Multi-level data fusion of laser-induced breakdown spectroscopy and X-ray fluorescence for arsenic determination in...

Pteris vittata, an arsenic-hyperaccumulating fern, is widely employed for phytoremediation of arsenic (As). Rapid, accurate assessment of As in P. vittata is crucial for evaluating its accumulation ability. In this study, P. vittata was analyzed using a spectral fusion of laser-induced breakdown spectroscopy (LIBS) and X-ray fluorescence (XRF). A total of 60 biological samples (roots and fronds) were collected and prepared as 180 compressed tablets for spectroscopic analysis, covering an As concentration range of 88-1956 mg kg^-1. Multivariate analysis methods were employed for full spectra and feature spectra, including partial least squares regression (PLSR), least squares support vector machine (LSSVM), extreme learning machine (ELM), random forest (RF), and adaptive weighting normalization-linear weighted network (AWN-LWNet). The best single-modality model, an XRF-based PLSR model built upon feature spectra selected by the Competitive Adaptive Reweighting Sampling (CARS) algorithm, achieved a prediction performance of R²_P = 0.969, RMSE_P = 54.13 mg kg^-1, and MAE_P = 43.00 mg kg^-1. Spectral fusion further enhanced prediction accuracy, with high-level fusion outperforming low- and mid-level methods. The proposed feature-spectra-based decision fusion model achieved the best performance (R²_P = 0.980, RMSE_P = 43.69 mg kg^-1, MAE_P = 32.49 mg kg^-1), corresponding to reductions of 19.3% in RMSE_P and 24.4% in MAE_P compared to the best single-modality model. These results demonstrate that spectral fusion effectively integrates complementary information, improving the accuracy of As quantification in complex plant matrices. The proposed approach provides a rapid and non-destructive strategy for monitoring arsenic accumulation in phytoremediation plants.