Close Window

A basin-wide assessment of the GOES and MODIS active fire products for the Brazilian Amazon

Wilfrid Schroeder, University of Maryland, (Presenting)
Ivan Csiszar, University of Maryland,
Elaine Prins, University of Wisconsin,
Chris Schmidt, University of Wisconsin,
Alberto Setzer, INPE,
Karla Longo, CPTEC/INPE,
Saulo Freitas, CPTEC/INPE,
Jeffrey Morisette, NASA,
Jason Brunner, University of Wisconsin,

This LBE-ECO Phase III study is designed to assess the performance of active fire products which have been used to delineate the fire dynamics in the Brazilian Amazon basin and which are routinely used to feed biomass burning emissions models for the region. The initial analyses are focused primarily on the creation of a validated long term (1995-present) record for the WFABBA active fire product using GOES East geostationary satellite data. For comparison purposes we also included the MODIS/Terra “Thermal Anomalies” (MOD14) data in our validation analyses. We found that at the 50% detection probability mark (p<0.001), the GOES fire product requires four times more active fire area than it is necessary for MODIS to achieve the same probability of detection. However, the higher observation frequency of GOES resulted in less than 40% omission error compared to 80% with MODIS. Basin-wide commission errors for MODIS and GOES were approximately 15 and 17%, respectively. Commission errors were higher over areas of active deforestation due to the high thermal contrast between the deforested sites and the adjacent green forests which can cause multiple false detections. Burnt area estimates were also produced based on ETM+ data to assess the average burnt area size associated with the coarse resolution active fire data above. Burn scar polygons were digitized and intersected with the MODIS/Terra and Aqua active fire data. 50% of all polygons containing active fires in the MODIS imagery showed a burnt area size larger than 300ha. Burnt areas of less than 100ha in size represented 15% of all cases analyzed. Further work will be pursued to create a unified fire diurnal cycle map that can be used to model time dependent variables (e.g.: emissions). Emission modeling studies will also be addressed at the subsequent phases of this research project by feeding models with the optimized data sets described above.

Science Theme:  LC (Land Use and Land Cover Change)

Session:  3C: Land Use and Fire

Presentation Type:  Oral

Abstract ID: 36

Close Window