Estimation for sparse vegetation information in desertification region based on Tiangong-1 hyperspectral image.

In order to estimate the sparse vegetation information accurately in desertification region, taking southeast of Sunite Right Banner, Inner Mongolia, as the test site and Tiangong-1 hyperspectral image as the main data, sparse vegetation coverage and biomass were retrieved based on normalized difference vegetation index (NDVI) and soil adjusted vegetation index (SAVI), combined with the field investigation data. Then the advantages and disadvantages between them were compared. Firstly, the correlation between vegetation indexes and vegetation coverage under different bands combination was analyzed, as well as the biomass. Secondly, the best bands combination was determined when the maximum correlation coefficient turned up between vegetation indexes (VI) and vegetation parameters. It showed that the maximum correlation coefficient between vegetation parameters and NDVI could reach as high as 0.7, while that of SAVI could nearly reach 0.8. The center wavelength of red band in the best bands combination for NDVI was 630nm, and that of the near infrared (NIR) band was 910 nm. Whereas, when the center wavelength was 620 and 920 nm respectively, they were the best combination for SAVI. Finally, the linear regression models were established to retrieve vegetation coverage and biomass based on Tiangong-1 VIs. R2 of all models was more than 0.5, while that of the model based on SAVI was higher than that based on NDVI, especially, the R2 of vegetation coverage retrieve model based on SAVI was as high as 0.59. By intersection validation, the standard errors RMSE based on SAVI models were lower than that of the model based on NDVI. The results showed that the abundant spectral information of Tiangong-1 hyperspectral image can reflect the actual vegetaion condition effectively, and SAVI can estimate the sparse vegetation information more accurately than NDVI in desertification region.

[Analysis of sensitive spectral bands for burning status detection using hyper-spectral images of Tiangong-01].

To obtain the sensitive spectral bands for detection of information on 4 kinds of burning status, i. e. flaming, smoldering, smoke, and fire scar, with satellite data, analysis was conducted to identify suitable satellite spectral bands for detection of information on these 4 kinds of burning status by using hyper-spectrum images of Tiangong-01 (TG-01) and employing a method combining statistics and spectral analysis. The results show that: in the hyper-spectral images of TG-01, the spectral bands differ obviously for detection of these 4 kinds of burning status; in all hyper-spectral short-wave infrared channels, the reflectance of flaming is higher than that of all other 3 kinds of burning status, and the reflectance of smoke is the lowest; the reflectance of smoke is higher than that of all other 3 kinds of burning status in the channels corresponding to hyper-spectral visible near-infrared and panchromatic sensors. For spectral band selection, more suitable spectral bands for flaming detection are 1 000.0-1 956.0 and 2 020.0-2 400.0 nm; the suitable spectral bands for identifying smoldering are 930.0-1 000.0 and 1 084.0-2 400.0 nm; the suitable spectral bands for smoke detection is in 400.0-920.0 nm; for fire scar detection, it is suitable to select bands with central wavelengths of 900.0-930.0 and 1 300.0-2 400.0 nm, and then to combine them to construct a detection model.

[The changes of forest canopy spectral reflectance with seasons in Xiaoxing'anling].

The ASD FieldSpec portable spectrometer was adopted to collect canopy reflectance spectrum data of the 9 main tree species in study area by a long-term observation to get the data of the four seasons Then the smoothed reflectance curve and the first derivation curve from 350 to 1400 nm and several commonly used vegetation spectral characteristic parameters were generated to analyse seasonal change characteristics and variation of the 9 tree species in visible and near-infrared band and to explore the best band characteristics and period for species identification. The results showed that different trees had different and rather unique spectral features during the four seasons. The spectral characteristics of the deciduous trees have regular changes with the cycle of the seasons, whereas those of the evergreen tree species have no significant changes in one year. As well changes in the spectral characteristics could effectively reflect forest phenology changes, and it is proposed that the optimal strategy for tree species classification may be the integration and analysis of multi-seasonal spectral data. Evergreen trees and deciduous trees in the winter have obvious differences in the canopy spectral characteristics and the best single-season remote sensing data for tree species recognition is in summer.

[Estimation of sparse vegetation coverage in arid region based on hyperspectral mixed pixel decomposition].

Based on Hyperion hyperspectral image data, the image-derived shifting sand, false-Gobi spectra, and field-measured sparse vegetation spectra were taken as endmembers, and the sparse vegetation coverage (< 40%) in Minqin oasis-desert transitional zone of Gansu Province was estimated by using fully constrained linear spectral mixture model (LSMM) and non-constrained LSMM, respectively. The results showed that the sparse vegetation fraction based on fully constrained LSMM described the actual sparse vegetation distribution. The differences between sparse vegetation fraction and field-measured vegetation coverage were less than 5% for all samples, and the RMSE was 3.0681. However, the sparse vegetation fraction based on non-constrained LSMM was lower than the field-measured vegetation coverage obviously, and the correlation between them was poor, with a low R2 of 0.5855. Compared with McGwire's corresponding research, the sparse vegetation coverage estimation in this study was more accurate and reliable, having expansive prospect for application in the future.