CHARACTERISING VEGETATED SURFACES USING MODIS MULTIANGULAR SATELLITE DATA

Abstract. Bidirectional Reflectance Distribution Functions (BRDF) seek to represent variations in surface reflectance resulting from changes in a satellite's view and solar illumination angles. BRDF representations have been widely used to assist in the characterisation of vegetation. However BRDF effects are often noisy, difficult to interpret and are the spatial integral of all the individual surface features present in a pixel.

This paper describes the results of an approach to understanding how BRDF effects can be used to characterise vegetation. The implementation of the Ross Thick Li Sparse BRDF model using MODIS is a stable, mature data product with a 10 year history and is a ready data source. Using this dataset, a geometric optical model is proposed that seeks to interpret the BRDF effects in terms of Normalised Difference Vegetation Index (NDVI) and a height-to-width ratio of the vegetation components. The height-to-width ratio derived from this model seeks to represent the dependence of NDVI to changes in view zenith angle as a single numeric value.

The model proposed within this paper has been applied to MODIS pixels in central Australia for areas in excess of 18,000 km². The study area is predominantly arid and sparsely vegetated which provides a level of temporal and spatial homogeneity. The selected study area also minimises the effects associated with mutual obscuration of vegetation which is not considered by the model.

The results are represented as a map and compared to NDVI derived from MODIS and NDVI derived from Landsat mosaics developed for Australia's National Carbon Accounting System (NCAS). The model reveals additional information not obvious in reflectance data. For example, the height-to-width ratio is able to reveal vegetation features in arid areas that do not have an accompanying significant increase in NDVI derived from MODIS, i.e. the height-to-width ratio reveals vegetation which is otherwise only apparent in NDVI derived from the higher resolution Landsat NCAS dataset.