Applications of Machine Learning in Chemical and Biological Oceanography

Abstract

Machine learning (ML) refers to computer algorithms that predict a meaningful output or categorize complex systems based on a large amount of data. ML is applied in various areas including natural science, engineering, space exploration, and even gaming development. This review focuses on the use of machine learning in the field of chemical and biological oceanography. In the prediction of global fixed nitrogen levels, partial carbon dioxide pressure, and other chemical properties, the application of ML is a promising tool. Machine learning is also utilized in the field of biological oceanography to detect planktonic forms from various images (i.e., microscopy, FlowCAM, and video recorders), spectrometers, and other signal processing techniques. Moreover, ML successfully classified the mammals using their acoustics, detecting endangered mammalian and fish species in a specific environment. Most importantly, using environmental data, the ML proved to be an effective method for predicting hypoxic conditions and harmful algal bloom events, an essential measurement in terms of environmental monitoring. Furthermore, machine learning was used to construct a number of databases for various species that will be useful to other researchers, and the creation of new algorithms will help the marine research community better comprehend the chemistry and biology of the ocean.

Figure 1

Use of ML algorithms to identify, predict, and classify marine phytoplankton from images: (a) An underwater video plankton (VPR) system towed by a ship moved side, up, and down to capture plankton images, and the video recording was analyzed onboard by LVQ (extract image, identify major taxa and its distribution). Reprinted in part with permission from ref (57). Copyright 2004, Inter Research Science. (b) Plankton taxa are grouped in automatic classification with their taxonomic and morphological classes, followed by the comparison manual and automatic classification accuracy for living particles (size range 3–100 μm). The dot and solid line represent the autoimage, and square and dashed line represent fluorescence-triggered samples. Adapted and reprinted in part with permission from ref (61). Copyright 2012, Oxford University Press. (c) Sipper images classified using SVR. (d) Classification performance of VGG16 represented in the boxplot; the red diamonds indicate the mean value, and outliers are indicated by the black dots. Reprinted with permission from ref (72). Copyright 2020, Springer Nature. (e) Use of simple CNN to classify FlowCAM and FlowCytobot images. (f) Plankton classification using YOLOV3.

Figure 2

Use of ML algorithms to measure and estimate the ocean surface chlorophyll-a from satellite data. (a) Prediction of Chl-a using SVR and (b) multilayer perceptron neural network. (c) SVR model prediction for the hourly spatial distribution of chlorophyll-a concentration on the west coast of South Korea. Reprinted in part with permission from ref (76). Copyright 2014, Taylor & Francis. (d) The overall workflow for the prediction of Chl-a using CNN. Reprinted in part with permission from ref (82). Copyright 2021, MDPI.

Figure 3

ML algorithms are used to identify or classify zooplankton through images captured by various underwater devices. (a) Complete workflow for identification of ZP, starting with image segregation, training, and classification by CNN, followed by validation (classification accuracy), which distinguishes the samples into low and high probability and a confusion matrix showing the model classified images into 108 classes from 75,000 random ZP images. Reprinted in part with permission from ref (110). Copyright 2018, ASLO. (b) RF used for classifying tintinnopsis. (c) Basic CNN and pipeline for classification of foraminifera. Reprinted in part with permission from ref (115). Copyright 2019, Elsevier. (d) Automatic prediction of the spatial distribution of copepods well correlated with reference distribution, x-axis distance from the shore, and the y-axis represents the depth in meters. Reprinted in part with permission from ref (107). Copyright 2016, Elsevier. (e) Pipeline representing the steps and ML algorithms used for automated plankton identification and counting. Reprinted in part with permission from ref (112). Copyright 2019, PLoS.

Figure 4

ML algorithms are used to identify and understand the behavior of marine fishes and mammals from acoustic data, (a) Graphical representation of acoustic data recording. (b) Workflow to detect right whale calls using LeNet CNN where feature maps were generated by convolution and max-pooling and (c) prediction of right whale upcalls by deep net with probability >0.8. Reprinted in part with permission from ref (130). Copyright 2020, Springer Nature.

Figure 5

Different ML algorithms were used to identify marine fishes and mammals. (a) Images were created that resembles Deep vision photography, where different numbers of fish images were cropped and pasted into the empty background (at a random spot, orientation, and size) and used to train the ML model for real-time identification of fishes. Reprinted in part with permission from ref (140). Copyright 2019, Oxford University Press. (b) Classification of swimming fish from the background and drifting particles, where ML identified fishes (red box) and nonliving particles (blue box) successfully. Reprinted in part with permission from ref (136). Copyright 2014, Elsevier. (c) Workflow proposed for identifying fishes, where CNN extracts features, than pooling and finally classification by linear SVM. Reprinted in part with permission from ref (137). Copyright 2016, Elsevier. (d) Counting of whales by CNN-step-2, based on CNN step-1, which locates the whale in the grid cell green boxes, whereas the red box represents the false negative result. Map data were obtained from Google and DigitalGlobe. Reprinted in part with permission from ref (144). Copyright 2019, Springer Nature.