A Deep Learning approach for modeling the spatial distribution of contaminants in the Black Sea

Abstract

Black Sea (BS) is an important ecosystem, which is affected by various anthropogenic pressures, such as shipping activities, wastewater inputs from large coastal cities and most importantly loads by major rivers (e.g., Danube, Dniester, Dnieper). The chemical pollution that rivers transfer to the BS is significant considering that the Danube river alone discharges 6550 m3/s to the BS. This study focuses on the Ukrainian shelf (the northwestern part of the Black Sea) and investigates the river sources of chemicals in the shelf. To achieve this objective, data generated by the Joint Black Sea Surveys (JBSS) was used. JBSS took place in 2016 and 2017 in context of the EU/UNDP EMBLAS II project (www.emblasproject.org). During the JBSS campaign, seawater samples were collected, extracted and analyzed by highthroughput analytical methods such as liquid chromatography high-resolution mass spectrometry (LC-HRMS). The analysis resulted in data, which was processed using open-source algorithms to generate a dataset with the detected chemical signals and their intensity in the sampling stations. The dataset was used to generate images, representing the spatial distribution of the signals. The figures were then used as an input to a deep learning convolutional neural network classification model. The aim of the study was to create an end-to-end solution for the estimation of the pollution potential of the major contributing rivers (Dnieper and Danube) in the Ukrainian shelf. Finally, a dashboard to facilitate data visualization and results’ evaluation was built. The generation of such models can also serve to the prioritization of unknown chemical signals, which is the key for non-target screening.