Artificial intelligence models to predict acute phytotoxicity in petroleum contaminated soils

https://doi.org/10.1016/j.ecoenv.2020.110410Get rights and content

Highlights

  • We studied TPH phytotoxicity in eleven soils from Sakhalin island.

  • Barley root bioassay response to TPH is highly complex and non-linear.

  • ANN and SVR are able to predict phytotoxicity in the TPH polluted soils.

Abstract

Environment pollutants, especially those from total petroleum hydrocarbons (TPH), have a highly complex chemical, biological and physical impact on soils. Here we study this influence via modelling the TPH acute phytotoxicity effects on eleven samples of soils from Sakhalin island in greenhouse conditions. The soils were contaminated with crude oil in different doses ranging from the 3.0–100.0 g kg−1. Measuring the Hordeum vulgare root elongation, the crucial ecotoxicity parameter, we have estimated. We have also investigated the contrast effect in different soils. To predict TPH phytotoxicity different machine learning models were used, namely artificial neural network (ANN) and support vector machine (SVM). The models under discussion were proved to be valid using the mean absolute error method (MAE), the root mean square error method (RMSE), and the coefficient of determination (R2). We have shown that ANN and SVR can successfully predict barley response based on soil chemical properties (pH, LOI, N, P, K, clay, TPH). The best achieved accuracy was as following: MAE – 8.44, RMSE –11.05, and R2 –0.80.

Introduction

Crude oil contamination, arising from oil production and transporting procedures (Larive, 2008), has a devastating impact on the surrounding terrestrial ecosystems entailing agricultural production and thus human health (Khan et al., 2018). Crude oil includes various aliphatic and aromatic hydrocarbons, which are rich in petroleum hydrocarbon and non-hydrocarbon compounds, yet are deficient in any nutritional elements (CCME, 2008). Since total petroleum hydrocarbons (TPH) structurally belong to the group of complex chemical compounds that are either found within crude oil (CCME, 2008) they can be used as a means of measuring soil contamination with crude oil. The level of TPH in polluted soils largely depends on soil properties and can rise almost by 10 times as compared to the background level. What is more important, TPH input to the soil may trigger immediate changes in the environment and thus induce adverse biological and ecological effects (Garcia et al., 2019; Hunt et al., 2019).

Among different soil quality indicators, acute phytotoxicity is a conventional yet very efficient method to assess the extent to which soil is polluted (Gerber et al., 2017). It is often measured by calculating the seed germination inhibition, root growth inhibition, or any other adverse effects on plants. Several studies demonstrated TPH contamination to be closely related to soil phytotoxicity (Siddiqui and Adams, 2002; Molina-Barahona et al., 2005; Kanarbik et al., 2014; Kaur et al., 2017). The mechanism by which TPH may induce soil phytotoxicity is rather complex as the effects produced by various environmental factors and soil properties are synergistic. Therefore, it is essential to perform an accurate simulation of the threefold relationship between plant response, TPH content, and physicochemical factors. Thus, quantifying and predicting TPH phytotoxicity in the soil is a crucial issue for soil planning and remediation.

Meanwhile, experimental measurements assessing the correlation between soil properties and phytotoxicity of the contaminants are time-consuming and complicated, and, what is more, it seems impossible to test all the naturally existing variants with utmost accuracy. Hence, the quest to find a universal approach to determine a non-linear correspondence between bio-indicators and the physicochemical data observed in the soil is necessary. These demands are met by machine learning techniques (ML) that demonstrated a rapid surge of interest to designing models to simulate and predict soil processes (Goodarzi et al., 2016; Olawoyin, 2016; Cipullo et al., 2019; Sayyad et al., 2019). Modern supervised machine learning methods, such as support vector machine (SVM) and artificial neural networks (ANN), are considered to be promising and efficient tools aimed at interpreting high-dimensional and high-nonlinear data in environmental science. Previously, SVM was used to predict spatial distribution of soil organic carbon, soil nitrogen stock (Kou et al., 2019), and soil salinity (Wu et al., 2018). ANNs were successfully applied to model soil physical properties in case of temperature fluctuations (Ozturk et al., 2011) and erosion (Gholami et al., 2018). It was useful in determining soil chemical properties (Fernandes et al., 2019) as well as soil biological activity (Jha and Ahmad, 2018; Ebrahimi et al., 2019).

In this study, we aim to show that a few observations and measurements of the soil properties provide the possibility to predict their phytotoxicity judging by the range of TPH concentration in the soil. More specifically, we compare the performance of ANN and SVM in predicting barley phytotoxicity considering TPH concentration and soil properties as the driving factors. We also share the newly obtained data on studying the soils from Sakhalin and their phytotoxicity under various petroleum dose applications.

Section snippets

Material and methods

Top soils samples from eleven field sites on Sakhalin island were brought to the laboratory, where they were thoroughly mixed and quartered (Table 1). The total weight of one wet soil sample was 4 kg. The soils were stored in airtight containers at a room temperature before analysing. The soil samples were classified according to WRB (Chesworth, 2008).

