Korean Journal of Remote Sensing (대한원격탐사학회지)

Korean Society of Remote Sensing (대한원격탐사학회)
권호 표지
DOI QR코드

DOI QR Code

An Artificial Intelligence Method for the Prediction of Near- and Off-Shore Fish Catch Using Satellite and Numerical Model Data

  • Yoon, You-Jeong (Undergraduate Student, Department of Spatial Information Engineering, Pukyong National University) ;
  • Cho, Subin (Undergraduate Student, Department of Spatial Information Engineering, Pukyong National University) ;
  • Kim, Seoyeon (Undergraduate Student, Department of Spatial Information Engineering, Pukyong National University) ;
  • Kim, Nari (Postdoctoral Researcher, Geomatics Research Institute, Pukyong National University) ;
  • Lee, Soo-Jin (PhD Student, Major of Spatial Information Engineering, Division of Earth Environmental System Science, Pukyong National University) ;
  • Ahn, Jihye (PhD Student, Major of Spatial Information Engineering, Division of Earth Environmental System Science, Pukyong National University) ;
  • Lee, Eunjeong (Assistant Manager, Korea Environmental Science and Technology Institute) ;
  • Joh, Seongeok (President, Korea Environmental Science and Technology Institute) ;
  • Lee, Yang-Won (Professor, Department of Spatial Information Engineering, Pukyong National University)
  • Received : 2020.02.10
  • Accepted : 2020.02.14
  • Published : 2020.02.28
Download PDF
⟨ Previous Next ⟩

Abstract

The production of near- and off-shore fisheries in South Korea is decreasing due to rapid changes in the fishing environment, particularly including higher sea temperature in recent years. To improve the competitiveness of the fisheries, it is necessary to provide fish catch information that changes spatiotemporally according to the sea state. In this study, artificial intelligence models that predict the CPUE (catch per unit effort) of mackerel, anchovies, and squid (Todarodes pacificus), which are three major fish species in the near- and off-shore areas of South Korea, on a 15-km grid and daily basis were developed. The models were trained and validated using the sea surface temperature, rainfall, relative humidity, pressure,sea surface wind velocity, significant wave height, and salinity as input data, and the fish catch statistics of Suhyup (National Federation of Fisheries Cooperatives) as observed data. The 10-fold blind test results showed that the developed artificial intelligence models exhibited accuracy with a corresponding correlation coefficient of 0.86. It is expected that the fish catch models can be actually operated with high accuracy under various sea conditions if high-quality large-volume data are available.

Keywords

1. Introduction

In recent years, the total annual production of near and off-shore fisheries in South Korea could not reach one million tons for the first time, only recording 0.91 and 0.93 million tons in 2016 and 2017, respectively. Fishery resources are becoming scarce due to changes in the fishing environment,such as high water temperatures caused by global warming, overfishing of young fish, and the pollution of fish habitats due to the inflow of land pollutants to near- and off-shore areas (Cho, 2006). In addition, the increase in fishing costs due to rising labor and oil costs is decreasing the competitiveness of the near- and offshore fisheries. In these situations, the prediction of fish catch that spatiotemporally changes according to the sea state is crucial in maintaining and developing fisheries (Markridakis and Hyndman, 1974).

In previous studies, the relationships between fish catch and marine environmental factors were qualitatively analyzed (Kim et al., 2017; Song, 2018; Yoo et al., 2018). Regression analysis was performed using water temperature and zooplankton as explanatory variables to estimate CPUE (catch per unit effort) (National Institute of Fisheries Science, 1992; Hwang et al., 2012; Park, 2013). Fish catch has a complex and nonlinear relationship with various marine, meteorological, and biological factors in terms of time and space (Yoo and Zhang, 1993). Such complexity and nonlinearity have been considered lately, and attempts have been performed to predict the squid (Todarodes pacificus) catch for the SAU (sea area unit, approximately 50-km grid) as three categories (large, medium, and small) using artificial intelligence methods(Korea MeteorologicalAdministration, 2016) or to predict the monthly catch of mackerel and squid using deep learning (National Institute of Fisheries Science, 2018). These studies used various marine environmental data as input variables. However, monthly predictions on a 50-km grid are not sufficient information for fishing vessels which sail every day.

