土壤的发射率具有较大的不确定性,为了准确提取土壤的发射率,利用ASTER光谱库中的58条土壤光谱,模拟产生了热红外高光谱数据集,利用这些数据进行了土壤的发射率提取试验,分析了较为典型的几种温度发射率分离方法,如NEM、ISSTES、α剩余法、MMD、TES在土壤发射率提取中的适用性、稳定性和精度,并根据分析的结果对各种算法在土壤发射率反演中的应用进行了相应改进.对于NEM方法,给出了最优的最大发射率;对于MMD方法,提出了一种比原平均-最小最大发射率之差更好的经验关系;在TES方法中,使用ISSTES代替原先的NEM方法,获得了精确的发射率初始值.基于模拟数据的算法分析结果表明,对于地面测量高光谱数据的土壤发射率信息提取,ISSTES准确度最高.最后给出了使用这5种方法由地面实测高光谱数据提取的土壤发射率光谱实例,提取的发射率光谱的分布情况很好印证了基于模拟数据的算法分析结果.

Temperature and emissivity are two mi portantparameters of thermal infrared remote sensing. Surface emitted radiance is a function of both its kinetic temperature and its spectral emissivity. Temperature and emissivity separation from radiometricmeasurements relates to the problems ofsolvingN+1 parameterswithN equations. Some approxmi ations or assumptionsmust be taken to reduce the number of unknown parameters and make the equation complete. Many temperature and emissivity separation algorithms have been put forward according to the different strategies. Mostof these temperature and emissivity separation algorithms are designed for processingmulti-spectral data. As far as hyperspectral FTIR data is concerned, their applicability needs to be evaluated. Moreover, we exploreswhether there is an optmi al algorithm for retrieving soil emissivity from hyperspectralFTIR data.Five typical temperature emissivitymethods (e.g.NEM,ISSTES, alpha residualmethod, MMD and TES) were investigated in this study by smi ulated dataset. The smi ulated dataset is composed of two parts, smi ulated ground-leaving radiance and smi ulated atmospheric downward radiance. Totally 58 soildirectionalhemispherical reflectancewere obtained from the ASTER spectral library,and were converted to emissivities based on Kirchhoffs' law. The soil temperaturewas assigned as300K. The atmospheric downward radiancewas smi ulated byMODTRAN4.0 inwhich the1976US atmosphere modelwas used. The smi ulated datawas added a random Gaussnoisewith zeromean and standard deviation of3.14e-9W/cm2/sr/cm-1,whichwas theNoiseEquivalentSpectralRadiance (NESR) ofourspectrometer BOMENMR 304measured in laboratory. In order to evaluate these algorithms sensitivity of response to the instrument random noise, the smi ulated data added with zero mean and standard deviation of 2,4,6,8,10,15,20 tmies of instrumentNESR were also considered.On the basis of the result, we draw some valuable conclusions. ForNEM,an optmi almaxmi um emissivity of0.985 is suggested, the RMSE of derived soil emissivities and mean absolute temperature is minmi um with this maxmi um emissivity.A better empirical relationship has been discovered to substitute the original mean-minmi um maxmi um difference relationship in MMD method. The alpha residual method is not suitable to retrieve soil emissivity from hyperspectralFTIR data. By comparing the accuracy of NEM and ISSTES, we find that the RMSE of derived soil emissivities suing NEM under ideal condition is two tmi es than ISSTES, so the original NEM module hasbeen replaced by ISSTES to acquire the accurate initial value of emissivity in TES, the original power relationship inMMD module of TES has been replaced by a linear relationship forhigher fitprecision. As a conclusion, we find the ISSTES is the bestmethod with the true instrument noise leve,l the RMSE of derived soil emissivities is only 0.0007 and the mean absolute temperature bias is only 0.02K.The RMSE of derived soil emissivities and the mean absolute temperature bias monotonically growwith the increase of instrumentnoise leve.l Finally, we presentan example ofsoilemissivity extraction using fivemethodsmentioned abovewith ground-basedmeasured hyperspectraldata, whichweremeasured atour field test sitewith BOMENMR 304 spectrometer on the afternoon of September 26, 2005. The distribution of derived emissivity spectra verifies the results of algorithm analysis.