Algorithm Research of Soil Emissivity Extraction
程洁,肖青,李小文,柳钦火,杜永明
DOI: https://doi.org/10.3321/j.issn:1007-4619.2008.05.004
2008-01-01
Abstract:Temperature and emissivity are two important parameters 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 radiometric measurements relates to the problems of solving N+1 parameters with N equations.Some approximations or assumptions must 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.Most of these temperature and emissivity separation algorithms are designed for processing multi-spectral data.As far as hyperspectral FTIR data is concerned,their applicability needs to be evaluated.Moreover,we explores whether there is an optimal algorithm for retrieving soil emissivity from hyperspectral FTIR data.Five typical temperature emissivity methods(e.g.NEM,ISSTES,alpha residual method,MMD and TES) were investigated in this study by simulated dataset.The simulated dataset is composed of two parts,simulated ground-leaving radiance and simulated atmospheric downward radiance.Totally 58 soil directional hemispherical reflectance were obtained from the ASTER spectral library,and were converted to emissivities based on Kirchhoff's law.The soil temperature was assigned as 300K.The atmospheric downward radiance was simulated by MODTRAN4.0 in which the 1976 US atmosphere model was used.The simulated data was added a random Gauss noise with zero mean and standard deviation of 3.14e-9 W/cm2/sr/cm-1,which was the Noise Equivalent Spectral Radiance(NESR) of our spectrometer BOMEN MR 304 measured in laboratory.In order to evaluate these algorithms' sensitivity of response to the instrument random noise,the simulated data added with zero mean and standard deviation of 2,4,6,8,10,15,20 times of instrument NESR were also considered.On the basis of the result,we draw some valuable conclusions.For NEM,an optimal maximum emissivity of 0.985 is suggested,the RMSE of derived soil emissivities and mean absolute temperature is minimum with this maximum emissivity.A better empirical relationship has been discovered to substitute the original mean-minimum maximum difference relationship in MMD method.The alpha residual method is not suitable to retrieve soil emissivity from hyperspectral FTIR 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 times than ISSTES,so the original NEM module has been replaced by ISSTES to acquire the accurate initial value of emissivity in TES,the original power relationship in MMD module of TES has been replaced by a linear relationship for higher fit precision.As a conclusion,we find the ISSTES is the best method with the true instrument noise level,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 grow with the increase of instrument noise level.Finally,we present an example of soil emissivity extraction using five methods mentioned above with ground-based measured hyperspectral data,which were measured at our field test site with BOMEN MR 304 spectrometer on the afternoon of September 26,2005.The distribution of derived emissivity spectra verifies the results of algorithm analysis.