Using Experimental Models and Multitemporal Landsat-9 Images for Cadmium Concentration Mapping
Shilan Felegari, Alireza Sharifi, Mohammad Khosravi, Sergei Sabanov
Abstract
Remote sensing technology and its integration with machine learning are examples of effective and low-cost acquisition, particularly in the field of environmental studies and earth sciences for optimal management. The purpose of this study is to accurately map the cadmium (Cd) concentration and introduce the most suitable regression model among different models, including support vector regression (SVR), partial least square regression (PLSR), and artificial neural networks (ANN). In this study, instead of using a single-date Landsat-9 image, multi-temporal images were used, and the extraction of suitable features to estimate the Cd concentration was selected as the input of the regression model. For this purpose, twenty Landsat-9 images and 100 soil samples were prepared from the Ust-Kamenogorsk region in northeastern Kazakhstan. The comparison of multi-temporal images and single-date images of Landsat-9 showed that multi-temporal images have a better performance for monitoring the concentration of heavy metals compared to single-date images. Among various features investigated in this study, the OB (original band) was found to be the most appropriate feature for grouping regressions. The results suggested that SVR model with OB as the input leads to the most accurate estimate of the Cd concentration in the area, which was in the range of 8 to 26 mg/kg.