Hyperspectral estimation model of dust deposition content on plant leaves

Abstract: Dustfall is an important indicator to characterize the regional atmospheric environment quality. The dustfall status and regional environmental quality can be reflected directly by dust deposition content on plant leaves. The acquisition of hyperspectral data measured at ground surface is more and more convenient in recent years with the development of hyperspectral technology. Study on inversion model for foliar dust deposition content based on hyperspectral data will improve the efficiency of atmospheric dust monitoring and spatial sampling. And the model can not only be used as an effective complement to traditional atmospheric dust monitoring means, but also improve the time accuracy and spatial accuracy of dustfall monitoring. The aim of exploring the construction of hyperspectral estimation model of foliar dust deposition content is to promote the application of hyperspectral and remote sensing techniques on dustfall monitoring, and provide theoretical basis for the quantitative monitoring of regional dustfall based on the ground hyperspectral data. The adult Beijing poplar leaves were collected in Beijing urban area during the period from September 17 to September 18 in 2014. In order to collect the hyperspectral data and the data of per unit area dust deposition content on leaf samples, the work was carried out in the following sequence: spectral measurement, weighing, dust removal, weighing, spectral measurement, and measurement of leaf area. We finally got 59 valid sample data. We analyzed the influence of foliar dust deposition content on the spectral reflectance and trilateral parameters of poplar leaves. And the relationships between leaf spectral characteristics and foliar dust deposition content were studied. Then, the estimation model of foliar dust deposition content based on spectral parameters was established. The results showed that foliar dust deposition content enhanced the reflectivity of 400-700 nm band and inhibited the reflectivity of 710-1110 nm band. And foliar dust deposition content had no obvious effect on red edge position, yellow edge position and blue edge position. The linear relationship between spectral reflectance of the near infrared wavelengths (730-1000 nm) and foliar dust deposition content was obvious, and the coefficient of each band was higher than 0.7. The reflectance of green band was not sensitive to the influence of leaf dust deposition content. And the relationships between red edge amplitude, red edge area and leaf dust deposition content achieved significant relation. Three correlation coefficients' matrices were constructed by the indices calculated based on different spectra reflectance and foliar dust deposition content. The maximum value in each matrix was higher than the maximum value of correlation coefficient between single band and foliar dust deposition content. The highest value of correlation coefficients of the 3 matrices was 0.7615, which was in the matrix of correlation coefficient of normalized index and foliar dust deposition content. Models based on multivariate linear regression, principal component regression and partial least squares regression all had stronger ability to predict. The partial least squares regression model with the independent variables, which included spectral reflectance at the band of 749, 644 and 514 nm, red edge slope, red edge area, normalized difference index composed by the band of 924 and 1010 nm, difference index composed by the band of 713 and 725 nm, and normalized difference vegetation index composed by the band of 749 and 644 nm, had the highest estimate accuracy with the modeling decision coefficient of 0.734, the forecasting decision coefficient of 0.731, and the root mean square error of 0.311 for prediction.