Study On Dust-retention Capacity Of The Common Greening Plant In Guangzhou By Using Hyperspectral Remote Sensing

With the rapid development of industrialization and urbanization,urban air pollution is becoming increasingly serious.Urban greening plants can effectively block dust in the air and improve the quality of urban ecological environment.At present,the dust-retention effect of plants has become an important indicator for screening urban green plants.How to assess the dust-retention effect of plants has become the focus of current research.Hyperspectral remote sensing provides a highly efficient,convenient,and non-destructive technical means for the quantitative study of plant's dust-retention.In order to promote the application of remote sensing technology in the dust-retention effect of plants,this paper selected three common green plants—Ficus microcarpa L.f.cv'Golden leaves'(GL),Loropetalum chinensis(R.Br)Oliv.Var rubrum Yieh(LC),and Cordyline fruticosa(L.)A.Cheval(CF)in Guangzhou as the research object,according to the differences in blade surface characteristics.By using the hyperspectral remote sensing technology,the spectral characteristics of plant leaves under dust pollution were studied.The important hyperspectral variables were first identified by applying the random forest(RF)algorithm.Three estimation models were then developed using the support vector machine(SVM),classification and regression tree(CART),and RF algorithms.Based on the validation results,different methods are compared and analyzed to explore the optimal method for model inversion.Conclusions are as follow:(1)Dust-retention content increases with time,but after reaching saturation,the dust-retention content will no longer increase and even decrease.The saturated dust-retention content of the three plants were ranked as follows:CF(2.23 g/m2)>LC(2.21 g/m2)>GL(2.12 g/m2).(2)The visible and near-infrared bands(350-13 60nm)can effectively reflect the spectral characteristics of the leaf dust-retention.It are important bands for studying spectral characteristics,and are also important interval for the selection of inversion bands.And the first derivative of spectral reflectance is more suitable for inversion of the model.(3)The RF algorithm can effectively reduce the dimension of hyperspectral data.Compared with the methods,the machine learning algorithm is obviously superior to the Pearson correlation method.The SVM algorithm and RF algorithm are suitable for the inversion of dust-retention content,and the specific optimum is related to the type of plant.(4)There are different bands and methods for constructing high-precision inversion models for different plant species.For LC,which the number of inversion bands is 150,the optimal inversion method is RF algorithm.For CF,which the number of inversion bands is 74,the optimal inversion method is SVM algorithm.For GL,which the number of inversion bands is 80,and the optimal inversion method is SVM and RF algorithms,both can be adopted.