The I-band fire detections, shown as blue flame icons, represent the nominal, at-nadir pixel footprint (375m). Because of the higher resolution of the I-band, the icons are intended to draw the user’s attention to the detection(s). Zooming in will result in the blue flame(s) being replaced with color-coded pixel(s) which represent the calculated brightness temperature (red=hottest, yellow=coolest). Clicking a colored pixel will display a pop-up with the potential range of temperature in Kelvin. The M-band fire icons can also aid in identifying I-band detections, but may result in some I-band detections being missed because cases often arise where the I-band detects a fire and the M-band does not, owing to it's coarser resolution.
The color-coded satellite icon shows the center of the satellite granule.
Click the satellite icons to view an I-band RGB quicklook. Right click to “save image as”. The I-band PNG shows the detections overlaid on top of a false-color RGB (I-band 3, 2, 1 respectively) along with lat/lon graticules. (I-bands and corresponding wavelength centroid – 3: 1.61 μm; 2: 0.861 μm, 1: 0.640 μm)
You can also download just the RGB as a GeoTIFF, as well as the M-band detections for the entire day (cover all of CONUS) as an ASCII file or KMZ.
Figure 1: VIIRS I‐band (color‐coded polygons) and M‐band (flame icon) detections for August 23, 2015. Note that several I‐band detections do not have corresponding M‐band detections, owing to the higher resolution capabilities of the Imaging channels (discussed more below).
Figure 2: VIIRS 375 m (I) and 750 m (M) channels used in support of active fire detection, complemented
by MODIS 1 km (B) channels used in the Fire and Thermal Anomalies product (MOD14 & MYD14).
The VIIRS imager has five channels spanning the visible, shortwave-, middle and thermal-infrared spectral regions (Figure 2). Compared to the VIIRS 750 m M13 channel and MODIS 1 km channels 21/22 driving the respective fire products, VIIRS I4 channel has a shorter wavelength and broader spectral interval. Consequently, there is an approximate three-fold increase in the solar component on channel I4 compared to the other two coarser resolution data sets. However, any potential impact on the performance of the active fire algorithm may be reduced with the use of pixel classification tests designed to identify highly reflective fire-free areas.
Figure 3: VIIRS imager data acquired at 04:22UTC on 04 January 2013 over Tasmania/Australia. Panel (a) shows channels 3-2-1 RGB composite showing large active near the center of the image subset, including a large smoke plume extending towards the southeast. Panel (b) shows the zoomed active fire extracted from (a). Panel (c) shows the middle-infrared channel I4 with nominally saturated pixels highlighted in green, and folded pixels showing an artificial brightness temperature of 208 K marked in yellow. Panel (d) shows the thermal infrared channel with nominally saturated pixels highlighted in green. Red outline on panels (c & d) coincides with other artificially low channel I4 brightness temperature values also attributed to saturation scenarios accompanied by nominal data quality on channel I5.
VIIRS Imager Active Fire Algorithm
The VIIRS 375 m active fire algorithm builds on the heritage MODIS Fire and Thermal Anomalies product (MOD14 & MYD14), using a contextual approach based on all five imager channels to detect active fires during daytime, and on the middle- and thermal-infrared channels to detect active fires burning at night. Visible and shortwave infrared channel daytime data are used to screen for clouds and water bodies, and to help separate active fires from other fire-free bright pixels characterized by high solar reflection. At night, the sources of confusion are by and large eliminated, therefore the night fire detection is primarily driven by channel I4 and I5 input data. However, channel I4 data noise associated with the South Atlantic Magnetic Anomaly (SAMA) can lead to false alarms across that region, which extends over 110o W<> 11o E and 7o N<> 55o S. In order to address that, VIIRS M13 channel data are used in order to verify fire detections coincident with the SAMA region.
The theoretical minimum detectable fire using VIIRS 375 m data was derived using simulated active fires ranging between 2 <> 250 m2 and 400 <> 1,200 K, applied to actual global imager data representing different geographic regions. Figure 1 shows the 50% probability of detection curve derived for both day and nighttime algorithms, as a function of fire area and temperature.
Figure 4: Theoretical minimum detectable fire using VIIRS 375 m day and nighttime data represented by the 50% probability of detection curve as a function of fire area and temperature.
An experimental fire was implemented on 08 July 2013 at a site located approximately 90 km northeast of Rio de Janeiro, Brazil (43o00’00”W 22o23’36.6”S). The burn consisted of a bonfire measuring 1.25 m in radius containing dry biomass (~6 feet tall) (Figure 2). The fire was ignited approximately 10 minutes prior to the VIIRS overpass at 1:23am local time. The temperature was estimated at approximately 1,000 K at the time of the VIIRS observation, resulting in a +10 K increase of the fire pixel brightness temperature in the middle infrared I4 channel compared to the fire-free background. Consequently, this small fire met the primary contextual tests included in the nighttime fire detection algorithm corroborating the theoretical detection envelope above.
Initial assessment of the VIIRS 375 m fire detection data indicates superior overall detection performance for both small and large fires compared to the operational 750 m fire data generated by the Interface Data Processing Segment (IDPS), resulting in three-fold increase in the absolute number of daytime fire pixels, and 25-fold increase in the absolute number of nighttime fire pixels detected. The difference is primarily the result of the finer spatial resolution of the 375 m data, which provides better spatial information and allows for the detection of smaller fires omitted by the 750 m product. The greater difference separating the number of nighttime fire pixels detected using the 375 m and 750 m data may due to diurnal variation in fire behavior (smaller and less intense fires burning at night), compelled by the lack of tuning of the current 750 m operational fire algorithm. The VIIRS 375 m fire data also showed improved performance compared to MODIS 1 km data, resulting in more coherent fire monitoring data for fires lasting multiple days (Figure 6).
