Estimating soil cadmium concentration using multi-source UAV imagery and machine learning techniques.
Journal:
Environmental monitoring and assessment
PMID:
40272583
Abstract
Urbanization and industrialization have led to widespread soil heavy metals contamination, posing significant risks to ecosystems and human health. Conventional methods for mapping heavy metal distribution, which rely on soil sampling followed by chemical analysis, are costly and time-consuming. This study proposes a novel approach for estimating soil cadmium (Cd) concentrations by integrating multi-source data captured by Unmanned Aerial Vehicle (UAV), including multispectral images, Digital Elevation Model (DEM), and high-resolution RGB aerial imagery, with machine learning algorithms. Environmental factors such as proximity to pollution sources, terrain attributes, and remote sensing indices were extracted from the UAV multi-source data and used to train machine learning models for soil Cd concentration estimation. Among the tested models, the Gradient Boosting Decision Tree (GBDT) algorithm demonstrated the highest accuracy in estimating soil Cd levels. The proposed method achieves higher predictive accuracy than Kriging interpolation, with Mean Squared Error (MSE) and Mean Absolute Error (MAE) reduced by 37% and 22%, respectively. Additionally, the approach quantified the relative importance of each explanatory variable using feature importance scores derived from machine learning regression models, revealing that proximity to pollution sources was the most influential factor affecting soil Cd concentrations in the study area. This study demonstrates the potential of UAV-based multi-source data, combined with machine learning techniques, as a complementary approach to conventional soil contamination mapping methods. The proposed methodology improves soil contamination assessment efficiency, aiding hotspot detection and targeted remediation. These findings suggest UAV-based remote sensing could support environmental monitoring and land management.