Prediction of Pb and Zn in urban soil using VIS-NIR-SWIR spectroscopy

Abstract

<p>Heavy metals serve as a subset of chemical elements with higher density than iron. Besides, these environmental pollutants are constant and nonbiodegradable elements that can cause toxicity and genetic mutations to the live cells. Depending on the study area, an increase in soil heavy metals from a specific level often created by human activities can lead to many adverse effects on individuals, soil, and plants. In case of their existence in the food chain or transfer to groundwater resources, human health is seriously threatened. Over numerous years, being affected by a colossal number of pollutant resources such as world war and household waste, industry, transportation systems, and urbanization has changed Berlin to a city at risk of soil pollution by heavy metals. That is why carrying out a study on heavy metals in this city is of great significance. Chemical analysis is the first and most traditional ways to measure soil heavy metals. Despite high precision, this method is complicated, time-consuming, costly, and ineffective on a large scale. However, the spectral data facilitates the rapid and cost-effective assessment of these elements. Therefore, in this study, the ability of spectral data to predict heavy metals in Berlin’s soil is examined.</p><p>When it comes to the data required, there are two categories: 1) heavy metals (Pb and Zn) related to more than 600 soil samples collected from 2016 to 2018 and measured in the laboratory, and 2) the spectral data measured for each sample in the range between 350 to 2500nm in a spectrometry lab. All data is divided into training (80%) and testing (20%) to reach this aim. Next, the first group is used to train the machine learning algorithms, including partial least square regression (PLSR), support vector regression (SVR), and random forest (RF). Moreover, the second group is used to test the models. Finally, the accuracy of models is evaluated by correlation of determination (R<sup>2</sup>), and Root mean square error (MSE). As a part of the results, R2 and MSE were achieved 0.25, and 4394.45 for Pb, and 0.18 and 6558.49 for Zn.</p>