Quantitative elemental mapping of heavy metals translocation and accumulation in hyperaccumulator plant using laser-induced breakdown spectroscopy with interpretable deep learning

Abstract

As the material basis of phytoremediation for heavy metal-contaminated soils, hyperaccumulator plants are capable of metal hyperaccumulation. Rapid and accurate detection and visualization of heavy metals in hyperaccumulators are essential for investigating metal translocation and monitoring phytoremediation. However, existing analytical and mapping techniques are time-consuming and inefficient in quantification, hindering remediation tracking and timely decision-making, while their high cost and limited accessibility reduce economic feasibility for agricultural and environmental applications. Laser-induced breakdown spectroscopy (LIBS) is a promising alternative for elemental mapping due to its complementary analytical performance and cost-effective instrumentation. Here, LIBS assisted with artificial intelligence was used to establish a quantitative elemental mapping method for Cd and Zn in Sedum alfredii, an important Cd/Zn co-hyperaccumulator. A dataset of 288 samples, comprising shoots and roots of hydroponically and soil-cultivated S. alfredii, was used to develop and test quantitative models. Traditional machine learning encountered overfitting and poor performance in low-concentration samples. To overcome the challenge, a convolutional neural network (CNN) with feature fusion was applied to enhance predictability, reduce matrix effect interference, and enable integrated detection across shoots and roots (Cd: coefficient of determination (R²ₜₑₛₜ) = 0.9887, residual prediction deviation (RPD) = 9.46; Zn: R²ₜₑₛₜ = 0.9887, RPD = 15.21). SHapley Additive exPlanations (SHAP) offered model interpretability by quantifying feature importance, promoting future development of portable LIBS devices. Based on optimal models, quantitative mapping across the leaf-stem-root system was generated, revealing Cd and Zn uptake through root tips and lateral roots, with translocation from roots to shoots and older to younger shoots. The LIBS-based quantitative elemental mapping provides an effective tool for monitoring heavy metals in hyperaccumulators, guiding phytoremediation, and securing agricultural production.

Code and data availability

公開状態または取得可能な本文経路を確認できませんでした。

No evidence-backed public reproduction asset is currently recorded.