Hyperspectral Remote Sensing Information Extraction And BRDF Model Of Soil

With the development of remote sensing,hyperspectral remote sensing has been widely applied in a growing number of fields.The development of modem farming urgently requests that remote sensing technique to offer timely and accurate ground information.Hyperspectral remote sensing has the characteristics of narrow width and high band continuity,it can get continuous spectra of object,which makes the interested information extend to spectral dimension.Therefore, the hyperspectral data can be used to discriminate the subtle spectral features with high spectral resolution,and achieve the aim of accurate monitoring and inversion,consequently.Moreover, water content,organic matter content,surface roughness and texture of soil are the important information to agricultural practice.Many of studies indicate that the spectral characteristic highly related to the physical and chemical properties of soil..Therefore,the hyperspectral remote sensing has great potential of quantitatively retrieving the soil characteristics.Bidirectional reflection is one of the most basic macroscopical phenomenon,there have significant differences between object's surface reflectance as illumination and observation angle changes,some of the structural characteristics can be inferred from the variety of the observed shadow.Bidirectional reflection characteristic also plays important roles in the research of remote sensing model and inversion.Study of bidirectional reflection properties of soil is meaningful to the quantitative remote sensing and the development of remote sensing technology,which must be solved for the inversion problem of surface temperature,surface roughness,albedo,etc.Moreover, directional distributions of soil reflectance potentially carry information of soil properties such as moisture content,organic matter content,iron-oxide content,mineralogy,particle size distribution and surface roughness.Therefore,BRDF(Bidirectional Reflectance Distribution Function) mathematical model,model validation and multi-angle model inversion are the hot spot and difficulty for the research of quantitative remote sensing.This thesis focused on soil information extraction from hyperspectral data and the analysis methods.Then took hyperspectral remote sensing as technical support and put great emphasis on the study of retrieving soil characteristics from laboratory spectral data using different processing and modeling methods.The prediction models were established and some of the soil characteristics were inversed successfully.Bidirectional reflectance measured in the laboratory and field were simulated based on the radiative transfer model and geometric optics model, respectively,and parameters of these models were fitted well.The impacts of the soil surface conditions on model parameters was also studied in this thesis.Main content and conclusion as follows:(1)Smoothing effect to soil spectral reflectance with different filtering methodMoving average(MA),Middle-value(MV),Savitzky Golay(SG),Low-pass filter(LP), Gaussian filter(GS)and Wavelet denoising(WD)were used to smooth soil spectral curve,and smooth index(SI),horizontal features retain index(HFRI),vertical features retain index(VFRI) were established to evaluate the effect of smoothing.The results showed that the stronger of smoothing capacity inevitably lead to the poorer maintaining ability of horizontal and vertical features,and the better ability to maintain horizontal features lead to the better ability to maintain vertical features.Moreover,GS had the least SI,HFRI and VFRI,WD and MV were the best smoothing methods with better ability to maintain the characteristic of spectral curve,which could balance the contradiction of maintaining the characteristics and smoothing.The smoothing effects of MA and LP were not so good because of their strong ability of maintain the characteristics.Partial least squares regression method(PLSR)was used to build up the prediction model of saline soil sand content with the data filtered by above-mentioned different methods.The results showed that model of WD had the highest precision with less factors,conversely,the model of MA,GS and SG,which used more factors,had lower prediction precision.The model of WD and MV,which with good balance of smoothing and features maintaining,predicted the sand content with higher accuracy.The model of MA,which with worst smoothing ability,had the lowest accuracy of prediction.It was suggested that the smoothing ability was the main factor affecting the precision of sand content prediction.But the effects of features maintain ability could not be ignored.(2)Sand content prediction of saline soil using different spectral data processing and modeling methods After filtered by the WD and resampled with 10 nm interval,soil reflectance data were processed with normalization(NOR),first differential(FD),baseline correct(BL),standardization (SNV)and multiple scattering correct(MSC),with the original data(NO),were modeled to predict the sand content of saline soil using two linear model,PLSR and Principal component regression(PCR),and two nonlinear model,including Artificial neural networks(ANN)and Support vector machine(SVM).The prediction precision was affected obviously by the different data preprocessing methods.The SNV and MSC methods gave the best prediction precision,FD method was worst.It was noted that the prediction accuracy were decreased after processed by NOR,FD and BL.To the linear model,discrimination of prediction precision was not so far after different preprocessing methods except BL,it indicated that these two linear model were more stable.Moreover,data preprocessing methods had great impact on the two nonlinear model,which made them very unstable.(3)Organic matter content prediction of paddy soil using different spectral data processing and modeling methodsOrganic matter(OM)content prediction models of paddy soil were established using soil spectral reflectance and its different transformations.Results of single correlation analysis revealed that the bands,which correlated significantly with organic matter content,were all in visible waveband.The correlation between first differential of reflectance's reciprocal and OM content had been significantly improved but did not increase the accuracy of the prediction model accordingly,and model based on the reciprocal of logarithm of the reflectance predicted the OM content with the highest accuracy.Different transformations of spectral reflectance lead to different prediction precision to the modeling samples and test samples.Prediction precision of the PLSR models,which needs less factor and convergence faster,was superior to the PCR models.