Detection of Fires At Night Using DMSP-OLS Data

Christopher D. Elvidge
NOAA National Geophysical Data Center
325 Broadway, Boulder, CO 80305

Ingrid Nelson, Vinita Ruth Hobson, Jeff Safran
Cooperative Institute for Research in Environmental Sciences
University of Colorado
Boulder, Colorado 80305 USA

Kimberly E. Baugh
Analytical Imaging and Geophysics LLC
Boulder, Colorado 80303 USA

GOFC Wildfire Book Chapter

January 13, 2000
Revised September 18, 2000

ABSTRACT

It has been known since the early 1970's that fires can be detected at night using low light imaging data from the Defense Meteorological Satellite Program (DMSP) Operational Linescan System (OLS). DMSP data have not been widely used for fire detection due to lack of near real time access to the digital data and the lack of algorithms for processing the image data. We describe a set of algorithms for the generation of georeferenced fire products from nighttime OLS data and provide a series of example products. The U.S. Air Force has reduced the hold on OLS data distribution from 72 hours to three hours. Nighttime OLS data are now being distributed to fire monitoring projects in near real time.

INTRODUCTION

The detection of fires at night with low light imaging data was first noted by Croft (1973) in data from the U.S. Air Force Defense Meteorological Satellite Program (DMSP). This appears to be one of the earliest reports indicating that fires could be monitored in near real time using satellite remote sensing. The realization of this capability using DMSP Operational Linescan System (OLS) low light imaging data has taken nearly thirty years as data access and processing issues were resolved. In this chapter we review the DMSP-OLS data characteristics, outline of processing steps used in fire detection, fire detection results obtained with DMSP-OLS data, and recently developed capabilities for delivery of the data using high speed internet links.

The DMSP-OLS is a two band (visible and thermal) imaging system designed for global observation of cloud cover. At night the visible band is intensified with a photomultiplier tube to permit detection of clouds illuminated by moonlight. The light intensification enables the observation of faint sources of visible- near infrared emissions present at night on the Earth's surface (Figure 1), including cities, towns, villages, gas flares, heavily lit fishing boats and fires (Croft 1973, 1978, 1979). By analyzing a time series of DMSP-OLS images, it is possible to define a reference set of "stable" lights, which are present in the same location on a consistent basis. Fires are identified as lights detected on the land surface outside the reference set of stable lights.

The earliest DMSP low light imaging system was the SAP (Sensor Aerospace Vehicle Electronics Package) which were flown from 1970-76. This was followed by the OLS sensors which began flying in 1976 and are expected to continue flying until approximately 2010. Digital DMSP data were not distributed outside of DoD or preserved during the first 22 years of the program. An OLS film archive at the University of Colorado, National Snow and Ice Data Center provided the scientific community with access to a subset of OLS data in analog form. The only systematic use of film data for the analysis of fires was the inventory of African fires reported by Cahoon et al. (1992).

A digital archive for the DMSP-OLS data was established in 1992 at the U.S. National Oceanic and Atmospheric Administration (NOAA) National Geophysical Data Center (NGDC). Publications describing the use of digital DMSP data for fire detection include Kihn (1996), Elvidge et al. (1996, 1998a, 1998b, 2000) and Fuller and Fulk (2000).

THE OPERATIONAL LINESCAN SYSTEM

The Air Force generally operates two DMSP satellites in sun-synchronous orbits, one in a dawn-dusk orbit, the other in a day-night orbit. The Operational Linescan System (OLS) is an oscillating scan radiometer with two spectral bands (visible and TIR) and a swath of ~3000 km. The "visible" bandpass straddles the visible and near-infrared portion of the spectrum (0.5 to 0.9 um). The thermal band pass covers the 10.5 to 12.5 um region. Satellite attitude is stabilized using four gyroscopes (three axis stabilization), a star mapper, Earth limb sensor and a solar detector. The OLS visible band signal is intensified at night using a photomultiplier tube (PMT), for the detection of moonlit clouds (Figure 1). The low light sensing capabilities of the OLS at night permit the measurement of radiances down to 10^-9 watts/cm ²/sr/um. This is four orders of magnitude lower than the OLS daytime visible band or the visible-near infrared bands of other sensors, such as the NOAA AVHRR or the Landsat Thematic Mapper. Fires present at the Earth's surface at the time of the nighttime overpass of the DMSP are readily detected in the visible band data (Figure 2a). In contrast, fires rarely show up as hot spots in the OLS thermal band data (Figure 2b). The OLS thermal band position (10.5 to 12.5 um) is not well placed for fire detection.

