Assessing the Role of Environmental Factors on Baltic Cod Recruitment, a Complex Adaptive System Emergent Property

Abstract

For decades, fish recruitment has been a subject of intensive research with stock-recruitment models commonly used for recruitment prediction often only explaining a small fraction of the inter-annual recruitment variation. The use of environmental information to improve our ability to predict recruitment, could contribute considerably to fisheries management. However, the problem remains difficult because the mechanisms behind such complex relationships are often poorly understood; this in turn, makes it difficult to determine the forecast estimation robustness, leading to the failure of some relationships when new data become available. The utility of machine learning algorithms such as artificial neural networks (ANNs) for solving complex problems has been demonstrated in aquatic studies and has led many researchers to advocate ANNs as an attractive, non-linear alternative to traditional statistical methods. The goal of this study is to design a Baltic cod recruitment model (FishANN) that can account for complex ecosystem interactions. To this end, we (1) build a quantitative model representation of the conceptual understanding of the complex ecosystem interactions driving Baltic cod recruitment dynamics, and (2) apply the model to strengthen the current capability to project future changes in Baltic cod recruitment. FishANN is demonstrated to bring multiple stressors together into one model framework and estimate the relative importance of these stressors while interpreting the complex non-linear interactions between them. Additional requirements to further improve the current study in the future are also proposed.

Enhancing Europe's capability in end-to-end marine ecosystem modelling for societal benefit

“…ANN have been proposed for understanding the processes impacting on ecosystem state (Barreiro et al, 2018), to predict the distribution of phytoplankton groups over the global ocean (Palacz et al, 2013), and fish recruitment in the marine environment (e.g. Krekoukiotis et al, 2016). Other popular applications of machine learning to marine ecosystem models include statistical analyses such as Bayesian Beliefs Networks (BBN), which have been used for the identification of trade-offs among multiple uses of marine ecosystems and random forest algorithms.…”

Using machine learning to link spatiotemporal information to biological processes in the ocean: a case study for North Sea cod recruitment

“…Various attempts have already been made to use ML methods for the prediction of stock−recruitment relationships, including neural networks (Chen & Ware 1999, Megrey et al 2005, Krekoukiotis et al 2016, random forest (Hansen et al 2015, Smoliński 2019) and naïve Bayes (Fernandes et al 2010(Fernandes et al , 2015. Although it is believed that these methods can improve predictions of recruitment (Megrey et al 2005), they are not a standard tool among fisheries scientists.…”