Study On The Verification Of Optical Characteristics Tracing Sediment Sources

Sediment fingerprints have been widely used in source identification studies.Optical characteristics have become a powerful substitute for traditional fingerprints in recent years due to their fast measurement and low-cost analysis.However,the accuracy of optical fingerprinting methods has received little attention.Here,this study combined artificial mixing,simulated scouring,and indoor sieving experiments with optical fingerprinting methods(visible,near-infrared,and mid-infrared spectroscopic fingerprinting methods;color parameter fingerprinting methods)to analyze and optimize the main factors that affected the prediction performance of optical fingerprinting methods.Then the accuracy of four different optical characteristics(visible,near-infrared,and mid-infrared spectroscopy and color parameters)in predicting sediment sources were compared and evaluated.This study further analyzed the influence of particle sorting on the accuracy of optical fingerprinting methods.This research could provide a reference for improving the accuracy of optical characteristics fingerprinting of sediment source,and was of great significance to further study the application of optical fingerprinting method in source identification.The main conclusions are as follows:(1)The main factors affecting the prediction performance of the optical fingerprinting methods were analyzed and then were optimized.The results showed that preprocessing techniques and modeling methods had a significant impact on the predictive performance of spectral fingerprinting methods.Among them,the prediction performance of the spectroscopic tracing models established by the preprocessed spectrum has been improved.The support vector machine model established by using visible and near-infrared spectral data and partial least squares regression model established by using mid-infrared spectroscopy had the best prediction performance.The selection of unmixing models had a significant impact on the prediction performance of fingerprinting methods for color parameters.Color parameters combined with Walling model and Mix SIAR model provided the same accuracy,while Mix SIAR model showed higher stability than Walling model.The prediction accuracy of the optimized parameters was verified based on fingerprinting results of artificial mixtures by using best spectral tracing models and the best color parameter fingerprints.The results showed that visible,near-infrared,and mid-infrared tracing models showed excellent prediction accuracy(MAE < 0.10).Color parameters combined with Mix SIAR model could provide good prediction effect(MAE < 0.20).(2)The accuracy of tracing sediment sources by four optical characteristics were compared and evaluated,and the influence of particle sorting caused by water flow on the accuracy of optical tracing sediment source was further analyzed.The results showed that the visible,near-infrared and mid-infrared spectral fingerprinting methods had high prediction accuracy(MAE = 0.06)when tracing the sediment sources.In contrast,the color parameters had weak performances in simulated scouring experiments(MAE = 0.16).Additionally,similar fingerprinting results of the artificial mixtures(MAE = 0.08)and scouring mixtures(MAE = 0.09)indicated that particle sorting caused by water flow(slight degree)had little effect on the accuracy of optical fingerprinting results.(3)The impact of particle sorting on the accuracy of optical characteristics for fingerprinting sediment sources was analyzed,and the accuracy of the optical fingerprinting methods were verified under the condition of small differences in sediment sources.Comparing with the changes of particle size distribution and optical characteristics in three sieving particles(< 125,< 63,< 31 μm),the spectral curves and characteristic peak positions of the same group of samples did not significantly change as the soil particles gradually become smaller,while the spectral absorbance value gradually decreased and the color parameter value gradually increased.Best spectral tracing models and color parameter fingerprints selected were used to predict source contribution of sieving mixtures in two test setup groups.The results showed that the spectroscopic fingerprinting methods had high accuracy when the small differences between sediment sources.The difference in fingerprinting results among different sieving particles confirmed that the selection of particles had an important impact on the accuracy of optical characteristics for tracing sediment sources.The < 125 μm sieved particle was the most suitable particle for the visible,near-infrared spectroscopy,and color parameters tracers.The < 31 μm sieved particles could provide the most accurate tracing results by using mid-infrared spectroscopy.Additionally,different fingerprinting results of the simulated particle sorting setup groups(MAE = 0.22)and no particle sorting setup groups(MAE = 0.09)indicated that extreme particle sorting had a negative effect on the accuracy of optical fingerprinting results.This study verified the good applicability of optical characteristics in sediment source research,analyzed the influence of different particle sorting degrees on the optical characteristics tracing sediment sources,provided a reference basis for improving the accuracy of optical characteristics tracing sediment sources,and was of great significance for the indepth study of the application of optical tracing method in sediment sources.