There are two spatial resolution modes in which OLS data can be acquired. The full resolution data, having nominal spatial resolution of 0.56 km, is referred to as "fine". On board averaging of five by five blocks of fine data produces "smoothed" data with a nominal spatial resolution of 2.7 km. Most of the data received by NOAA-NGDC is in the smoothed spatial resolution mode. Acquisitions of the fine resolution data can be made (on a non-interference basis) through request to the Air Force. There are spatial details present in the fine resolution data that are absent in the smoothed data. Figure 3 shows examples of smoothed versus fine nighttime visible band data of fires in S.E. Asia from 1999.

The PMT gain is programmed to track changes in lunar illumination, with the objective of producing uniform (constant contrast) imagery of clouds for as much of each lunar cycle as possible. The lowest gain setting coincide with the full moon, at which time clouds and the outline of land surface features such as coastlines and major drainage patterns can be readily observed in visible band data (see Figure 1). The highest visible band gain settings occur during the darkest 10 nights of each lunar cycle, when lunar illumination at the time of the DMSP overpass is less than half of the monthly maximum. Even with the high gain settings, clouds and most of the earth's surface remain dark in OLS data acquired during the darkest nights of the lunar cycle, leaving visible-near infrared emission sources as the primary features observable in the nighttime visible band data. As a result, fainter areas of visible-near infrared emission are detectable in OLS data acquired under new moon conditions than under full moon conditions.

DMSP-OLS FIRE DETECTION

Below is a brief description of the algorithms used to generate DMSP-OLS fire products. The steps are outlined in the flow chart shown in Figure 4. The procedures rely on the existence of a stable lights data set derived from a time series of nighttime OLS observations. The basic procedures used to generate the stable lights have been described by Elvidge et al. (1997). The fire product processing can be grouped into two divisions: processing done on raw OLS data and processing done on georeferenced OLS data. Certain steps are logically applied while the data are in their raw scanline format. In addition, the raw data has less data volume, which allows the processing to run faster. Processing steps applied to the raw data include: orbit assembly and suborbiting, cloud identification, light detection, geolocation and gridding. Following the light detection there are options for the removal of single pixel light detections and light detections that coincide with clouds on heavily moonlit nights. Pixels identities (e.g. lights, clouds, clouds with lights, missing scan lines) are marked in a flag file which overlays the OLS image data. Steps performed on the georeferenced data include the removal of lights associated with stable lights, lights over water surfaces, and final editing by an image analyst.

Orbit Assembly and Suborbiting:

The incoming OLS data from the Air Force is assembled into orbits. The on-board ephemeris for each scanline is stripped off and updated using highly accurate ephemeris derived from an Air Force orbital model optimized for the DMSP platforms. The model is parameterized with bevel vectors derived from radar sightings of the satellite (provided by Naval Space Command). The satellite heading is estimated by computing the tangent to the orbital subtrack. Suborbits are cut out of the orbital strips based on rectangular latitude / longitude outlines for areas of interest.

Cloud Identification:

Because of the low level of lunar illumination present in the nighttime data, it was not possible to use the visible band to assist in the identification of clouds. The cloud screening was based entirely on thresholds set on the TIR band. Clouds are generally colder than the earth's surface. However, the separation of cloud pixels from earth surface pixels using TIR thresholding is complicated by seasonal, latitudinal, and altitudinal variations in the background earth surface temperature. The separation of clouds from earth surface pixels is relatively easy at low latitudes where there is generally a large temperature difference between pixels of cloud tops and pixels containing land or ocean. Because of the strong latitudinal effects on the TIR threshold for cloud screening in our data, we segmented each of the orbital sections into a series of latitudinal bands for determination of a TIR threshold for discrimination and tagging of the cloud pixels.

