Fig. 1. Sampling points of AM in this work

Fig. 2. (a) The relationship between fluorescence intensity of excitation/emission wavelength of 280/355 nm and extractant type and extractant ratio; (b) The relationship between fluorescence intensity of excitation/emission wavelength of 355/430 nm and extractant type and extractant ratio; (c) Variation of fluorescence intensity with ultrasonic time

Fig. 3. Rayleigh scattering eliminated from EEM data by interpolation
(a): Raw data with visible Rayleigh scattering; (b): Defect data after deducting Rayleigh scattering; (c): Data repaired by interpolation

Fig. 4. Typical EEM contour map of AM samples in different provinces after deducting scattering
(a): Anhui AM; (b): Hunan AM; (c): Zhejiang AM

Fig. 5. The normalized excitation spectra profiles (a), the normalized emission spectra profiles (b) and the relative concentration spectra profiles (c) obtained by ATLD analysis of three-way matrix with N=3

Fig. 6. The normalized excitation spectra profiles (a) and normalized emission spectra profiles (b) of 7 standard substances with fluorescence signals that may be contained in AM

Fig. 7. Three-way scatter diagram of training samples based on the scores of the first three latent variables of PLS-DA

Table 1. Geographical and weather conditions and the number of samples when picking AM (The prediction set samplesare not included)

Table 2. Chinese and English names, molecular weights, concentrations, structural formulas, purity and production companies of 17 standard substances that may exist in AM

Table 3. The complexity parameters (LVs/k) of the PLS-DA and kNN classification models, the classificationaccurate rate, specificity and sensitivity obtained by cross-validation, training set, test set and prediction set

Table 4. Confusion matrix of training set, test set and prediction set obtained by PLS-DA and kNN classification methods in AM classification model