The development of daily catch prediction information in a finer spatial resolution will provide more precise data, and in practice, be more helpful for fisheries. The purpose of this study is to develop artificial intelligence models using marine meteorological variables and satellite data which affect fish catch as input data to predict the daily catch of mackerel, anchovies, and squid for the SSAU (small sea-area unit, approximately a 15-km grid). Fig. 1 shows the flow chart for the daily catch prediction of mackerel, anchovies, and squid, which are three major fish species in the near- and offshore areas of South Korea. Spatiotemporal matchup between the satellite, marine, and meteorological data and the fish catch statistics of Suhyup (National Federation of Fisheries Cooperatives) was performed, and three artificial intelligence models based on SVM (support vector machine), RF (random forest), and DNN (deep neural network) were trained and validated.

OGCSBN_2020_v36n1_41_f0001.png 이미지

Fig. 1. Flow chart for prediction of daily CPUE (catch per unit effort).

2. Theoretical Background

1) Ecology of near- and off-shore fish species

Near- and off-shore areas generally refer to shallow sea with a water depth of up to 200 m. Mackerel, anchovies, and squid are three major fish species that represent the highest proportions in near- and off-shore fisheries in South Korea. They are popular with high economic value. Mackerel is mostly distributed in tropical and temperate sea areas. As a fish species that swims near the sea surface, mackerel inhabits the sea surface within a 200 m water layer. For them, the habitat water temperature range is 7-25°C and the optimal water temperature range is 15-19°C. They mostly spawn in the surrounding waters of Jeju Island and Tsushima Island in April and May (Cha et al.,2002).Anchovies are warm-water fish distributed in all waters surrounding South Korea. They are coastal catadromous fish that live on the continental shelf up to 20 m water depth. For them, the habitat water temperature ranges from 8 to 30°C and the optimal water temperature ranges from 13 to 23°C. Their main spawning season is from May to July, and spawning and growth occur mostly on the southern coast (National Institute of Fisheries Science, 2005). Squid is a warm-water catadromous fish that lives within approximately 200 m from the sea surface. It is an annual fish species that dies after spawning (Roper et al., 1969). Its habitat water temperature range is 4- 27°C and the optimal water temperature range is 12- 18°C. It mostly spawns near the seabed in the shallow waters of the East China Sea and the southern part of the East Sea.

2) Sea state variables affecting fish catch

The sea state refers to the state of the sea that combines marine and meteorological factors that affect fish catch, such as water temperature, salinity, seawater flow, nutrient salts, and plankton. In near- and off-shore areas, the variability in sea state is larger compared to the deep sea. Such physical, chemical, and biological properties of seawater are very important for estimating fish catch (Tom, 1983).

(1) Water temperature

The fifth report of the IPCC (Intergovernmental Panel on Climate Change) predicted that the air, ocean temperatures, and sea level of the earth would continuously rise (CoreWritingTeamet al., 2014). Such climate change is prone to affect marine ecosystems, thereby causing changes in fish species composition, fish habitats, and fish catch (Cochrane et al., 2009; Lu and Lee, 2014). Over the last 100 years (1918- 2017), the air temperature and sea surface temperature increased by 1.55°C and 0.62°C on average globally. In addition, the sea surface temperature of South Korea has increased by approximately 1.1°C over the last 50 years. It has also increased by 1.7°C in the East Sea, 1.4°C in the South Sea, and 0.3°C in the West Sea (Statistics Korea, 2018). These changes in water temperature have increased warm-water fish species and decreased cold-water fish species in the near- and off-shore areas of South Korea since 1990.

(2) Salinity

All marine organisms live in saline water and control osmotic pressure. Salinity is subject to spatiotemporal changes depending on the influence of wind and rainfall (Thorson, 1971). The mackerel catch tended to increase when salinity was between 32 and 34.4‰ (Lee, 2009), and the anchovy catch tended to increase when salinity was between 31.5 and 34‰ (Seo, 1999). As different fish species have different salinities suitable for habitation and spawning, it is necessary to analyze the spatial distribution of salinity and identify its change patterns for predicting fish catch.