Glare and Sunlit Data Removal:

Nighttime visible band data acquired at high latitudes near the summer solstice may be contaminated by sunlight. The OLS data stream includes solar elevations for the earth's surface at the nadir of each scanline. Scanlines with solar elevations exceeding -14 degrees are flagged as sunlit and are not processed for fire detection. One adverse effect of the light intensification is that the OLS is quite sensitive to scattered sunlight. Under certain geometric conditions, portions of the OLS are illuminated by sunlight. Scattering of sunlight into the optical path results in visible band detector saturation (Figure 1), a condition referred to as glare. The exact shape and orbital position of the glare changes through the year. To remove glare from OLS images, an algorithm searches the OLS data for occurrence of 100 by 100 blocks of pixels with t saturated pixels (DN=63). Once such a block of data is encountered, all adjacent pixels having DN values greater than 40 are flagged as glare.

Identification of Visible-Near Infrared (VNIR) Emission Sources:

Because of brightness variations that occur within and between orbits, it is not possible to set a single digital number (DN) threshold for identifying visible-near infrared emission sources. We have developed an algorithm for automatic detection of visible-near infrared emission sources (lights) in nighttime OLS data using thresholds established based on the local background (Elvidge et al., 1997a). Lights are identified in 20 by 20 pixel blocks, with the local background being drawn from the surrounding 100 by 100 pixel block. This "light picking" algorithm selects a dark background pixel set, which is analyzed to establish a threshold for the detection of pixels containing light sources. The resulting threshold is applied to the central 20 by 20 pixel block inside the 100 by 100 pixel block. Processing of an image proceeds by tiling the results from adjacent 20 by 20 pixel blocks. Only pixels that were not previously tagged as glare or sunlit can be tagged as VNIR emission sources.

Removal of Single Pixel Lights:

There is an option to remove single pixel lights which can be activated or turned off by the user. This feature is designed to remove noise which may be present under high gain settings or when the satellite passes through fields of high energy particles in the ionosphere. The area of the world where we found this option most useful is the Brazil region. There is a persistent pattern of noise in the low light imaging data of this region associated with South Atlantic Ionospheric Anomaly.

Geolocation and Gridding:

The DMSP fire processing uses 30 arc second grids to match the format of the stable lights data. This corresponds to grid cell size of about a square kilometer at the equator. The geolocation algorithm operates in the forward mode, projecting the center point of pixels onto the Earth's surface. The geolocation algorithm estimates the latitude and longitude of pixel centers based on the geodetic subtrack of the satellite orbit, satellite altitude, OLS scan angle equations, an Earth sea level model, and digital terrain data. The algorithm uses an oblate ellipsoid model of sea level and a terrain correction based digital elevations from the Global Land One-km Base Elevation (GLOBE) Project (http://www.ngdc.noaa.gov/seg/topo/globe.shtml). The DN values of the visible and thermal band data are used to fill 30 arc second grids using near neighbor resampling.

Setting Cloud Screen:

Another option which may be exercised is the elimination of light detections coinciding with clouds. This option is set based on lunar illumination levels, which are provided for each OLS scanline. The primary reason for this is that it is possible for small heavily moonlit clouds to be misidentified at lights by the light detection algorithm. Typically this only occurs when lunar illumination is over 70% of the monthly maximum. This phenomenon does not occur when lunar illumination is less than 50%. In general practice we set the lunar illumination threshold for elimination of cloud-covered lights at 50%. The lunar illumination threshold can be altered to meet the users requirements. The cloud screen is performed on the georeferenced data, as opposed to the scanline data, to accommodate plans to automate the cloud identification in the georeferenced thermal band data through comparison with modeled surface temperature grids.

