Validation of Remotely Sensed Fire Detections Using Ground and Aircraft Reports

A daily fire analysis for North America is prepared by NOAA/NESDIS utilizing seven NOAA and NASA geostationary and polar orbiting satellites. The analysis incorporates automated fire detections into an analyst quality control procedure. Limited validation on the analysis has been performed to date. One effort utilized high resolution ASTER sensor data on the NASA Terra spacecraft and another used ground reports from Montana and Idaho. Owing to inherent limitations in both approaches further validation has been performed. The current study expands on the ground report method. Daily ground reports (authorizations) have been obtained from the Florida Division of Forestry. Additional data from Montana/Idaho has been obtained as well as a small set of ground reports from Washington state. Aircraft data was also obtained for agricultural burns in Manitoba. The use of these additional data sets has expanded the validation to include a greater variety of land types/uses as well as geographic locations that have varying geostationary sensor viewing angles. The results are consistent with the previous ground based validation that only included Montana/Idaho and indicate a probability of detection (POD) of 20-25%. This low POD is at least partly due to cloud cover obscuring reported fires during active burning and also due to the small size of many fires. The analyst quality controlled product for each data set had a higher POD than for the automated detections only. The nature of the data sets precludes the determination of commission errors. POD information combined with fire size estimates are important considerations for emission modeling. These results suggest that emission estimates based on remotely sensed fire detections may be too low.