| Among different Gobi characteristics, surface material composition(gravel size composition) not only directly affect the nature of other characteristics, but also largely determine the difficulty of transformation and utilization, and it is the basis and premise to carry out the study of Gobi. The traditional method for obtaining the surface gravel size in Gobi is the field survey method, which is only applicable to small scale areas. The development of remote sensing technology provides a new technical support for obtaining the surface particle size, and provides possibility for the acquisition of gravel size distribution in large areas accurately. However, because of the low spectral contrast between different size of gravels, and the gravel spectra are greatly affected by environmental factors, the accurate remote sensing inversion of the surface grain size still faces the following problems:(1) current researches focus on the grain size of sandy surface and sea(river), while the lack of researches about the grain size of gravel in Gobi. And the remote sensing data used in inversion are mainly multi spectral images, which can not discriminate the spectrum of different size of gravels in complex environment effectively;(2) there are many researches on the physical and chemical properties of soil, vegetation index, etc. However, there are few research on hyperspectral inversion of Gobi surface gravel using specific spectral absorption parameters. In addition to the problems of the inversion of surface grain size, there are few studies on the spatial distribution of surface gravel in Gobi.Hyperspectral remote sensing technology is an important technical breakthrough in the field of remote sensing in the late 20 th century. Hyperspectral remote sensing image has a large number of bands, and the bandwidth is narrow(band width <10nm). The researchers can obtain continuous spectral curve of ground objects from hyperspectral image. In recent years, hyperspectral inversion of ground objects has great development. And there are great development of ground object retrieval technology based on field measured spectrum and Airborne hyperspectral data. However, there are lack of inversion studies combined with satellite hyperspectral data. The high spectral resolution characteristics of the satellite borne hyperspectral images have advantages in the quantitative inversion of the Gobi surface gravel size.In view of the above reasons, the gravel particle size of Gobi area were divided into six grades(d =0.8cm,d =3.4cm,d =16.3cm,d =41cm,d =53cm,d =83cm), EO-1 Hyperion image and spectral mixture analysis method were used to carry on the analysis. So that abundance images of different size of gravel can be obtained through image unmixing. In this study, we found that spectral bands and iron oxides and Al-OH absorption features that are sensitive to gravel size can be extracted after derivative transform and continuum removal. Therefore, in this study, the inversion models of gravel size was established based on sensitive spectral bands and the spectral absorption parameters respectively. And the spectral absorption parameters of the band that has greatest correlationship with gravel size were selected and applied to the abundance image after the unmixing of the Hyperion data to establish inversion model of surface gravel particle size in Gobi area. And the spatial distribution map of gravel size was rendered. Then different inversion methods and their accuracy were compared with each other. At the same time, in combination with the spatial statistics, the spatial heterogeneity of different size of gravels in the study area was analyzed through the analysis of the granularity parameters of the gravels. The results of this study are as follows:(1) the correlation between the spectral reflectance and the particle size was better after the derivative transformation. The best correlation bands were 908 nm, 983 nm and 985 nm. And it was found that the one variable cubic regression model had a better fitting precision, and the logarithmic derivative performed best(R2=0.851) in the regression analysis, and the prediction accuracy was higher(75.27%). Thus, the inversion model of surface gravel size in Gobi area was established based on the spectral differential transformation of the gravel spectra. The formula was: Y= 110.667–39312.858 b908+ 286870.05 b9082– 5.77E11 b9083;(2) The empirical model used to express the relationship between the absorption parameters(absorption depth, width and area) in NIR and Al-OH regions of the continuum removal spectra and the size of the gravel samples showed good correlation. Through the correlation coefficient R2 and RMSE, we found that the simulation results of derivative spectra and the logarithmic differential model of gravel size(RMSE=24.9,R2=75.27%) is lower than the absorption parameters(RMSE < 8, R2> 80%). The parameter of optimal inversion model of gravel size in the study area was band depth(BD) of the 908 nm band in NIR region, which had good prediction accuracy(R2=0.8048);(3) spectral mixture analysis method was applied on EO-1 Hyperion data to generate abundance images of different gravel size, and the spectral absorption parameters of the sensitive bands of surface gravels were applied to the abundance images. After the inversion study of gravel size in the study area, we found that in the prediction models of 6 absorption feature parameters, the determination coefficient R2(0.8817)of the band depth(BD) of 908 nm in NIR region was the biggest, and the RMSE in modeling and forecasting were the smallest(0.039 and 0.047). The Gobi surface grain size inversion model that was established based on the abundance images was: Y = 89.38BD2- 95.522BD+ 13.912. According to the calculated results of the model, we found that the trend of grain size distribution obtained from the inversion modle and the actual sampling results were basically consistent.(4) in the sample sites selected in this study, the gravel content showed that: medium gravel content decreased and coarse gravel content increased gradually with altitude increasing. The performance of average particle size showed that the average particle size increased with altitude increasing. The sorting coefficient showed that most of gravel had good separation level(0.28 ~ 0.53). And the surface gravel coverage distribution was relatively uniform, with Gobi of medium gravel coverage dominating, which also has the most extensive distribution area.(5) the 0.8cm, 41 cm and 83 cm gravels were not suitable for the use of variogram model in predicting spatial variation of grain size. It illustrated that the gravel size was mainly affected by random factors, and the spatial autocorrelation was weak; the regionalized variables of 0.8cm, 3.4cm, 16.3 cm and 53 cm gravels were strongly autocorrelated, and spatial heterogeneity caused by random factors only only accounted for a small proportion; the spatial autocorrelation range of the 41 cm gravel was maximum(1340m). The uniformity of saptial distribution was gradually weakened with the spatial autocorrelation getting smaller. On the other hand, the uniformity of spatial distribution of gravel increased.Overall, in this study, we found and made full use of the absorption parameters of different gravel spectra, and the grain size inversion model based on the EO-1 Hyperion data was in high precision, and the application range of the hyperspectral remote sensing was extended. At the same time, the spatial distribution of the Gobi surface gravels was concluded, which could provide reference for the sand prevention and control engineering. |
PDF Full Text Request