Quantitative Estimation of Zn and Pb in Soil Using Multivariate Analysis and Remote Sensing Technique

Heavy metal pollution is one of the most important environmental problems worldwide and measuring their concentration in the soil is the first step in providing a solution to reduce their risks. However, conventional methods for estimating the amount of heavy metals in the soil are time-consuming and costly. Imagery and remote sensing techniques have shown good potential to be an alternative to the routine approaches for monitoring heavy metal pollution in soil. The aim of this study was to evaluate the efficiency of multivariate analysis methods and remote sensing technique to quantify soil contamination with zinc and lead under different land uses. Previous studies revealed the high concentrations of zinc and lead in Zanjan’ soils cause a serious concern regarding the health of humans in this province.

The study was conducted between the latitude 36° 20' 48" N to 36° 47'45" N, and longitude 48° 15' 51" E to 49° 00' 55" E in Zanjan province. The average daily temperature in the study area is 15.7°C and the mean annual precipitation is 335 mm. Rangeland, agriculture, industrial and residential areas are the main land uses in Zanjan. A total of 230 soil samples were collected at depths of 0–10 cm under different land uses from an area of ​​3424 Km2 at two grids intervals of 1.5 kilometers in industrial and residential areas, and 3 kilometers in rangeland and agriculture. Before chemical analysis, soil samples were air-dried and sieved. Soil lead and zinc contents were measured using an atomic absorption spectrometry (Perkin- Elmer: AA 200). Radiometric and geomatic corrections were performed on satellite images and the “special resampling” procedure was used to resample the 20 m images to 10 m. Thirty spectral indices were determined using Sentinel 2 satellite images. These indices were reported as practical indices for remote sensing assessment of land condition. The indices included; NDVI, NDRE, MTVI, MCAR, MNLI, GNDVI, SAVI, LCI, MTCI, PSRI, CI-RedEdge, CI-Green, NLI, TVI, EVI, STAVI, GRI, LSWI, MSAVI, BI, BI2, RI, CI and seven spectral ratios. Spectral estimation models were developed and evaluated by three methods of principal component regression (PCR), partial least squares regression (PLSR) and support vector machine regression (SVMR). The samples were randomly divided into the validation set (30%) and the calibration set (70 %). Therefore, 161 samples were used for calibrating and 69 samples for validating models. Levene’s test was performed to test the variance homogeneity between calibration and validation sets. The accuracy of models was evaluated using the coefficient of determination (R2), root mean square error of prediction (RMSEP) and the ratio of predicted deviation (RPD). Martens’ uncertainty test was used to identify important wavelengths for zinc and lead estimations.

The range of lead in the studied soils was 40 to 364 mg kg-1 and the range of zinc was 96 to 824 mg kg-1. The accuracy of spectral models was categorised into excellent (RPD ≥ 2.5 and R2 ≥ 0.8), good (2 ≤ RPD < 2.5 and R2 ≥ 0.7), moderate (1.5 ≤ RPD < 2 and R2 ≥ 0.6) and poor accuracy (RPD < 1.5 and R2 < 0.6) based on the soil spectral accuracy classification, that is presented by Askari et al., (2015 and 2019). The highest concentration of metals was observed in industrial and residential land uses. The SVMR model (RPD≥2.6 and R2≥0.84, RMSE≤ 40), had a better spectral estimation for both lead and zinc than the PLSR model (RPD≥ 1.9 and R2≥  0.7, RMSE≤ 53) and the PCR model  (RPD≥1.3 and R≥ 0.45, RMSE≤ 75). Red-edge and infrared range were identified as the most effective wavelength ranges for monitoring the contamination of lead and zinc in soil. Brightness and modified triangular vegetation index were the most effective indicators for spectral estimation of lead and zinc in the studied soils. The SVMR model showed high accuracy and the PLSR model showed acceptable accuracy for evaluating and monitoring lead and zinc contamination using Sentinel 2 images. Comparing the predicted and measured values of heavy metals with a 1:1 line showed an overestimation for low values of lead and zinc, and an underestimation for high values of lead and zinc.

This study revealed that the method of multivariate analysis and remote sensing data could provide a practical approach for rapid and quantitative assessment of soil heavy metal pollution in Zanjan province and areas with similar soil conditions. An accurate prediction of heavy metal pollution can be acquired using freely available Sentinel-2 multispectral imagery system.