Elimination of Stable Lights:

The georeferenced lights from single suborbits are overlain with the stable lights data sets. An algorithm then removes all lights from the georeferenced OLS data that are in contact with a stable light source. This approach has been developed to permit the algorithm to operate in cases where a simple mask removal based on the stable lights would not provide for the complete removal of fixed light sources. This procedure accounts for slight misregistrations between the stable lights and incoming lights. Atmospheric conditions such as smoke and light cloud cover makes lights appear larger than they do under clear sky conditions. Lights that are found outside and unattached to the set of stable lights are candidates for being identified as fires.

Removal of Lights Over Water:

Lights occurring over water are removed based on a 30 arc second land sea mask.

Editing:

As a final step the remaining lights are overlain with the stable lights data and thermal band image and visually reviewed to identify and remove light sources that are not from fires. This may include bad scan lines, lightning, city lights not fully removed, or heavily lit clouds. The editing step is expected to be simplified in the near future with the implementation of automatic algorithms for the detection and removal of lighting and bad scan lines.

FIRE DETECTION RESULTS

DMSP fire products can be generated for single suborbits, mosaics from multiple suborbits to cover extended geographic regions, or as temporal composites. As an example from a single suborbit, Figure 5a shows a georeferenced visible band image and Figure 5b the resulting fire product from the September 7, 1999 nighttime OLS orbit over Bolivia, Paraguay, and parts of Brazil during the peak of the fire season. Large numbers of fires were observed in Paraguay and Bolivia on this date.

By compositing nightly fire detections over time for individual years or fire seasons and normalizing the results based on the total number of observations, it is possible to compare the relative severity and pattern of fires over time. Figure 6 shows the cumulative burn observed during the August through December time period for Madagascar for each year from 1992 through 1999. Beginning in 1994 the government of Madagascar has issued annual no-burn edicts. The OLS observations suggest this policy may have been partially successful in 1994.

NGDC has produced a global fire product for a six-month period, from October 1, 1994 through March 31, 1995. The product tallies the frequency with which fires were detected in cloud-free OLS data from the dark half of lunar cycles during the six month time period. Several dense regional clusters of fires were observed, including the Sahel region of Africa, south of the Sahara Desert (Figure 7a), Southeast Asia (Figure 7b), Australia (Figure 7c), and northern South America (Figure 7d). Fires were observed in the global product to latitudes of 60 degrees in the northern hemisphere, indicating that DMSP's fire detection capabilities are not limited to tropical and temperate regions. The dearth of fire detection in places such a Brazil and southern hemisphere Africa can be attributed to the fact that the analysis window (October through March) missed the primary burning season in these areas.

FACTORS WHICH INFLUENCE DMSP FIRE DETECTION

There are a number of factors or conditions which will impede the detection of fires with DMSP-OLS data, including: high levels of lunar illumination, cloud cover, and solar glare. OLS fire detection is susceptible to errors of commission due to the fact that the fire detection occurs in a single spectral band which also detects large number of other light sources and moonlit clouds. If the stable lights set used is incomplete then lights from cities or towns can pass into the fire product. Thus it is important that the stable lights database be kept up to date and that it is constructed with sufficient numbers of cloud-free observations to ensure that all stable lights are included. Because of the manner in which the stable lights are removed, fires that are adjacent to stable lights can not be readily detected (errors of omission).

Because final steps in generating a DMSP fire product are performed on 30 arc second grids instead of OLS pixels, the are ambiguities in the tally of OLS fire pixels. Since a single OLS pixel usually occupies nine 30 arc second grid cells, a simple estimate of the number of OLS fire pixels can be made by dividing the total number of 30 arc second cells tagged as fire by nine.

It is clear that the OLS detects fires which are subpixel in size. This is also the case for AVHRR, GOES, MODIS and other systems. Methods for making subpixel estimates of fire size in GOES and AVHRR have been developed based on the brightness of fires in multiple thermal bands. Kihn (1996) developed a method for estimating the area of active fire using nighttime OLS data. However, the application of the Kihn technique relies on the detection of the fire in both the visible and thermal band data, a rare occurrence with OLS. As with AVHRR, OLS visible band data often saturates when a fire is present. If a fire is undetected in the thermal and saturated in the visible, no direct estimate of the fire size can be made. In cases where the OLS visible band data for a fire pixel is unsaturated, the probability of detection in the thermal band is low. As a result, the visible band DN values are probably best viewed as indicators but not predictors of fire size or fire intensity.

