Spectral unmixing model to assess land cover fractions in Mongolian steppe regions
スポンサーリンク
概要
- 論文の詳細を見る
The land cover fractions (LCFs) and spectral reflectance of photosynthetic vegetation (PV), nonphotosynthetic vegetation (NPV), and bare soil were measured at 58 sites in semi-arid and arid regions of Mongolia in the summers of 2005 and 2006. These data sets allowed a detailed assessment of the impact of measurement geometry as represented by the solar zenith angle θs, sensor view zenith angle θv and azimuth view angle phi in the estimation of LCF values by means of the spectral unmixing model (SUM). The bidirectional distribution function (BRDF) was fitted to the reflectance data and then used to produce reflectance at various measurement geometries. LCFs from these reflectance data for a given combination of θs, θv, and phi were compared with visually determined LCFs. It was found that θs in the range of 30–45° produced a better agreement of LCFs. For θv, the agreement is not very sensitive to the choice of angle for the range 30–70°, although θv = 50° showed a slightly better performance. The azimuth view angle does not have strong influences to the LCF estimation, except for the case of phi = 180° (view toward the sun), which does not allow precise fitting of BRDF function over a tall vegetation site. Overall, this study verified the results of earlier studies obtained mostly for the American continents that SUM is capable of producing LCF estimates accurately and also found that its accuracy was, in general, much better than that by the more traditional approach of the supervised classification method (SCM) applied to images of a digital camera.