(3) Currents and waves

Wind generates surface currents and waves due to its friction with the sea surface (Tom, 1983). For catadromous fish that move in a certain sea according to the season, such as mackerel, anchovies, and squid, the production varies depending on the surface current patterns. If the wind blows continuously in a constant direction over the sea surface of a coastal area, upwelling may occur due to the Ekman spiral. This phenomenon transports deep water with abundant nutrient salts and affects the growth of phytoplankton. It may also contribute to the fishery formation by inducing fish clusters (Ryther, 1969; Cho et al., 2003; Williamson et al., 2009). In this study, the sea surface wind velocity and significant wave height, which is the average of the values that correspond to the upper third of the total wave height, were used as input data.

(4) Air-sea interaction

On the sea surface between the atmosphere and the sea, the exchange of heat, water, and momentum affect the physical phenomena of both the atmosphere and the sea. The pressure gradient force generated by the pressure difference is a factor that generates ocean currents. Water exchange by evaporation is an important factor that affects heat exchange. The atmospheric pressure, relative humidity, and rainfall are variables that can be considered in the analysis of fish catch (Nam and Noh, 2012; Song, 2013).

3. Study Areas and Data

1) Study areas

Among the fisheries of mackerel, anchovies, and squid, representative SAUs were selected as study areas (Fig. 2). SAU 224, 232, 233, and 234 in the vicinity of Jeju Island were selected for mackerel, and SAU 97, 98, 104, and 105 near the South Sea were selected for anchovies. SAU 76, 82, 87, and 93 on the east coast were selected for squid. One SAU was again composed of 3×3 SSAUs. From 2014 to 2016, daily fish catch data were collected for each SSAU. Mackerel are mostly found near Jeju Island in autumn and winter, and anchovies generally stay near the South Sea from spring to autumn. Squid is mostly observed on the east coast from July to November.Therefore, the frequency of seasonal data was constructed with differences depending on the fish species.

OGCSBN_2020_v36n1_41_f0002.png 이미지

Fig. 2. Study area for fish catch prediction by species.

2) Numerical weather prediction data

The numerical weather prediction model of KMA (Korea Meteorologica Administration)includes global, regional, and local models as shown in Table 1 (http://www.kma.go.kr/aboutkma/intro/supercom/model/model _category.jsp). Among them, the RDAPS (Regional Data Assimilation and Prediction System) data with a 12-km grid, which is most similar to the SSAU 15-km grid, was used. RDAPS receives boundary conditions from the global model every three hours, and performs predictions four times a day (00, 06, 12, and 18 UTC) through 4-dimensional variational data assimilation. The RDAPS data are provided in the GRIB2 (Gridded Binary 2) format, a WMO (World Meteorological Organization)standard. The pressure, relative humidity, and rainfall data required for CPUE prediction were extracted.

Table 1. Data Assimilation and Prediction System by Korea Meteorological Administration

OGCSBN_2020_v36n1_41_t0001.png 이미지

Table 2. Wave models by Korea Meteorological Administration

OGCSBN_2020_v36n1_41_t0002.png 이미지

3) Regional wave model data

The wave model of KMA provides forecast data, such as on significant wave height, wave aspect, and wave period, for global, regional, and regional coasts as shown in Table 2 (http://www.kma.go.kr/aboutkma/intro/supercom/model/model_category.jsp). The RWW3 (Regional Wave Model) with an 8-km grid, which is most similar to the SSAU 15-km grid, performs predictions twice a day (00 and 12 UTC). From the model, the wind speed and significant wave height variables were extracted and used.

4) Sea surface temperature data

The OSTIA (Operational Sea Surface Temperature and Sea Ice Analysis) data of the GHRSST (Group of High-Resolution Sea Surface Temperature) are provided as a daily mean-field of 6 km resolution by combining the SST and in-situ buoy data obtained from the following sensors: AMSR-E (Advanced Microwave Scanning Radiometer-EOS) of the Aqua satellite, AATSR (Advanced Along-Track Scanning Radiometer) of the EnviSat satellite, SEVIRI(Spinning Enhanced Visible and Infrared Imager) of the MSG (MeteoSat Second Generation) satellite, AVHRR-3 (Advanced Very High-Resolution Radiometer-3) of the NOAA-16, NOAA-17, and NOAA-18 satellites, and TMI (TRMM Microwave Imager) of the TRMM(Tropical Rainfall Measuring Mission)satellite (https:// www.metoffice.gov.uk).