During the full moon, the OLS visible band gain is set to a low level. This raises the minimal detectable radiance relative to the dark portion of the lunar cycle and fewer fire detections (Figure 9). The detection of fires with OLS data works best during the dark half of each lunar cycle. Fewer fires will be detected during the brightest nights of lunar illumination cycle.

Heavy cloud cover blocks the transmission of visible-near infrared light from the earth's surface, blocking the observation of fires. Many fires are detected under light cloud and smoke cover, however their outlines appear diffused. Other factors impeding the detection of fires include data dropouts, sunlight, and solar glare. Data dropouts occur randomly. Thus there is always some chance that DMSP-OLS data of a particular event will not transmitted to NGDC. Solar glare results in unusable nighttime visible band data. This phenomenon precesses and thus will affect portions of an orbit and gradually shift to the north or south. In some cases it is possible to use data from the edge of scan in an adjacent orbit to observe visible-near infrared emissions that are blocked by solar glare in the orbit. Sunlight will impact the detection of fires at high latitudes at or near the summer solstice.

NGDC recently tested the geolocation accuracies achieved with DMSP-OLS data from satellites F-10, F-12, F-14, and F-15. This testing was done using a time series of images from each satellite to track geolocation shifts for a subpixel light source in the Mojave Desert of California. Our results indicate that the geolocation accuracy for F-10, F-12 and F-15 OLS data are less than an OLS pixel. NGDC identified geolocation errors for F-14 exceeding an OLS pixel. The F-14 geolocation can be improved to +/- one pixel by increasing the scanner offset by one count. The source for the geolocation errors observed with F-14 OLS data has not been determined.

COMPARISON TO AVHRR AND GOES FIRE DETECTION RESULTS

To date there have been very few direct comparisons of DMSP fire detection and results obtained with other satellite systems. Using a set of known fires in New Mexico during 1996, Elvidge et al. (1998a) found that DMSP fire detection was comparable to the results obtained from AVHRR and GOES. Fuller and Fulk (2000) found a strong spatial and temporal correspondence between DMSP and AVHRR fire detection patterns in Kalimantan, Indonesia during 1997. While this topic deserves further study it is anticipated that a multisensor approach to fire monitoring will provide a better overall depiction of fire events than reliance on data from a single system.

DMSP DATA DISTRIBUTION

The distribution of real time DMSP data is restricted through the use of signal encryption. Like the NOAA Polar Orbiting Environmental Satellites (POES), DMSP broadcasts real time data continuously. Since this signal is encrypted the real time data are available only to DoD real time receiving stations. NGDC receives the data from the DMSP tape recorders, which are downlinked to groundstations in an encrypted format. The primary ground station is Thule, Greenland (Figure 9). The satellite readouts are relayed to the Air Force Weather Agency (AFWA) located at Offutt Air Force Base, Omaha, Nebraska. Following decryption the data are passed to NGDC on a dedicated T-1 data line. Data arrive at NGDC one to two hours after their collection. NGDC organizes the data into orbital swath images and generates improved satellite ephemeris using and orbital model parameterized using bevel vectors derived from radar sightings of the DMSP spacecraft.

From 1992 until late 1999 the Air Force placed a 72 hour hold on the distribution of OLS data. In December of 1999, the 72 hour hold was relaxed to three hours. Based on this policy change NGDC established a near real time ingest and processing chain in order to provide shipment of 3+ hour old OLS data. These data shipments are made using Abilene (Internet-2).

