Application of artificial neural networks for identifying poorly formalized patterns in experimental become widespread nowadays. In the present work, we elaborated a prototype of a model for detection of water contamination with heavy metals. The model is a fully connected neural network (multilayer perceptron) designed by using the Python programming language and the TensorFlow software (Keras). It is intended for application in environmental monitoring of natural water bodies using chlorophyll fluorescence measurements which are considered as highly informative approach for probing photosynthetic activity in vivo and in situ. Fluorescence rise induced by application of a strong light pulse to the dark-adapted plant or algae (the OJIP transient) reflects a stepwise transition of the photosynthetic electron transport chain from the oxidized to the fully reduced state. To provide a quantitative analysis of the OJIP transient, a JIP test was introduced whose parameters describe energy fluxes through the photosynthetic electron transport chain. Our model uses OJIP transients and/or JIP-test parameters, measured in phytoplankton communities, as input data. As a result, it determines the probability of water pollution by heavy metals. In order to test the model, phytoplankton samples were taken from 9 water bodies of Pskov region and then treated with chromium and cadmium under laboratory conditions. For that, phytoplankton samples were exposed to cadmium and chromium salts (CdSO4 and K2Cr2O7) at two concentrations (20 and 50 μM) for three days, and OJIP curves were recorded and JIP-test parameters calculated at different stages of the experiment. In total, 419 curves were collected, and a whole dataset was analyzed. Results showed that accuracy of detecting the toxic effects of Cd2+ (after 2 or more hours of incubation) and Cr2O72– (after 9 or more hours of incubation) by the model achieved 90%. The highly accurate determination of the toxicity of heavy metals indicates a promising prospect for the application of machine learning technology in environmental monitoring.
ISSN 1995-4301
Select viewing options