Figure 5: Small experimental 1.25 m radius bonfire implemented on 08 July 2013 near Rio de Janeiro, Brazil. The sequence of photos illustrate the pre-fire set up, the peak flaming activity (approximately 7 minutes prior to VIIRS overpass), and the post-fire moment. The 04:23UTC VIIRS middle infrared (channel I4) image subset shows the bright active fire pixel contrasting with the cooler fire-free background.
Figure 6: Wildfire detected by 1 km Terra/MODIS (left), 375 m VIIRS (center), and 1 km Aqua/MODIS (right) over the Taim Ecological Reserve in Southern Brazil during 26-31 of March 2013 (Julian days 85-90).
VIIRS Imager Data Characteristics
The finer resolution VIIRS five-channel imager has native spatial resolution of 0.371 x 0.128 km at nadir and an aggregated resolution of 0.371 x 0.388 km at nadir. Figure 1 below illustrates the native pixel resolution with distance from nadir. The aggregation of native pixels is performed along scan resulting in three distinct image sections. In the first section extending from 0o (nadir) to ±31.59o scan angle three native pixels are aggregated over the radiance domain to form one effective sample. In the second image section extending from ±31.59o to ±44.68o scan angle two native pixels are aggregated to form one effective sample, and finally in the third image section extending from ±44.68o to the maximum scan angle (±56.06o) no aggregation is performed and one native pixel represents one effective sample (Figure 1). An array of 32 detections arranged along track generates 6,400´32 samples for each rotation of the telescope.
Similar to the 750 m M-band data, bowtie processing is performed onboard in order to reduce data redundancy away from nadir. Consequently, the four outermost sample rows of an individual scan (two on each end) are replaced with fill values across the second image section, whereas the eight outermost sample rows of an individual scan (four on each end) are replaced with fill values across the third image section. This procedure results in approximately 50% reduction in data redundancy in the Level 1B sensor data record (SDR). Figure 2 shows an example of a 5-min orbit segment displayed in satellite (swath) and geographic (resampled) projection. The onset shows the zoomed area taken from the swath image where the bowtie deletion pixels can be seen across the second and third image sections.
Figure 7: VIIRS imager native pixel and aggregated sample spatial resolution with distance from nadir (top). VIIRS imager sample size and area as a function of scan angle (bottom).
Figure 8: VIIRS imager 5-min orbit segment of channels 3-2-1 RGB combination displayed in satellite (swath; top left) and geographic (top right) projection. The bottom panel shows the zoom area representing the red box in the swath image, where fill values (striped pattern) associated with bowtie processing can be seen coinciding with second and third image aggregation sections.
375 m Active Fire Mapping Examples
The images below highlight the efficacy of using the I-bands and the inherent benefit of having finer spatial resolution fire information in support of daily active fire monitoring.
Power Fire – California 2013
This first example shows data acquired on 06 August 2013 for the Power Fire in California. The fire was imaged on 06 August by the U.S. Department of Agriculture National Infrared Operations (NIROPS) at 21:36h PDT, approximately 7 h after the afternoon VIIRS overpass on 06 August, and 5 h before the early morning VIIRS overpass on 07 August. Figure 1 shows the spatial subset corresponding to the fire, with VIIRS 375 m and 750 m fire pixel footprints and NIROS fire perimeter and intense heat areas highlighted. Overall, VIIRS 375 m day and nighttime fire detection data showed excellent agreement with the ~10 m resolution NIROPS airborne data. The operational VIIRS 750 m fire product also showed good overall agreement with the NIROPS data, although nighttime detection was limited to the northeast section of the fire, omitting the cooler southwest area highlighted as intense heat in the NIROPS data.
Figure 9: VIIRS 375 m and 750 m fire detection data acquired at 14:28h PDT on 06 August 2013 and at 2:50h PDT on 07 August 2013 over the Power Fire in California. USDA-Forest Service NIROPS airborne data acquired at 21:36h PDT on 06 August 2013 shows the mapped fire perimeter as well as the areas showing intense heat.
Rim Fire – California 2013
The second example selected shows data acquired over the Rim Fire in California, which burned for several days during August-September 2013. The animation below contains the sub-daily VIIRS 375 m fire detections accumulated over the period of 19 August – 12 September 2013. Active fire pixels are color-coded according to the pixel-level brightness temperature, here used as a proxy for fire intensity.
Multiple USDA-NIROPS airborne data were acquired for the Rim Fire. Figure 2 shows two instances of the fire development when NIROPS ~10 m and VIIRS 375 m active fire were generated within less than 90 minutes apart. NIROPS data show the accumulated fire perimeter up to the date of acquisition, along with areas of intense heat (striped pattern). The daily progression of the fire perimeter mapped using VIIRS 375 m shows excellent agreement with the maps derived from NIROPS data.
Figure 10: Instantaneous observations of the Rim Fire in California using near-coincident NIROPS ~10 m and VIIRS 375 m data. Fire perimeter outlined by the NIROPS data represents the accumulated fire-affected area at the time of sampling, whereas the striped pattern describes the instantaneous intense heat areas identified by NIROPS image analysts.