NGDC has provided near real time regional subsets of OLS data to the Pacific Disaster Center in Hawaii and the Government of Mexico CONABIO (Comision nacional para el conocimineto y uso de la biodiversidad) in Mexico City. Transpacific data shipments are made via STARTAP (Science, Technology, And Research Transit Access Point). The Asian Pacific Advanced Network (APAN) is used to transfer near real time OLS data to two locations: 1) the Japan Ministry of Agriculture, Forestry and Fisheries Research Network (MAFFIN) computer center in Tsukuba, Japan, and 2) the Australian Bureau of Meteorology via the Australian National University. The Singapore Advanced Research and Education Network (Singaren) is used to transfer OLS data to the National University of Singapore (NUS) Center for Remote Imaging Sensing and Processing (CRISP) and the Meteorological Service of Singapore (MSS). The OLS data are used for regional fire monitoring at CONABIO, MAFFIN, CRISP and MSS.

CONCLUSION

We have outlined the algorithms for nighttime fire detection with data from the DMSP-OLS and provided examples. In late December of 1999 the U.S. Air Force reduced the 72 hour hold placed on the distribution of OLS data down to three hours. This change and recent advances in the automated processing of OLS data open up new possibilities for the more widespread use of OLS data for fire detection. Because of it's long track record of continuous operation (back to the early 1970's) and prospects for continuos operation to the year 2010 or beyond, DMSP should be classed as one of the operational systems for global fire monitoring during the coming decade. DMSP will eventually be replaced by the National Polar Orbiting Environmental Satellite System (NPOESS) which will preserve the low light sensing capability initiated with the OLS. For several years at the end of the coming decade it is anticipated that both NPOESS and DMSP will be operating.

The nocturnal VNIR emissions from fires are frequently high enough to saturate the visible band OLS data at the gain settings that are typically used (see Elvidge et al., 1999).The saturated data can be used for fire detection, but cannot be converted into a radiance value that could be used to infer the fires size or temperature.This saturation problem is also common in GOES and AVHRR fire observations and is not unique to DMSP. However, capabilities to infer fire characteristics with DMSP data are further diminished by the fact that fires are generally not detected in the OLS thermal band data.

The mid-evening overpass of the DMSP is well off the daily peak in fire numbers, which typically occurs in the middle of the afternoon. Thus DMSP does not detect the peak number of fires in most cases. Many fires are extinguished or recede substantially after sunset. The mid-evening observation of fires by DMSP may in this sense seem inadequate. On the other hand, it is the worst fires which persist into the night and a capability to detect these fires may provide resource managers with a unique opportunity to identify the most destructive and out of control fires.

ACKNOWLEDGMENT

The authors acknowledge the DMSP Program Office and the U.S. Air Force Weather Agency for providing NOAA-NGDC with DMSP data used in this research. NGDC current work on improved automation of DMSP-OLS product generation is supported by the Program to Address ASEAN Regional Transboundary Smoke and Haze (PARTS) sponsored by the U.S. State Department East Asian Environmental Initiative. The Madagascar fire products were produced through the sponsorship of the U.S. Geological Survey, EROS Data Center and the U.S. Agency for International Development.

REFERENCES

Cahoon, D.R. Jr., Stocks, B.J., Levine, J.S., Cofer, W.S. III, O'Neill, K.P., 1992, Seasonal distribution of African savanna fires. Nature, v. 359, p. 812-815.

Croft, T.A., 1973, Burning waste gas in oil fields. Nature, v. 245, p. 375-376.

Croft, T.A., 1978, Nighttime images of the earth from space. Scientific American, v. 239, p. 68-79.

Croft, T.A., 1979, The brightness of lights on Earth at night, digitally recorded by DMSP satellite. Stanford Research Institute Final Report Prepared for the U.S. Geological Survey, pp. 57.

Elvidge, C.D., Kroehl, H.W., Kihn, E.A., Baugh, K.E., Davis, E.R., Hao, W.M., 1996, Algorithm for the retrieval of fire pixels from DMSP Operational Linescan System. In "Global Biomass Burning" edited by Joel S. Levine, MIT Press (Cambridge, Massachusetts). Pages 73-85.

