Estimating fractional vegetation cover (FVC) using satellite vegetation indices and digital photo image in Mongolia

Abstract:

Fractional vegetation cover (FVC) is a useful index of monitoring land cover dynamics and land surface energy partitioning into sensible and latent heats from satellite because it can be estimated by using satellite-based spectral vegetation indices (VI), such as NDVI and EVI. The relationship between FVC and vegetation indices is however variable depending on regional vegetation types and background soil types across different regions. In particular, arid and semi-arid region shows substantial uncertainty in the VI-FVC relations because of sparse vegetation cover and hence, important roles of local soil type in land-surface spectral reflectance. In this study, VI-FVC relations were investigated for arid and semi-arid regions of Mongolia. The FVC data was prepared from digital-camera image interpretation sheets taken at 160 sites in our field excursions from 2012 to 2014. In comparisons with visual inspections, the camera-based FVC showed good linear relations (r = 0.97, p < 0.001) with 6.83% of RMSE. Three satellite-based VIs were prepared, i.e. MODIS NDVI, EVI, and SAVI, which applied to produce linear regression models of FVC. The model parameters (i.e. slope and intercept) was obtained through iterated calculation process. Among the 160 sites, 120 sites were arbitrarily extracted to produce regression model and the remaining 40 sites were used for model validation. This iterated process were repeated 1,000,000 times and then, statistics were derived as averages of model (slope and interception) and validation (Pearson correlation coefficient and RMSE). In results, the regression models generally showed good agreements in model validations over r = 0.8 (p < 0.001). This study discussed problems in long-term FVC retrieval for the arid and semi-arid regions of Mongolia.