5) Ocean Grid in-situ data

The Ocean Grid service of the KHOA (Korea Hydrographic and Oceanographic Agency) (http:// www.khoa.go.kr/oceangrid) provides ocean observation data through an API (open application programming interface). Among them, the 1-minute salinity data of Hupo, Tongyeong, Yeosu, and Seongsanpo, which are monitoring stations close to the study areas, were acquired.

6) Data preprocessing and spatiotemporal matching

Since the input data have different spatiotemporal resolutions as shown inTable 3, data preprocessing for spatial and temporal matching is required. The temporal resolution was matched based on the daily mean to predict daily fish catch. For the salinity data, 1-minute data were averaged for 24 hours. For the RDAPS and RWW3, data provided twice or four times a day were averaged. For spatial matching, the grid points of the RDAPS, RWW3, and OSTIA data were matched using the nearest neighbor method based on the SSAU 15- km grid. The point data from KHOA Ocean Grid observation were also matched to the closest cell of SSAU.

Table 3. Summary of data used in this study

OGCSBN_2020_v36n1_41_t0003.png 이미지

Fish catch data, which are records reported by fishing vessels, were provided by Suhyup. The data have high reliability in terms of locations, but the fish catch report may be often missing (Korea Meteorological Administration, 2016). CPUE is total fish catch divided by the total effort, and it was expressed in ton/boat in this study.

4. Analysis Methods

Fish catch was predicted using the artificial intelligence models based on SVM, RF, and DNN, which used CPUE as a response variable and the sea surface temperature, salinity, pressure, relative humidity, rainfall, sea surface wind velocity, and significant wave height as explanatory variables. The results were compared and evaluated.

1) Support vector machine

SVM is a machine-learning algorithm for optimal grouping of data, and can be combined with a regression model for optimal grouping (Pal and Mather, 2003; Jensen et al., 2009; Duro et al., 2012). A hyperplane is used for optimal grouping of data (Mountrakis et al., 2011; Foody and Mathur, 2006;Van der Linden and Hostert, 2009) by maximizing the margin between groups. The MMH can be found by maximizing the margin between support vectors at the boundary of the data groups(Meyer, 2014) using kernel functions, such as linear and Gaussian RBF (radial basis function) (Fig. 3). We used a Gaussian RBF as the kernel function in this experiment.

2) Random forest

RF, which represents an improved version of CART (classification and regression trees), is an ensemble method that uses the bootstrap and bagging process(Breiman andCutler, 2014).It generates a large number of decision trees with slightly different features by extracting random samples from the training data, including a bootstrap that determines the suitability of the sampling distribution, and conducts resampling as needed. During the bagging process, bootstrap-based decision trees are aggregated to create a final solution using an ensemble method, such as the average (or majority vote) (Breiman, 2001; Ali et al., 2012). The final decision tree can be optimized by the tree pruning algorithm to determine an optimal size for the trees. In our experiment, the number of trees was set to 500, and the number of variables used for splitting nodes was set to n/3 (n = number of input variables). In addition, the out-of-bag error was used as the criterion of model suitability (Fig. 4).

OGCSBN_2020_v36n1_41_f0003.png 이미지

Fig. 3. Examples of data grouping using the SVM (support vector machine) with a (a) linear function and (b) Gaussian radial basis function.

OGCSBN_2020_v36n1_41_f0004.png 이미지

Fig. 4. Conceptual framework of RF (random forest).

