Analysis And Evaluation Of Heavy-metal Pollution Stress Based On Chlorophyll Content Of Rice Using Hyperspectral Data

Heavy metal pollution is one of the major ecological problems facing the world today, not only hinder the growth of crops, also cause a harm to human body through the food chain of transfer and enrichment indirectly. Hence how to obtain large area crop pollution of heavy metals stress state accurately is one of the major research direction of agricultural quantitative remote sensing. Chlorophyll content of crop leaf is usually considered as a good indicator to measure crop growth period, photosynthetic capacity and heavy metal stress condition. This paper, choosing four different pollution situation farmland of heavy metal in Changchun-Jilin area as experimental sample plots, obtained the spectral indexes which sensitive to changes of chlorophyll content as inversion model parameters for chlorophyll content by controlling the single variable method in the process of simulating crop leaf spectral with leaf optical radiation transmission model(PROSPECT), built RF- PROSPECT regression model for leaf chlorophyll content inversion through machine learning algorithms--random forests and evaluated precision of the model, divided the leaf chlorophyll content range shows different crops polluted by heavy metal stress degree according to the correlation analysis between the measured chlorophyll content and the pollution level of the sampling area, explained the relative importance of heavy metal elements to chlorophyll content through the random forest classification model combining with the measured soil heavy metal content. Which realize the pollution area crop monitoring of photosynthesis ability and pollution levels so that pollution control measures would be put forward as soon as possible, is of great significance to food security problem. The main research conclusions are as follows:1.The three indexes, modified chlorophyll absorption in reflectance( MCARI) 、chlorophyll absorption in reflectance(CARI)and green normalized difference vegetation index(GNDVI) are sensitive to the change of chlorophyll content and weak sensitivity of other ingredients among the six experience indexes in our transmission model of simulating crop leaf spectral with leaf optical radiation. So we choose these three indexes as the spectral indexes of inversion model.2.The model achieved a good inversion result, getting the R2 of RF- PROSPECT inversion model is 0.874, the average absolute error is 1.34(ug/cm2), the average relative error is 7.83% by training the model again and again in the regression model of random forest algorithm.3.Pollution could be rated into no pollution, mild pollution and moderate pollution, the chlorophyll content of corresponding domain respectively defined as greater than 30 ug/cm2, 20 ~ 30 ug/cm2, 0 ~ 20 ug/cm2 according to the correlation analysis between the measured chlorophyll content and the pollution level of the sampling area,using chlorophyll content value to evaluate heavy metal pollution of crops stress level indirectly and obtained the relative importance of heavy metal elements to chlorophyll content through the random forest classification model based on measured data as: Cr>Cu>As>Cd.