Analysis of POLDER-ADEOS data for the Australian continent: the relationship between BRDF and vegetation structure

We analysed the POLDER-ADEOS level 3 dataset (coefficients of a bi-directional reflectance distribution (BRDF) model fitted to reflectances) to interpret the spatial and temporal patterns of BRDF for the Australian continent in terms of landscape attributes. Significant noise was identified and systematically removed by transparent and mechanistically sound filtering and interpolation. We investigated the hypotheses that the spatial variation of the Roujean model coefficients over the continent is greater than the temporal variation over the eight month time series and that vegetation structure has more influence on the BRDF than does landform. We found systematic, interpretable relationships between the Roujean model coefficients and structural attributes for both woody and grassy vegetation. For woody vegetation, the model coefficients were closely related to the spacing to height ratio of vegetation elements. The Roujean model coefficients were generalized to ten types by an unsupervised classification which produced coherent spatial patterns. While these could be related to vegetation types, they indicated that there were other factors operating. Finally, the magnitude of the BRDF normalization of AVHRR to nadir viewing and fixed solar zenith angle was investigated for the continent under summer and winter illumination conditions. The magnitude of normalization was small during summer, but was as large as 100% at the winter solstice.