(4)Soil water content prediction using different spectral data processing and modeling methodsWater content prediction models of different soils were established using soil spectral reflectance and transformations.The results showed that the effect of water content on the reflectance was same as the conclusions of previous studies:for all the wavelengths and all the soils,the reflectance decreased when the moisture increased for low soil moisture levels. Conversely,after a critical point,soil reflectance increased with soil moisture.The wavebands that highly correlated with water content were also located nearby 1450 and 1950 nm,which were known as the classical water absorption wavebands.Logarithm of reflectance improved the correlation with water content and first differential reduce it.Water content prediction model established with the water absorption band nearby 1450nm was more effective than that nearby 1950nm.For the water content prediction using PCR and PLSR model,logarithm transformation of reflectance was superior to other spectral variables.(5)Soil bidirectional reflectance varied with illumination,observation angles and the effect factorsIt was showed that the soil bidirectional reflectance increased with increasing off-nadir view angles for all azimuth directions and azimuthally symmetric in the orthogonal plane.Additionally, the bidrectional reflectance was highest in backscattering direction,lowest in forward scattering direction,and increased with the solar zenith decreasing.All of these changes were related to shadow observed in field of view of detector varied with the illumination and viewing angles.Soil surface roughness and moisture were the most important factors affecting the directional reflectance of bare soil,it was found that bidirectional reflectance decreased with the increasing soil surface roughness,and the soil surface showed much more non-Lambert characteristic. Furthermore,impact of water content on the bidirectional reflectance was same as it on the reflectance observed in zenith.(6)Simulation of soil bidirectional reflectance based on the radiative transfer model and inversion of the model parametersThe optimisation procedure of parameter values of SOILSPEC model,which derived from Hapke's model based on radiative transfer theory,was not sensitive to the arbitrary initial estimates,and the given bidirectional reflectance could be simulated with good agreement.Single scattering albedo(SSA,ω)for wavelength between 400 nm and 1400 nm calculated by the model increased as becoming dry of the soil,andωwas independent of the illumination and observation conditions.Value of parameter h increased with rougher soil surface,which increased as the augment of particle size,and parameter h did not vary a lot with wavelength.Moreover,the scattering type of soil surface was relative to the surface condition.Bidirectional reflectance of soil with different surface condition measured in the laboratory could be simulated very well. However,simulation of bidirectional reflectance of raw saline soil measured in field was not so good,especially for the data measured in large solar zenith angles. (7)Simulation of soil bidirectional reflectance based on the geometric optics model and inversion of the model parametersBidirectional reflectance factor R predicted by Irons geometric optics model reached absolute maximum when phase angle at zero and strong backscattering in the anti-solar direction from the soil surfaces.Bidirectional reflectance factor decreased as sphere area index L increased and the sensitivity of the predicted reflectance factor to viewing zenith angle decreased as diffuse illumination f increased in the principal plane.Moreover,the position where R reached maximum was not sensitive to the variance of f.Both isotropic bidirectional reflectance factor P and L decreased with increasing soil water content.But after a critical point,P also increased with soil moisture.L increased as soil surface roughness increased and P decreased contrarily.Bidirectional reflectance measured in the laboratory could be simulated well using this model.However,the calculated bidirectional reflectance less than the measured values as viewing zenith angle increased.Accuracy of this model was relatively poorer than that of radiative transfer model.Four possible innovations or new developments were made as follows:(1)Different spectral reflectance filtering,preprocessing and modeling methods were synthetically utilized to establish the soil characteristics predicting models,and accuracy of these methods were also compared to identify the best prediction model.It could be used as new reference to the processing of soil spectral data and information extraction.(2)Regulation of the soil bidirectional reflectance varied with the viewing zenith angle, azimuth angle and solar zenith angle were studied based on the measurements of soil samples with different surface condition and raw saline soils.This study provided foundations for the development of new soil BRDF model and inversion of soil characteristics.(3)Parameters of two kinds of BRDF model were fitted using the bidirectional reflectance of soil samples with different surface roughness and moisture,and impacts of soil surface condition on study of the parameters retrieve and soil BRDF characteristic were revealed in this study.It could offer new research ideas to the research of BRDF characteristic of soil under natural state in the field and the inversion of surface properties,moreover,provide research foundations to improve the accuracy of quantitative remote sensing inversion of soil.(4)Bidirectional reflectance of raw saline soil measured in the field were simulate successfully using BRDF model based radiative transfer theory.It could exploit the practical potentials of new generation of multi-angle sensors to detect characteristics of soils under natural or tillage conditions,and provide new research foundation to improve the accuracy of remote sensing of soil, quantitative inversion of soil characteristics.Meanwhile,it offers the certain basis for the simulation of multi-angle image and the development of new sensors.