3) Deep neural network

The classic neural network method has a local minima problem in which an optimization process often stops at a locally, rather than globally, optimized state. In addition, generic machine learning methods sometimes have problems with overfitting, in which they cannot handle data with outliers due to excessive learning from the given dataset. Such problems can be resolved by DNNs through an intensive optimization process in a deep network structure (Fig. 5). To handle local minima and issues with overfitting, L1/L2 regularization can be performed to ensure sparsity (L1) and simplicity (L2) of the DNN model. Also, backward and forward optimization is conducted in the backpropagation algorithm to improve accuracy. The problem of vanishing gradients of loss functions, which may occur during the back-propagation process, can be managed by applying appropriate activation functions, such as sigmoid and RELU (rectified linear unit). The drop-out method deals with unexpected outliers via a learning mechanism, in which the DNN model becomes more robust to extreme cases through iterations of a type of handicapped training with randomly deleted links and nodes (Pham et al., 2014). In addition, a weight and bias set built in an existing DNN model can be imported as an initial value of a new DNN model for more custom-tailored training. This is called pre-training and transfer learning, which can improve the optimization of a DNN model. Finetuning can also be incorporated into the optimization process to adjust the weight and biasset in more detail, by including additional training data (Erhan et al., 2010). In our experiment, the hidden units consisted of 200-200-200, and the number of epochs was set to 300. We usedAdaDelta (adaptive delta)for an optimizer and RELU for an activation function.

OGCSBN_2020_v36n1_41_f0005.png 이미지

Fig. 5. Structure of DNN (deep neural network) model for this study.

4) Training and validation

A10-fold blind test method was used for the training and validation of artificial intelligence models. First, 4,113 spatiotemporal matchups between 2014 and 2016 were used in the SVM, RF, and DNN models for mackerel, anchovies, and squid to construct a total of nine models, and the 10-fold blind test results for each model were collected. The MBE (mean bias error), MAE (mean absolute error), RMSE (root mean square error), and CC (correlation coefficient) were used as validation statistics.

Table. 4. Multicollinearity of explanatory variables by fish species

OGCSBN_2020_v36n1_41_t0004.png 이미지

5. Results and Discussion

As multicollinearity among the explanatory variables may cause erratic results in response to small changes in the model or the data (Kim et al., 2018), it was checked using the VIF (variation inflation factor) before modeling. In general, aVIF > 10 indicates the presence of multicollinearity. All of the seven explanatory variables mentioned above were adopted in this study because there was no multicollinearity among them (Table 4).

The 10-fold blind test using the matchups between 2014 and 2016 showed that the DNN models exhibited relatively higher performance among the fish catch prediction models of mackerel, anchovies, and squid (Tables 5, 6, and 7). The correlation coefficients of the DNN models were 0.745 for mackerel, 0.864 for anchovies, and 0.842 for squid. The models exhibited high accuracy considering that they performed daily predictions for the SSAU 15-km grid. The RMSE of mackerel was higher than those of anchovies and squid. It may be implied that CPUE of mackerel were much larger than those of anchovies and squid, and thus the magnitude of the error was relatively higher. For the DNN models that exhibited the highest performances, the scatter plots of the observed and predicted values are shown in Fig. 6, 7, and 8. The blind test results of mackerel, anchovies, and squid showed that most of the cases were concentrated on the one-to-one lines in spite of slight overestimation or underestimation tendencies. Vertically arranged points occasionally observed in the scatterplots may be due to inescapable noise in the observed values. To mitigate such phenomenon, outlier detection techniques such as Cook’s distance test can be employed to find out the records which may distort the relationship between explanatory and response variables.

Table 5. Validation statistics of SVM, RF, and DNN models

OGCSBN_2020_v36n1_41_t0005.png 이미지

Table 6. Validation statistics of SVM, RF, and DNN models for anchovies

OGCSBN_2020_v36n1_41_t0006.png 이미지

Table 7.Validation statistics of SVM, RF, and DNN models

OGCSBN_2020_v36n1_41_t0007.png 이미지

OGCSBN_2020_v36n1_41_f0006.png 이미지

Fig. 6. Observed and predicted mackerel catches for DNN model in 10-fold blind tests.

OGCSBN_2020_v36n1_41_f0007.png 이미지

Fig. 7. Observed and predicted anchovy catches for DNN model in 10-fold blind tests.

OGCSBN_2020_v36n1_41_f0008.png 이미지

Fig. 8. Observed and predicted squid catches for DNN model in 10-fold blind tests.

