| Water is one of the most important conditions for maintaining the normal physiological functions of plants and ensuring life activities.Organic matter is an important material composition of the soil.It has important practical significance for crop growth and soil fertility maintenance.The traditional method of determination of soil moisture and organic matter are drying weighing method and laboratory chemical assay.Although the results obtained by these methods are better,it is time-consuming and energy consuming,and the representative is poor.Hyper-spectral remote sensing has many characteristics,such as real time,high efficiency,multi-band and large amount of data.It can get the detailed spectral characteristics of land objects.And it is of great practical significance to the efficient management of farmland.This study takes Tai'an city of Shandong Province as the study area,and takes 90 brown soil samples as the research object.The qualitative and quantitative analysis of the effects of water and organic matter on the original,transformed spectrum of the soil are analyzed according to the group comparison method,correlation analysis and two factor variance interaction analysis.Based on the law of interaction between soil water and organic matter,characteristic factors are selected to establish the spectral estimation model of soil organic matter by BP neural network,multiple stepwise regression and partial least-squares regression.The effectiveness of the factors selected according to the interaction law of the two is verified to improve the accuracy of the spectral estimation model.The main research contents and conclusions are as follows:?1?The effect and interaction of water and organic matter in brown soil on spectral are discussed.First,according to the experimental data,we draw the average spectral curve of nine groups and analyze the spectral characteristics of the group.Then,correlation analysis method is used to analyze the correlation between soil water and organic matter and soil spectrum.Finally,two-factor analysis of variance?ANOVA?is used to investigate whether the state changes of soil water content and organic matter will lead to the change of spectral reflectance index,and then explore their influence on soil spectral reflectance.The results show that the interaction between water and organic matter exists objectively.And the effect on soil spectrum from large to small is water,organic matter and interaction.The effect of water on soil spectrum is about 5 to 8 times in 4251 800 nm and 8 to 12 times in 1 9502 300 nm than that of organic matter.The effect of organic matter on soil spectrum is about 2 times as much as interaction in 3502 500 nm.It shows that organic matter has little effect on the spectral response of soil water,and its effect can be ignored when water content information is estimated.However,when estimating organic matter,the effect of water on the spectrum must be eliminated.?2?The sensitive bands and the inversion factors of organic matter of brown soil are determined.First,the reflective spectrum is transformed by using 6 kinds of spectral transform technology such as square root,square,logarithm,reciprocal,first-order differential and combination transformation for improving the correlation between soil organic matter and spectral reflectance.Then,the correlation coefficient between organic matter and spectral reflectance is calculated.Last,the characteristic factors of organic matter are selected based on two aspects.On the one hand,the characteristic factors are selected using the principle of maximum correlation.On the other hand,the characteristic factors are selected again according to the principle that organic matter has a larger role and water has a smaller effect and the interaction between the two is smaller.The results show that the correlation between soil organic matter and spectrum is significantly improved after the original spectrum is transformed by the above methods.The maximum value of the correlation coefficient is increased from 0.38 to 0.50.The spectral response area of organic matter is mainly located near the visible light wavelengths of 680720nm and near infrared wavelengths of7801200nm and 14001800nm.The feature factor selected based on the two methods is located in the above range.The feature factors are?R?'1091,R1?46 7,?RlnR?1?452,?RlnR?1?74 4,1?lnR?1?69 9,?R2?1?45 1,?R2?1?74 4,?eR?7?8 8 and?R?'718,R 7?8 8,?RlnR?7?8 8,?RlnR?1?74 4,1?lnR?7?8 8,?R2?6?9 7,?R2?1?74 4,?eR?1?70 2.?3?A hyper-spectral estimation model of soil organic matter based on the interaction of moisture and organic matter is established.The spectral estimation model of soil organic matter based on BP neural network,multiple stepwise regression and partial least squares regression are established respectively.For the feature factor selected based on the principle of maximum correlation,the determination coefficients of test samples are 0.540,0.676 and 0.676,respectively,and the average relative errors are 14.816%,7.899%and 8.055%,respectively.For the feature factor selected based on the interaction rule between them,the determination coefficients of test samples are 0.601,0.793 and 0.793,respectively,and the average relative errors are 12.751%,6.682%and 6.680%,respectively.The results show that the accuracy of the soil organic matter spectral estimation model based on the interaction law of moisture and organic matter has been greatly improved,which shows that the interaction of the two factors can not be ignored.Therefore,in the process of soil organic matter inversion,the interaction between the two must be considered. |
PDF Full Text Request