Testing the sensitivity of a MODIS-like daytime active fire detection model in Alaska using NOAA/AVHRR infrared data

C. A. Seielstad, J. P. Riddering, S. R. Brown, L. P. Queen, W. M. Hao

    Research output: Contribution to journalArticlepeer-review

    Abstract

    A MODIS-like daytime active fire detection model was tested in Alaskan biomes using NOAA-AVHBR infrared data, and its performance was assessed across a range of channel 3 (3.8 μm) brightness temperature and contextual standard deviation thresholds. Absolute thresholding of channel 3 (T3) and the channel 3/4 difference (T34) was more effective than contextual analysis in minimizing false detections, although detection sensitivity to actual fire pixels was lower. The contextual analysis became more effective in terms of fire detections as the T3 and standard deviation thresholds were loosened. However, enhanced fire detection capabilities were achieved at the expense of increased false detections associated primarily with cloud edges. False detections increased exponentially and detections of active fires increased linearly as thresholds were loosened. Furthermore, T3 and standard deviation thresholds suggested for the MODIS global fire detection product appear too high for Alaska. An optimal T3 threshold between 314K and 315K and a standard deviation threshold between 2.5 and 3.5 are proposed. These results suggest that each biome or region may require different thresholds to optimize algorithm performance, recognizing that optimization of the model depends upon user goals. Effective cloud removal is clearly the most significant issue facing this type of fire detection method.

    Original languageEnglish
    Pages (from-to)831-838
    Number of pages8
    JournalPhotogrammetric Engineering and Remote Sensing
    Volume68
    Issue number8
    StatePublished - 2002

    Fingerprint

    Dive into the research topics of 'Testing the sensitivity of a MODIS-like daytime active fire detection model in Alaska using NOAA/AVHRR infrared data'. Together they form a unique fingerprint.

    Cite this