A temporal-BRDF model-based approach to change detection

Remote Sensing provides the only practical means to monitor changes over large areas. This paper describes the development of a generic algorithm designed to map the temporal occurrence and spatial extent of areas exhibiting sudden change. The algorithm is demonstrated here applied to the problem of mapping fire-affected areas. The research further develops the work of, which implemented a bi-directional reflectance (BRDF) model-based change detection algorithm to map the approximate day and location of burning, using daily 500 m MODIS surface reflectance data. An original algorithm assumption is that the surface state remains static prior to the changes of interest. This is problematic in the presence of underlying change (for example, due to vegetation phenology) especially when there are missing and/or cloudy data. In an attempt to deal with this issue, an additional kernel has been added to the BRDF model in the form of a cubic function of time. In addition, a step function kernel has been introduced in order to more robustly detect step-like changes. These modifications and preliminary results over southern Africa using daily MODIS land surface reflectance data are presented