Burned area mapping using multi-temporal moderate spatial resolution data - a bi-directional reflectance model-based expectation approach

While remote sensing offers the capability for monitoring land surface changes, extracting the change information from satellite data requires effective and automated change detection techniques. The majority of change detection techniques rely on empirically derived thresholds to differentiate changes from background variations, which are often considered noise. Over large areas, reliable threshold definition is problematic due to variations in both the surface state and those imposed by the sensing system. A new approach to change detection, applicable to high-temporal frequency satellite data, that maps the location and approximate day of change occurrence is described. The algorithm may be applied to a range of change detection applications by using appropriate wavelengths. The approach is applied here to the problem of mapping burned areas using moderate spatial resolution satellite data. A bi-directional reflectance model is inverted against multi-temporal land surface reflectance observations, providing an expectation and uncertainty of subsequent observations through time. The algorithm deals with angular variations observed in multi-temporal satellite data and enables the use of a statistical measure to detect change from a previously observed state. The algorithm is applied independently to geolocated pixels over a long time series of reflectance observations. Large discrepancies between predicted and measured values are attributed to change. A temporal consistency threshold is used to differentiate between temporary changes considered as noise and persistent changes of interest. The algorithm is adaptive to the number, viewing and illumination geometry of the observations, and to the amount of noise in the data. The approach is demonstrated using 56 days of Moderate Resolution Imaging Spectroradiometer (MODIS) land surface reflectance data generated for Southern Africa during the 2000 burning season. Qualitatively, the results show high correspondence with contemporaneous MODIS active fire detection results and reveal a coherent spatio-temporal progression of burning. Validation of these results is in progress and recommendations for future research, pending the availability of independent validation data sets, are made. This approach is now being considered for the MODIS burned area algorithm. (C) 2002 Elsevier Science Inc. All rights reserved.