Derivation Of Remote Sensing Indices For Lithologic Recognition Using A Linear Approximation To The Planck Function And Development Of Lithologic Discriminant Space

Lithologic fine recognition is an essential and difficult issue in remote sensing petrology. This study aims to develop methodology, which is targeted and stable to the environment background, for lithologic fine recognition using remote sensing data. The main methods conclude the derivation of lithologic indices and construction of lithologic discriminant space. The study area is located in Qilian County, Qinghai Province, China, and belongs to the Qinghai-Tibet Plateau. The remote sensing images used in this study are ASTER and fully polarimetric RADARSAT-2 data.Thermal radiance of ASTER TIR bands correlates highly with each other. A linear equation that represents the relationship of the radiance between any two TIR bands was derived from the thermal infrared radiation transfer equation using a linear approximation to the Planck function. The linear equation can explain the correlation in theory. Moreover, the coefficients are determined by the emissivity of objects, atmosphere condition, etc. and are land surface temperature independent. It is to say that different objects may have different linear equations due to their different emissivity features, thus the linear equations can be used to detect specific objects. Because constrain of the parameters in the linear equations is complicated, this study used a statistic regression method as an alternative. Theoretical indices were derived based on the linear regression equation and regression residual characteristics for any two ASTER TIR bands. Two mafic-ultramafic rock indices(MI) and two quartzose rock indices(QI) were proposed, they satisfactorily map these rock units. In addition, the indices are insensitive to the variation of land surface temperature.However, the MI and QI can only recognize specific rocks, and there are also influences form other surface objects on these indices. For further lithologic mapping, multistage spectrum enhancement and multiple indicators are necessary. This study developed a 2-dimensional spectrum feature space using a QI and MI as the two dimensions of the feature space. All the samples show an approximate triangular shape in this space. Mafic-ultramafic rock, quartz-rich rock and carbonate rock occupy the three vertex regions separately, while felsic rock is located in the central region of the triangle. Silicate rocks show a linear belt in the space. Several classification functions were deduced based on the statistical characteristics of the samples. Confusion matrix of the classification results demonstrates the performance of this feature space is good. The producer‘s accuracies of all the surface objects, except for felsic rock, exceed 90%. The accuracy of recognizing felsic rock is about 80%. Comparison of the feature space with other felsic rock recognition method also shows the advantage of the feature space for extracting felsic rocks.This study introduced a feature-based synergy of remote sensing data for further lithologic recognition. Several features were extracted from the ASTER and RADARSAT-2 SLC images, including the relative content of mafic minerals(MI), the relative content of quartz(QI), the Cloude polarimetric decomposition parameters, i.e. scattering entropy H, and the mean scattering angle α. Four 2-dimensional feature spaces were generated for collaboratively analyzing these features. Bayes linear discriminant analysis was conducted on the four feature spaces. The H-MI space shows a good performance of recognizing serpentinite, with a producer‘s accuracy of 88.5%. Single use of the ASTER and RADARSAT-2 data cannot achieve such accuracy. The results give a prototype proof on the advantage of the feature-based synergy of the multisource remote sensing data for lithologic recognition.