NRL Monterey, Marine Meteorology Division
|Fires or Hot-Spots in this Product Appear as Red, With Strongest Signals Appearing Yellow|
|During the Gulf War, Iraqi troops set fire to several oil wells during their retreat from Kuwait. The thick black smoke generated by these fires cloaked the sky, posing particular challenges to U.S. and Coaltion Force operations. Using high resolution infrared information collected by MODIS and AVHRR sensors, we are able to reveal the location of many fires and hot-spots from space, including the natural gas waste flares atop most of the oil wells throughout the Middle East. The current product highlights the location of active hotspots.|
|The algorithm used to detect fires or hot-spots (where no flame exists; hereafter, both varieties will be referred to as hot spots) is based on the high sensitivity to heat in the 3.9 micrometer channel available on various satellite radiometers (e.g., MODIS, AVHRR, GOES). This high sensitivity means that a hot spot covering only a small portion of a satellite pixel often will be detected. The spatial variability of this signal is used to flag and color-code these hot spots, which are then overlaid upon satellite imagery (visible during the daytime, infrared at night). Owing to solar reflection contributions during the day, the thresholds used to avoid false alarms are far more conservative. This results in fewer hot spots detectable during the daytime passes compared to night. The user therefore should use this product as a tool in determining where fires exist, as opposed to where they do not exist.|
|Fire Product Matches With Smoke Observed In True Color Imagery|
|The fire product enables the analyst to identify the location of active hot spots in the imagery and thereby monitor activities remotely. The product is available for both MODIS (day/night) and AVHRR (night only), providing multiple looks per day over most regions. The current product is somewhat more aggressive than the level-2 MODIS Science Team algorithm, in that signals over brighter backgrounds and under optically thin clouds are included. The penalty for this aggression is occasional (but infrequent, and usually easy-to-identify) false alarms as illustrated in examples below.|
|Image Near Swath Center||Image Near Swath Edge|
|As discussed above, detection of hot spots will vary between day and night
and also according to position of the swath over the current region of
interest. In the example above, collected from imagery only four hours
apart, the detrimental swath-edge effects are readily apparent. The
example at left, created from a near overhead satellite pass, shows
numerous hot spots throughout souther Iraq, Kuwait, parts of western Iran,
and various positions in the Norther Arabian Gulf (off-shore oil
platforms). The image at right depicts the same scene but from data near
the edge of the passing satellite swath. In addition to fewer fires
detected, those that are detected appear blurred and distorted, owing to
larger pixels and an increased chance of intra-channel spatial mismatching.
Some hot-spots appear as rings, owing to the failure of the spatial
variance filter in these situations (due to the smearing of the hot spot
information over a larger physical area).|
While the product is able to detect fires through optically thin clouds (e.g., refer to the topmost image of this tutorial, where several hot spots appear beneath a veil of cirrus), optically thick clouds will mask the heat signature--leaving them undetected.
The tuning of this product is currently on the aggressive end, which means that occasionally a cloud edge may light up as a hot spot. Be wary of any such features that appear linear and along a cloud edge.
The AVHRR product is available only during the nighttime hours, when it uses 3.9 micron (Ch3). During the day it uses 1.6 micron (Ch3a), which provides insufficient sensitivity to hotspots to avoid numerous false alarms.
The examples presented below will illustrate additional caveats associated (albeit infrequently) with this product, encountered predominantly during daytime images. In most instances, familiarity with these false-alarm sources will enable the analyst to rule them out.
|Daytime Caveat: Sunglint Zone|
|The region of specular reflection by the solar disk upon water surface is called the sunglint zone. Strong contributions from 3.9 microns combined with sometimes sharp gradients in the specular signature and a stable 11.0 micron thermal background contribute to false alarms under these special conditions. The example above illustrates a line of false fire alarms corresponding to a boundary of particularly strong glint brightness gradient. Comparison with the true color and high/low cloud companion imagery is recommended in these situations.|
|Daytime Caveat: Bright Shorelines|
|Sometimes the sunglint can transform muddy flat shorelines into lines of false fire pixels. An example is shown above. The brightness variation across the water body is a tell-tale sign that sunglint is at play in this image. Such unrealistic features, while uncommon, can usually be identified as false alarms without considerable effort. Examination of true color imagery can also be useful in ruling these regions out on the basis of no observable smoke.|
|Day/Night Caveat: Deep, Cold Convection||IR Imagery Indicates a Very Cold Cloud Top|
|While 3.9 micron is very sensitive to hot temperatures, it is all the more insensitive to cold temperatures. Deep convection produces extremely cold temperatures approaching or falling beneath the noise limits of the 3.9 micron channel. 11.0 micron, on the other hand, has good sensitivity to cold temperatures and tracks the deep convection temperatures very well. The result of a difference between the noisey 3.9 measurement and 11.0 can give rise to spikes in the fire detection algorithm which then manifest as false fires. An example of this artifact is shown above for a deep convective cloud, together with real fires that appear through the surrounding anvil cirrus. The accompanying 11.0 micron image of this scene shows an extremely cold (temperatures near -80C) cloud top, more than enough to introduce the noise artifacts described above.|
|Model Wind Overlays|
|Inferring the trajectory of smoke plumes emanating from hot spots in a snap-shot image requires wind speed and direction information (and an assumption that changes in this wind field over the time in question are minimal). An overlay of COAMPS or NOGAPS surface wind vectors, color-coded for speed, is shown above. The southwesterly winds analyzed over southern Iraq agree with the dark plumes streaming from the line of natural gas waste plumes hot spots. Were this windfield to remain constant over a sufficient duration and the plumes remained at a fixed level, one might infer that they would begin to drift more southerly over Kuwait. Slight disagreement can be due either to forecast error (increasingly likely with higher values of model Tau) or to the plumes being at a altitude whose winds are not represented by the surface wind field. Because hot spots are sometimes more difficult to pick out when color vectors are included in the image, a hot-spot-only product is produced along with the wind-overlay.|
Author: Steve Miller
Last Updated: Wed Mar 19 09:53:56 2003
Produced by: The Composer (Ver: 1.4 )