6. Conclusion

In this study, the daily catch of mackerel, anchovies, and squid, which are three major fish species in the near- and off-shore areas of South Korea, were predicted and validated on the SSAU 15-km grid through artificial intelligence modeling with marine, meteorological, and satellite data. While previous studies predicted the fish catch on the SAU 50-km grid or monthly basis, this study is significant in that it further improved temporal and spatial resolutions. In addition, the correlation coefficient of the validation statistics ranged from 0.745 to 0.864, indicating excellent accuracy. Approximately 4,000 fish catch data between 2014 and 2016 were available for our experiment. However, it was not sufficient for artificial intelligence modeling, and some data might include inescapable noise. In the future, it is expected that the accuracy of fish catch prediction models will be further improved if higher-quality large-volume data are available. In South Korea, fishery resources have been managed since 1999 by implementing the TAC (total allowable catch) system. If the prediction of fish catch becomes reliable, it will be more efficient to implement the TAC system. It is also expected to be used as valuable information in various directions, such as base data for estitmating the amount of loss for the case of damage to a fishery. Moreover, an application on the mobile web can help practical use of the prediction information for the people in the fishery industries.

Acknowledgements

This research was funded by the LINC Project(2018and 2019) supported by Pukyong National University and Ministry of Education. The authors acknowledge the assistance of Ui-Hwan Bae and Hyunjung Choi for helping data collection.

