Remote Sensing of Biomass Burning in Tropical Regions: Sampling Issues and Multisensor Approach

Abstract

Several global datasets on fire distribution are being generated from remotely sensed data to support research on the ecological impacts of biomass burning. This article examines the strengths and weaknesses of a number of approaches to the monitoring of biomass burning at a regional scale and suggests how to best combine the information content on fire distribution provided by different earth observation satellites. Remotely sensed data acquired over Central Africa from a variety of sensors (airborne video camera, SPOT XS, Landsat Thematic Mapper, NOAA Advanced Very High Resolution Radiometer, and ERS-1 Along Track Scanning Radiometer) were used to provide quantitative measurements of the spectral separability, and temporal and spatial sampling associated with the detection of burnt areas and active fires. Three main strategies to the monitoring of biomass burning were analyzed: detection of burnt areas at fine spatial resolution, detection of burnt areas at coarse spatial resolution and high temporal frequency, and detection of active fires at a coarse spatial resolution and high temporal frequency. In each case, we assess the detectability of the selected biomass burning indicator, the statistical representativity in time and space of the sample detected and whether the sample observations are an unbiased estimator of the total biomass burning events in the region. We conclude that while active fire detection remains important in defining the seasonality, timing, and interannual variations in biomass burning, the most reliable strategy for estimating biomass burning at a regional scale is a multisensor approach in which regional burnt area estimates from coarse spatial resolution data are calibrated on the basis of a sample of fine spatial resolution estimates of burnt areas, using a double sampling with regression estimator approach.