Estimation Of Exposed Carbonate Rock Fraction With Remote Sensing Imagery

Karst rocky desertification,as one of the most serious problems in land degradation,occurs and develops quickly under the impact of anthropogenic activities in Yunnan.There is the need for more work dedicated to understanding human-altered karst landscapes via assessing alterations of natural systems and the repair of karst ecosystems,measuring karst disturbance and sustainability,and mapping and monitoring karst land cover.The desertification has been characterized as the processes of forest destruction,soil erosion and basement rock exposure.Therefore,the estimation of fractional exposed carbonate rocks cover(FECR)is essential for the need.Remote sensing techniques provide an important way for understanding karst environment.In recent years,remotely sensed data have been applied extensively to monitor karst environment in large geographic areas.Despite all this,the mapping of exposed carbonate rocks with remote sensing is still a challenging task because of the highly heterogeneous landscapes in the desertification areas and the limitation in the spatial and spectral resolution of remote sensing images.This is especially true in the context of multi-spectral images with medium spatial resolution such as those provided by Landsat satellites.Because of the presence of spectral mixing in the pixels of these images.Therefore,it is very necessary to study the theories and methods for the quantitative estimation of FECR by using moderate or low resolution remote sensing data.Karst rocky desertification is very serious in Jianshui County,in the province of Yunnan,China,which,therefor,has become our study site.The typical surface constituents(including exposed carbonate bedrocks,soils,green vegetation and NPV)were collected and measured by a portable spectrometer(Analytical Spectral Devices(ASD)Inc.,FieldSpec 4,Boulder,CO,USA).The remote sensing data include Landsat-8 OLI imagery and Worldview 2 imagery.Our study was implemented following a procedure and detailed in the sections that follow.First,we compared the main types of remote sensing methods for estimation of fractional land cover,and summarized their advantages and disadvantages.Subsequently,we sorted out the details on Linear Spectral Estimation of Fractional Cover(LSEFC).Second,several spectral analysis functions,such as continuum-removal and class separability measure,were used to explore the optimal spectral features for distinguishing carbonate rock in karst,which were used to develop hyperspectral or multispectral carbonate rock indices for the estimation.In the next step,we explored the potential of metric learning in the estimation of fractional land cover using remote sensing data,and the cosine similarity metric learning was introduced into estimating FECR.Finally,the proposed methods in this thesis were analyzed and validated by simulation experiments and a case study in southern Yunnan using remote sensing imagery.The conclusions in this thesis are summarized as follows:(1)Three common methods of estimation of fractional land cover using remote sensing data were outlined,including empirical statistical models,semi-empirical spectral index models,and spectral mixture analysis models.We examined the pros and cons of each method.For utilization of the advantages of various types of methods,the physical mechanisms and convex geometry of the spectral linear spectral mixture analysis model(LSMA)were detailed firstly.Accordingly,the relationship between estimation of fractional land cover and LSMA was established by a large number of mathematic reductions.With the help of the properties of Metric Space and Measurement Topology in point set theory,we built up a general guideline and mathematical expression for LSEFC,which became the main guidance on how to estimate FECR in karst.(2)In this study,we examined the spectra of major surface constituents in karst areas for direct evidence of absorption features attributable to exposed carbonate rock fraction.Utilizing linear continuum removal,we applied spectral feature analysis to examine absorption features in the mixed spectra.Then we observed that there are overlapping spectral absorption in 2.149-2.398?m by soils,non-photosynthetic vegetation and exposed carbonate rock.These overlapping features complicated the carbonate absorption feature near 2.340?m in synthetic mixed spectra.According to LSEFC criteria,we removed the overlapping signal by adjusting the sensitive bands and developed two hyperspectral carbonate rock indices(HCRIs).(3)In order to develop multi-spectral methods for FECR estimation,we analyzed the multi-spectral features of carbonate rocks,soil and vegetation in KRD area,it was found that,comparing with the soil and vegetation,carbonate rock had smaller reflectance difference between the blue and near-infrared(NIR)bands.At the same time,after discussion of the class separability between carbonate rock and background and their scatter distribution in different OLI band feature spaces,the blue and NIR band combination was considered as the best one for FECR estimation among the all two OLI band combinations.Under the LSEFC guideline,tangent transformation of the two bands was introduced to construct the multi-spectral carbonate rock index for FECR estimation.(4)With the general criterion and mathematical expression of LSEFC,the Cosine Similarity Metric Learning(CSML)method was introduced to the FECR estimation and a bilinear method(CosEFC)was developed.The method can effectively deal with the estimation error caused by the inner difference in the same class,and it also improved the reliability and stability of FECR estimation.The case of Jianshui County showed that the FECR estimation result with CosFEC from OLI imagery was reasonable,and which was largely consistent with the expectations of exposed carbonate rock distribution based on our knowledge of the region.In this study,on the basis of the LSMA model,we put forward a general criterion and mathematical expression of linear spectral estimation of fractional cover,and it then became an important guidance for the estimation of fractional land cover including FECR.