References

  1. Ali, J., R. Khan, N. Ahmad, and I. Maqsood, 2012. Random forests and decision trees, International Journal of Computer Science Issues, 9(5): 272-278.
  2. Breiman, L., 2001. Random forests, Machine Learning, 45(1): 5-32. https://doi.org/10.1023/A:1010933404324
  3. Breiman, L. and A. Cutler, 2014. Random forests, http://www.stat.berkeley.edu/-breiman/RandomForests/cc_home.htm, Accessed on Dec. 15, 2018.
  4. Cha, H.K., Y.M. Choi, J.H. Park, J.Y. Kim, and M.H. Sohn, 2002. Maturation and spawning of the chub Mackerel, Scomber japonicus Houttuyn in Korean Waters, Journal of the Korean Society of Fisheries Resources, 5: 24-33 (in Korean with English abstract).
  5. Cho, K.D., D.S. Kim, and S.E. Park, 2003. Characteristics of oceanographic conditions in an area suitable for the construction of artificial upwelling, Korean Journal of Fisheries and Aquatic Sciences, 36(2): 187-192 (in Korean with English abstract). https://doi.org/10.5657/kfas.2003.36.2.187
  6. Cho, Y., 2006. Time series analysis of general marine fisheries, Journal of Rural Development/Nongchon-Gyeongje, 29(1): 123-134 (in Korean with English abstract).
  7. Cochrane, K., C. De Young, D. Soto, and T. Bahri, 2009. Climate change implications for fisheries and aquaculture, FAO (Food and Agriculture Organization of the United Nations) Fisheries and Aquaculture Technical Paper, 530: 212.
  8. Core Writing Team, R.K. Pachauri, and L.A. Meyer, 2014. Climate Change 2014: Synthesis Report, Contribution of Working Groups I, II and III to the Fifth Assessment Report of the Intergovernmental Panel on Climate Change, IPCC, Geneva, Switzerland.
  9. Duro, D.C., S.E. Franklin, and M.G. Dube, 2012. A comparison of pixel-based and object-based image analysis with selected machine learning algorithms for the classification of agricultural landscapes using SPOT-5 HRG imagery, Remote Sensing of Environment, 118: 259-272. https://doi.org/10.1016/j.rse.2011.11.020
  10. Erhan, D., Y. Bengio, A. Courville, P.A. Manzagol, and P. Vincent, 2010. Why does unsupervised pre-training help deep learning?, Journal of Machine Learning Research, 11: 625-660.
  11. Foody, G.M. and A. Mathur, 2006. The use of small training sets containing mixed pixels for accurate hard image classification: Training on mixed spectral responses for classification by SVM, Remote Sensing of Environment, 103(2): 179-189. https://doi.org/10.1016/j.rse.2006.04.001
  12. Hwang, K.S., S.K. Kang, T.Y. Oh, K.H. Choi, and D.W. Lee, 2012. Change in the Fishing Grounds and the Relationship between the Abundance of the Common Squid Todarodes pacificus and the Distribution of Zooplankton in the East Sea, Korean Journal of Fisheries and Aquatic Sciences, 45(2): 173-179 (in Korean with English abstract). https://doi.org/10.5657/KFAS.2012.0173
  13. Jensen, J.R., J. Im, P. Hardin, and R.R. Jense, 2009. Image Classification, In: Warner, T.A., Nellis, M.D., Foody, G.M. (Ed.), The Sage Handbook of Remote Sensing, SAGE Publications, Thousand Oaks, CA, USA, pp. 269-296.
  14. Kim, J., J. Park, C. Choi, and H.S. Kim, 2018. Development of regression models resolving high-dimensional data and multicollinearity problem for heavy rain damage data, Journal of The Korean Society of Civil Engineers, 38(6): 801-808 (in Korean with English abstract). https://doi.org/10.12652/KSCE.2018.38.6.0801
  15. Kim, S.W., J.S. Ahn, and Y. Lee, 2017. The Effect of Interannual Variations in Water Temperature on the Yellowtail Catch, Seriola Quinqueradiata, in the Eastern Part of the Korean Peninsula, Journal of the Korean Society of Marine Environment & Safety, 23(7): 909-917 (in Korean with English abstract). https://doi.org/10.7837/kosomes.2017.23.7.909
  16. Korea Meteorological Administration, 2016. Fusion services for big data in weather-disaster prevention, fisheries, https://data.go.kr/information/PDS_0000000000000492/extrl/recsroom.do, Accessed on Aug. 5, 2018.
  17. Lee, H.N., 2009. Catch and oceanographic characteristics for large purse seine fisheries, M. S. Thesis, Pukyong National University, Busan, Korea (in Korean with English abstract).
  18. Lu, H.J. and H.L. Lee, 2014. Changes in the fish species composition in the coastal zones of the Kuroshio current and China coastal current during periods of climate change: Observations from the set-net fishery (1993-2011), Fisheries Research, 155: 103-113. https://doi.org/10.1016/j.fishres.2014.02.032
  19. Markridakis, S. and R.J. Hyndman, 1974. Forecasting: methods and Applications, Operational Research Quarterly, 25(4): 648-649. https://doi.org/10.1057/jors.1974.117
  20. Meyer, D., 2014. Support vector machines, http://cran.rproject.org/web/packages/e1071/vignettes/svmdoc.pdf, Accessed on Jan. 20, 2019.
  21. Mountrakis, G., J. Im, and C. Ogole, 2011. Support vector machines in remote sensing: A review, ISPRS Journal of Photogrammetry and Remote Sensing, 66(3): 247-259. https://doi.org/10.1016/j.isprsjprs.2010.11.001
