Using machine learning to understand the implications of meteorological conditions for fish kills

You-Jia Chen,Emily Nicholson,Su-Ting Cheng

doi:10.1038/s41598-020-73922-3

You-Jia Chen, Emily Nicholson + Show 1 more

Open Access

https://doi.org/10.1038/s41598-020-73922-3

Copy DOI

Journal: Scientific Reports	Publication Date: Oct 12, 2020
Citations: 7	License type: open-access

Affiliation: Deakin University, National Taiwan University

Abstract

Fish kills, often caused by low levels of dissolved oxygen (DO), involve with complex interactions and dynamics in the environment. In many places the precise cause of massive fish kills remains uncertain due to a lack of continuous water quality monitoring. In this study, we tested if meteorological conditions could act as a proxy for low levels of DO by relating readily available meteorological data to fish kills of grey mullet (Mugil cephalus) using a machine learning technique, the self-organizing map (SOM). Driven by different meteorological patterns, fish kills were classified into summer and non-summer types by the SOM. Summer fish kills were associated with extended periods of lower air pressure and higher temperature, and concentrated storm events 2–3 days before the fish kills. In contrast, non-summer fish kills followed a combination of relatively low air pressure, continuous lower wind speed, and successive storm events 5 days before the fish kills. Our findings suggest that abnormal meteorological conditions can serve as warning signals for managers to avoid fish kills by taking preventative actions. While not replacing water monitoring programs, meteorological data can support fishery management to safeguard the health of the riverine ecosystems.

Highlights

Fish kills, often caused by low levels of dissolved oxygen (DO), involve with complex interactions and dynamics in the environment
Based on the clustering results of the self-organizing map (SOM), followed by a systematic risk analysis, our results suggested that the occurrence of fish kills can be categorized into different types, and different meteorological stressors can cause cumulative effects that increase the risk of fish kills
Excluding those clearly resulted from industrial pollution and those missing required meteorological observations, the final 19 events analyzed occurred across 19 sites in the months of April to September (Fig. 1) with estimated weights of dead fish ranging from 1500 to 35,000 kg (Table 1)

Summary

Introduction

Often caused by low levels of dissolved oxygen (DO), involve with complex interactions and dynamics in the environment. Non-summer fish kills followed a combination of relatively low air pressure, continuous lower wind speed, and successive storm events 5 days before the fish kills. Interactions between meteorological factors on the amount of oxygen dissolved in water are c omplex[15,16]: temperature controls the saturation concentration of D O17–19; precipitation washes oxygen-consuming material into r ivers[11,12]; and wind speed promotes DO through air–water oxygen diffusion by creating rough s urfaces[20,21]. Organic carbon (TOC)[17,22] These complex interactions present challenges in relating meteorological mechanisms to levels of DO in water, continuous meteorological observations are typically taken in many places around the world. Depending on the extent to which the meteorological factors can explain in relation to the DO conditions, any nonlinear relationship behind the fish kills might be revealed by new analytical approaches

Methods

Results

Conclusion