Leaf magnetic properties as a method for predicting heavy metal concentrations in PM2.5 using support vector machine: A case study in Nanjing, China.

The aim of this study was to establish a method for predicting heavy metal concentrations in PM2.5 (particulate matter with a diameter of less than 2.5 µm) using support vector machine (SVM) models combined with magnetic properties of leaves. In this study, PM2.5 samples and the leaves of three common evergreen tree species were collected simultaneously during four different seasons in Nanjing, China. A SVM algorithm was used to establish models for the prediction of airborne heavy metal concentrations based on leaf magnetic properties, with or without meteorological factors and pollutant concentrations as input variables. Results showed that the annual average PM2.5 concentration was 58.47 µg/m3. PM2.5 concentrations, leaf magnetic properties, and nearly all airborne heavy metals had higher concentrations in winter than in spring, summer, or fall. Ferrimagnetic minerals preponderant in dust-loaded leaves were sampled from the three tree species. Models using magnetic properties of leaves from Ligustrum lucidum Ait and Osmanthus fragrans Lour yielded better prediction effects than those based on the leaves of Cedar deodara G. Don, showing relatively higher correlation coefficient (R) values and lower errors in both training and test stages. Fe and Pb concentrations were well-simulated by the prediction models, with R values > 0.7 in both training and test stages. By contrast, the concentrations of V, Co, Sb, Tl, and Zn were relatively poor-simulated, with most R values < 0.7 in both training and test stages. Predictions for the main urban areas of Nanjing showed that the highest heavy metal concentrations occurred near industrial and traffic pollution sources. Our results provide a cost-effective approach for the prediction of airborne heavy metal concentrations based on the biomagnetic monitoring of tree leaves.

Response of magnetic properties to metal deposition on urban green in Nanjing, China.

Environmental magnetism is a simple and fast method that can be used to assess heavy metal pollution in urban areas from the relationships between magnetic properties and heavy metal concentrations. Leaves of Osmanthus fragrans, one of the most widely distributed evergreen trees in Nanjing, China, were collected from four different district types, i.e., residential, educational, traffic, and industrial. The magnetic properties and heavy metal concentrations were measured both for unwashed (dust-loaded) and washed leaves. Scanning electron microscopy with energy-dispersive X-ray spectroscopy confirmed that unwashed leaves accumulated much dust due to atmospheric deposition. The value of magnetic properties and heavy metal concentrations in unwashed leaves was significantly higher than those of washed leaves, indicating that these characteristics were mainly derived from atmospheric particulate matter. Saturation isothermal remanent magnetization (SIRM) values obtained from unwashed and washed leaves ranged from 209.14 × 10-6 to 877.85 × 10-6 Am2 kg-1 and from 69.50 × 10-6 to 501.28 × 10-6 Am2 kg-1, respectively. High concentrations of heavy metals, such as Pb and Fe, the Tomlinson pollution load index, and the SIRM of unwashed leaves occurred in the traffic and industrial districts. A preliminary principal component analysis identified the source categories and suggested that industrial activities may be more related to the release of particulate matter rich in Fe. The heavy metal concentrations and pollution load index showed significant positive correlations with the low-frequency magnetic susceptibility and SIRM of unwashed leaves, indicating that these properties can be used to semi-quantify atmospheric heavy metal pollution. Our study suggests that it is possible to employ magnetic measurements as a useful tool for the monitoring and assessment of atmospheric heavy metal pollution.

Magnetic Properties as a Proxy for Predicting Fine-Particle-Bound Heavy Metals in a Support Vector Machine Approach.

The development of a reasonable statistical method of predicting the concentrations of fine-particle-bound heavy metals remains challenging. In this study, daily PM2.5 samples were collected within four different seasons from a Chinese mega-city. The annual average PM2.5 concentrations determined in industrial, city center, and suburban areas were 90, 81, and 85 µg m-3, respectively. Environmental magnetic measurements, including magnetic susceptibility, anhysteretic remanent magnetization, isothermal remanent magnetization, hysteresis loops, and thermomagnetism, indicated that the main magnetic mineral of PM2.5 is low-coercivity pseudosingle domain (PSD) magnetite. Using a support vector machine (SVM), both the volume- and mass-related concentrations of heavy metals were predicted by the PM2.5 mass concentrations and meteorological factors, with or without magnetic properties as input variables. The inclusion of magnetic variables significantly improved the prediction results for most heavy metals. Predictions based on models that included the magnetic properties of the metals Al, Fe, Mn, Ni, and Ti were promising, with R values of >0.8 in both the training and the test stages as well as relatively low errors. Our results demonstrate that the inclusion of environmental magnetism in a SVM approach aids in the effective monitoring and assessment of airborne heavy-metal contamination in cities.

Prediction of size-fractionated airborne particle-bound metals using MLR, BP-ANN and SVM analyses.

Size-fractionated heavy metal concentrations were observed in airborne particulate matter (PM) samples collected from 2014 to 2015 (spanning all four seasons) from suburban (Xianlin) and industrial (Pukou) areas in Nanjing, a megacity of southeast China. Rapid prediction models of size-fractionated metals were established based on multiple linear regression (MLR), back propagation artificial neural network (BP-ANN) and support vector machine (SVM) by using meteorological factors and PM concentrations as input parameters. About 38% and 77% of PM2.5 concentrations in Xianlin and Pukou, respectively, were beyond the Chinese National Ambient Air Quality Standard limit of 75 µg/m3. Nearly all elements had higher concentrations in industrial areas, and in winter among the four seasons. Anthropogenic elements such as Pb, Zn, Cd and Cu showed larger percentages in the fine fraction (ø≤2.5 µm), whereas the crustal elements including Al, Ba, Fe, Ni, Sr and Ti showed larger percentages in the coarse fraction (ø > 2.5 µm). SVM showed a higher training correlation coefficient (R), and lower mean absolute error (MAE) as well as lower root mean square error (RMSE), than MLR and BP-ANN for most metals. All the three methods showed better prediction results for Ni, Al, V, Cd and As, whereas relatively poor for Cr and Fe. The daily airborne metal concentrations in 2015 were then predicted by the fully trained SVM models and the results showed the heaviest pollution of airborne heavy metals occurred in December and January, whereas the lightest pollution occurred in June and July.