Information of vegetation properties, including vegetation cover, leaf area index (LAI), vegetation water content (VWC) and vegetation dry biomass (VDM), has been widely used in ecology, geology, hydrology, meteorology, agriculture, forestry and even broader applied science and social science areas. The methodologies to evaluate the relationships between these vegetation properties using remote sensing were investigated in this study. The study area is Walnut Creek Watershed which is the focus of the Soil Moisture Experiments 2002.

Ground measurements proved strong linearity in coherent relations between leaf and stem components, and between VWC, VDB and LAI. Utilizations of spectral bands from different operational satellite sensors, namely Landsat 5 TM / Landsat 7 ETM+ and Terra - MODIS, were tested. Combining the strengths of atmospheric model and empirical model, a hybrid atmospheric correction model was applied to Landsat images and achieved comparable results to the conventional 6S method. Incorporation of TM and ETM+ was proposed to overcome the insufficient temporal coverage availability of Landsat imagery. Certain processing procedures had been followed to improve the quality of MODIS data. Both the Landsat- and MODIS-derived normalized difference water indices (NDWI; NDWI1240, NDWI1640 or NDWI2130 using the 1240 nm, 1640 nm or 2130 nm) were proposed and compared to the conventional Normalized Difference Vegetation Index (NDVI) in vegetation properties modelling thereafter.

Whereas NDVI saturated when LAI exceeded 3.0 (or VWC exceeded 3.5 kg/m2, or VDB exceeded 0.55 kg/m2, or vegetation cover is higher than 84%), it was evidenced that NDWI have sensitivity to LAI growth up to 5.0 (or VWC up to 4.5 kg/m2, or VDB up to 0.70 kg/m2, or vegetation cover up to 95%). Vegetation cover was mapped best by NDWI2130 via linear models. A linear reflectance mixture model was employed to remove subpixel soil reflectance contamination. Soil effect was found strongly influencing vegetation indices and consequently vegetation property estimations.

The soil effect corrected and uncorrected NDVI and NDWI were tested for their performances in evaluating LAI, VWC and VDB. It was discovered that LAI shows an expolinear relationships with vegetation indices. In LAI evaluation, the NDWI1640 achieved the best performance among the candidates. The correlation analysis between LAI and vegetation cover concluded a convincing expolinear relationship demonstrating a ‘two stages LAI growth pattern’: the horizontal expansion of leaves is predominant up to a certain level of vegetation cover (~62.5%) as an early stage, and the vertical leaf layer clumping plays a more dominant role at a later stage. In the relationships between LAI, vegetation cover and vegetation indices, some traces of vegetation-species independency was noted for corn and soybeans in this study. In VWC and VDB estimation, NDWI which utilize the water absorption featured NIR (1240 nm) and SWIR (1640 nm, 2130 nm) bands perform better than NDVI. With soil effect contamination, the NDWI1640 and NDWI2130 bands were preferable in VWC and VDB modelling. However, if soil effect can be corrected to some extents, the NDWI1240 and NDWI1640 should be considered as best candidates in linearly describing the relationships with VWC and VDB.
(Contact Email: jhuang@rsmas.miami.edu)