  22. Nam, J. O. and S.G. Noh, 2012. A study on forecast of Oyster production using time series models, Ocean and Polar Research, 34(2): 185-195 (in Korean with English abstract). https://doi.org/10.4217/OPR.2012.34.2.185
  23. National Institute of Fisheries Science, 1992. A study on the improvement of forecasting techniques on oceanographic and fishing conditions by application of remotely sensed data (III), http://www.ndsl.kr/ndsl/search/detail/report/reportSearchResultDetail.do?cn=TRKO200200006173, Accessed on Nov. 10, 2018.
  24. National Institute of Fisheries Science, 2005. The ecology and fishing grounds of coastal and offshore fish species in Korea, http://www.ndsl.kr/ndsl/search/detail/report/reportSearchResultDetail.do?cn=TRKO201800015682, Accessed on Aug. 5, 2018.
  25. National Institute of Fisheries Science, 2018. Major species of fish change analysis and prediction using deep learning/space statistics, http://www.google.co.kr/url?sa=t&rct=j&q=&esrc=s&source=web&cd=2&ved=2ahUKEwjsse3wyO7iAhXKIqYKHUEKAdkQFjABegQIABAC&url=http%3A%2F%2Fwww.nifs.go.kr%2Fdownload%3Fboard_idx%3D43694%26site_id%3Dnfrdi%26code_flag%3Dbbs%26code_rname%3Dtender%26file_idx%3D165061&usg=AOvVaw0PloNxkWEuI7M8vtNlZEYX, Accessed on Sep. 5, 2018.
  26. Pal, M. and P.M. Mather, 2003. An assessment of the effectiveness of decision tree methods for land cover classification, Remote Sensing of Environment, 86(4): 554-565. https://doi.org/10.1016/S0034-4257(03)00132-9
  27. Park, J., 2013. Correlation between codfishes (Gadus macrocephalus) catch and the water temperature, and estimated in Yeosu seas, M.S. Thesis, Chonnam National University, Gwangju, Korea (in Korean with English abstract).
  28. Pham, V., T. Bluche, C. Kermorvant, and J. Louradour, 2014. Drop-out improves recurrent neural networks for handwriting recognition, Proc. of 14th International Conference on Frontiers in Handwriting Recognition, Island of Crete, Greece, Sep. 1-4, pp. 285-290.
  29. Roper, C.F., R.E. Young, and G.L. Voss, 1969. An illustrated key to the families of the order Teuthoidae (Cephalopoda), Smithsonian Contributions to Zoology, 13: 1-32. https://doi.org/10.5479/si.00810282.13
  30. Ryther, J.H., 1969. Photosynthesis and fish production in the sea, Science, 166(3901): 72-76. https://doi.org/10.1126/science.166.3901.72
  31. Seo, Y. and D.S. Kim, 1999. Distribution of anchovy school catched by the lift net and environmental factors in the Kamak Bay. 1. Relation between distribution of anchovy school and temperature and salinity, Journal of the Korean Society of Fisheries and Ocean Technology, 35(3): 267-276 (in Korean with English abstract).
  32. Song, H., 2018. Fluctuations of common squid Todarodes pacificus catches in the Northwestern Pacific under changing climate and habitat temperature, Korean Journal of Fisheries and Aquatic Sciences, 51(3): 338-343 (in Korean with English abstract). https://doi.org/10.5657/KFAS.2018.0338
  33. Song, J., 2013. Entrainment mechanism of Todarodes pacificus winter cohort into the Yellow Sea, M. S. Thesis, Korea Maritime and Ocean University, Busan, Korea (in Korean with English abstract).
  34. Statistics Korea, 2018. Changes in catches of major fish species according to climate (water temperature), http://kostat.go.kr/portal/korea/kor_nw/1/1/index.board?bmode=read&aSeq=368502&pageNo=28&rowNum=10&amSeq=&sTarget=title&sTxt=OECD, Accessed on Jul. 17, 2018.
  35. Thorson, G., 1971. The Sea as a Habitat, Life in the Sea, McGraw-Hill, New York, NY, USA, pp. 20-45.
  36. Tom, B., 1983. Coastal Oceanography, Waves, Environ - mental Oceanography: An Introduction to the Behavior of Coastal Waters, Pergamon Press, Oxford, UK, pp. 1-28, 59-86.
  37. Van der Linden, S. and P. Hostert, 2009. The influence of urban structures on impervious surface maps from airborne hyperspectral data, Remote Sensing of Environment, 113(11): 2298-2305. https://doi.org/10.1016/j.rse.2009.06.004
  38. Williamson, N., A. Komiya, S. Maruyama, M. Behnia, and S. W. Armfield, 2009. Nutrient transport from an artificial upwelling of deep sea water, Journal of Oceanography, 65(3): 349-359. https://doi.org/10.1007/s10872-009-0032-x
  39. Yoo, J.T., Y.H. Kim, S.H. Song, and S.H. Lee, 2018. Characteristics of egg and larval distributions and catch changes of anchovy in relation to abnormally high sea temperature in the South Sea of Korea, Journal of the Korean Society of Fisheries and Ocean Technology, 54(3): 262-270 (in Korean with English abstract). https://doi.org/10.3796/KSFOT.2018.54.3.262
  40. Yoo, S. and C. I. Zhang, 1993. Forecasting of Hairtail (Trichiurs lepturus) landings in Korean waters by times series analysis, Korean Journal of Fisheries and Aquatic Sciences, 26(4): 363-368 (in Korean with English abstract).