A Comparative Study on Machine Learning Algorithms for Geochemical Prediction Using Sentinel-2 Reflectance Spectroscopy

The distribution of stream sediments is usually considered as an important and very useful tool for the early-stage exploration of mineralization at the regional scale. The collection of stream samples is not only time-consuming but also very costly. However, the advancements in space remote sensing has made it a suitable alternative for mapping of the geochemical elements using satellite spectral reflectance. In this research work, 407 surface stream sediment samples of the zinc (Zn) and lead (Pb) elements are collected from Central Wales. Five machine learning models, namely the Support Vector Regression (SVR), Generalized Linear Model (GLM), Deep Neural Network (DNN), Decision Tree (DT), and Random Forest (RF) regression, are applied for prediction of the Zn and Pb concentrations using the Sentinel-2 satellite multispectral images. The results obtained based on the 10 m spatial resolution show that Zn is best predicted with RF with significant R2 values of 0.74 (p < 0.01) and 0.7 (p < 0.01) during training and testing. However, for Pb, the best prediction is made by SVR with significant R2 values of 0.72 (p < 0.01) and 0.64 (p < 0.01) for training and testing, respectively. Overall, the performance of SVR and RF outperforms the other machine learning models with the highest testing R2 values.