and weighted analysis
Fish. Res.
(2020) - et al.
Standardizing catch and effort data: a review of recent approaches
Fish. Res.
(2004) The Astronomical Almanac’s algorithm for approximate solar position (1950-2050)
Sol. Energy
(1988)Marine life at Lagrangian fronts
Prog. Oceano
(2022)- et al.
Lagrangian study of transport and mixing in a mesoscale eddy street
Ocean Model
(2011) - et al.
Identifying Lagrangian fronts with favourable fishery conditions
Deep Sea Res. Part I Oceanogr. Res. Pap.
(2014) Guidance for decisions using the Vector Autoregressive Spatiotemporal (VAST) package in stock, ecosystem, habitat and climate assessments
Fish. Res.
(2019)