The experiment was conducted in greenhouse conditions (22±1 °C). Crude oil was characterized by the bulk density 0.83 g cm−3, 2.6% of the C5–C9

Changes in physicochemical and biological properties of the soil under crude oil pollution

The soils used for this study exhibited a wide range of soil properties (Table 1) and belonged to the main types of zonal and intrazonal soils of the Sakhalin island. Soil pH ranged from 4.30 at Soil #3 (Carbic Podzol) to 5.65 at Soil #5 (Livic Stagnosols Dystric), while in seven samples pH was lower 5.5. The LOI varied greatly: the maximum organic carbon was at Soil#1 (Fabric Histosols Dystric) 97.18%, while Soil#9 (Umbric Fluvisols Oxyaquic) had the lowest values < 2%. Clay content ranged

Discussion

Plant bioassay is an effective and popular tool for estimating the soil quality and assessing ecotoxicological risks (Ghosh et al., 2017). Rapid-cycling plants like barley (Hordeum vulgare) allow to reveal a short-term influence from organic and inorganic contaminants in the soil (Kim et al., 2019; Nikolaeva et al., 2019). According to the ISO 11269–1:2012 (ISO, 2012), root elongation may be a crucial endpoint for ecotoxicological assessment. Normally, elongation tests imply plant development

Conclusions

This study shows the complexity of soil treatment with TPH. As it was revealed, the range of the effect depend largely on the properties of the soil. In most tested soils, adding TPH in the doses over 15 g kg−1 tended to increase the barley root lengths and induced marked phytotoxicity. Nevertheless, in some cases the addition of TPH positively affected the growth of the roots as compared to the test with a non-polluted control.

We demonstrate the benefits of applying the machine learning

CRediT authorship contribution statement

Dmitrii Shadrin: Methodology, Validation, Formal analysis, Writing - original draft. Mariia Pukalchik: Conceptualization, Investigation, Writing - original draft, Funding acquisition. Ekaterina Kovaleva: Methodology, Investigation, Writing - review & editing. Maxim Fedorov: Supervision, Project administration.

Declaration of competing interest

The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Acknowledgments

This research was funded by the Russian Science Foundation (grant # 18-74-00015). We also thank reviewers for critical review of this manuscript.

References (63)

  • R. Hecht-Nielsen

    Theory of the backpropagation neural network

  • L.J. Hunt et al.

    Petroleum hydrocarbon PHC uptake in plants: a literature review

    Environ. Pollut.

    (2019)
  • S.Kr Jha et al.

    Soil microbial dynamics prediction using machine learning regression methods

    Comput. Electron. Agric.

    (2018)
  • N. Kaur et al.

    A review of germination and early growth as a proxy for plant fitness under petrogenic contamination — knowledge gaps and recommendations

    Sci. Total Environ.

    (2017)
  • M.A.I. Khan et al.

    Toxicity assessment of fresh and weathered petroleum hydrocarbons in contaminated soil- a review

    Chemosphere

    (2018)
  • D. Kou et al.

    Spatially-explicit estimate of soil nitrogen stock and its implication for land model across Tibetan alpine permafrost region

    Sci. Total Environ.

    (2019)
  • O. Nikolaeva et al.

    Ecotoxicological effects of traffic-related pollutants in roadside soils of Moscow

    Ecotoxicol. Environ. Saf.

    (2019)
  • R. Olawoyin

    Application of backpropagation artificial neural network prediction model for the PAH bioremediation of polluted soil

    Chemosphere

    (2016)
  • A.P. Sergeev et al.

    Combining spatial autocorrelation with machine learning increases prediction accuracy of soil heavy metals

    Catena

    (2019)
  • T.H. Tran et al.

    Germination, physiological and biochemical responses of acacia seedlings Acacia raddiana and Acacia tortilis to petroleum contaminated soils

    Environ. Pollut.

    (2018)
  • G. Wu et al.

    Machine learning models for predicting PAHs bioavailability in compost amended soils

    Chem. Eng. J.

    (2013)
  • I. Yilmaz et al.

    Multiple regression, ANN RBF, MLP and ANFIS models for prediction of swell potential of clayey soils

    Expert Syst. Appl.

    (2011)
  • E. Arendt et al.

    Barley in: Cereal Grains for the Food and Beverage Industries

    (2013)
  • J. Bori et al.

    Bioassays with terrestrial and aquatic species as monitoring tools of hydrocarbon degradation

    Environ. Sci. Pollut. Control Ser.

    (2016)
  • CCME

    Canadian Council of Ministers of the Environment Canada-wide Standard for Petroleum Hydrocarbons PHC in Soil: Scientific Rationale

    (2008)
  • B. Choubin et al.

    Multiple linear regression, multi-layer perceptron network and adaptive neuro-fuzzy inference system for forecasting precipitation based on large-scale climate signals

    Hydrol. Sci. J.

    (2016)
  • J.M. Cruz et al.

    Comparative study of phytotoxicity and genotoxicity of soil contaminated with biodiesel, diesel fuel and petroleum

    Ecotoxicology

    (2019)
  • W. de Vries et al.

    Impact of soil properties on critical concentrations of cadmium, lead, copper, zinc, and mercury in soil and soil solution in view of ecotoxicological effects

  • H. Drucker et al.

    Support vector regression machines

  • S. Gaskin et al.

    Screening of Australia native grasses for rhizoremediation of aliphatic hydrocarbon-contaminated soil

    Int. J. Phytoremediation

    (2008)
  • Cited by (0)

    View full text