Application of the Deep Learning Algorithm to Identify the Spatial Distribution of Heavy Metals at Contaminated Sites
Jun Man, Lingzao Zeng, Jian Luo, Weiliang Gao, Yijun Yao
Abstract
Accurate estimation of the spatial distribution of pollutants, such as heavy metals, is indispensable in investigation of contaminated sites. In this paper, we develop a novel deep learning (DL) algorithm, in which the data value of the target point is predicted by a nearest-neighbor neural network. Using 20 hypothetical sites and a real-world one, the developed DL is compared with several commonly used kriging algorithms (e.g., ordinary kriging). By taking the results of ordinary kriging as the benchmark, the interpolation accuracy can be improved by an average of 38.2% and 11.2–36.6%, respectively. The comparisons also indicate that the new method performs much better in cases with significant spatial variability by alleviating the smoothing effect and the edge effect in the kriging. Upon further examination of the developed method, the prediction accuracy is found to first increase and then decrease with the number of neighbor points. Moreover, the influence of the sampling density is limited if the number exceeds a certain threshold (e.g., n = 64 in our case). As a preliminary attempt of applying the DL algorithm at individual contaminated sites, this work provides a general alternative method to identify the spatial distribution of heavy metals.