Elvidge, C.D. Pack, D.W., Prins, E., Kihn, E.A., Kendall, J., and Baugh, K.E., 1998a, Wildfire Detection with Meteorological Satellite Data: Results from New Mexico During June of 1996 Using GOES, AVHRR, and DMSP-OLS. In "Remote Sensing Change Detection: Environmental Monitoring Methods and Applications" edited by Lunetta, R.S. and Elvidge, C.D. Ann Arbor Press (Chelsea, Michigan). Pages 74-122.

Elvidge, C.D., Baugh, K.E., Hobson, V.R., Kihn, E.A., Kroehl, H.W., 1998b, Detection of Fires and Power Outages Using DMSP-OLS Data. In "Remote Sensing Change Detection: Environmental Monitoring Methods and Applications" edited by Lunetta, R.S. and Elvidge, C.D. Ann Arbor Press (Chelsea, Michigan). Pages 123-135.

Elvidge,C.D., Baugh, K.E., Dietz, J.B., Bland, T., Sutton, P.C., Kroehl, H.W. 1999, Radiance calibration of DMSP-OLS low-light imaging data of human settlements. Remote Sensing of Environment, v. 68, p. 77-88.

Elvidge, C.D., Hobson, V.R., Baugh, K.E., Dietz, J., Shimabukuro, Y.E., Krug, T., Novo, E.M.L.M., Echavarria, F.R., 2000, DMSP-OLS estimation of rainforest area impacted by ground fires in Roraima, Brazil: 1995 versus 1998, submitted to the International Journal of Remote Sensing.

Elvidge, C.D., Baugh, K.E., Kihn, E.A., Kroehl, H.W, Davis, E.R, 1997, Mapping of city lights using DMSP Operational Linescan System data. Photogrammetric Engineering and Remote Sensing, v. 63, p. 727-734.

Fuller, D.O. and Fulk, M., 2000, Comparison of NOAA-AVHRR and DMSP-OLS for operational fire monitoring in Kalimantan, Indonesia. International Journal of Remote Sensing, v. 21, p. 181-187.

Kihn, E.A., 1996, Forest fire detection from DMSP Operational Linescan (OLS) imagery. In "Global Biomass Burning" edited by Joel S. Levine, MIT Press (Cambridge, Massachusetts). Pages 86-91.

LIST OF FIGURES

Figure 1. Nighttime visible band DMSP-OLS data of Africa and Europe acquired by satellite F-15 on September 11, 2000 at 20:40 GMT under high lunar illumination. Image features include moonlit clouds, gas flares, city lights, sunlight contamination and solar glare.

Figure 2a. Nighttime visible band DMSP-OLS data of southern Africa acquired September 23, 1997. Note the large number of fires detected, plus the lights of cities, such as Johannesburg.

Figure 2b. Thermal band data of southern Africa acquired simultaneous to the data shown in Figure 2a. Note the general lack of hotspots, except for a small set of points burning in Botswana.

Figure 3. Simultaneously acquired smoothed versus fine resolution nighttime visible band DMSP-OLS of fires in S.E. Asia, March 8, 1999.

Figure 4. Flow chart indicating sequence of steps used to extract fire pixels from nighttime DMSP-OLS data.

Figure 5a. Nighttime visible band image for Bolivia, Paraguay, and parts of Brazil observed in a single DMSP-OLS orbit acquired September 7, 1999.

Figure 5b. Fire product for Bolivia, Paraguay, and parts of Brazil observed in a single DMSP-OLS orbit acquired September 7, 1999. Fires are shown as black.

Figure 6. Comparison of cumulative cloud-free fire detections made during the August through December time period in Madagascar from 1992 through 1999. The images have been normalized to account for differences in the number of cloud-free observations in each year.

Figure 7. Images indicating the cumulative frequency of fire detections made during the October 1, 1994 through March 31, 1995 time period. This includes the Sahel region of Africa, south of the Sahara Desert (Figure 7a), Southeast Asia (Figure 7b), Australia (Figure 7c), and northern South America (Figure 7d).

Figure 8. Number of fire detections in forest areas in the State of Roraima, Brazil during the first three months of 1998. Lunar illumination levels are shown in the background. Note that there are three broad peaks in fire detection frequency, coinciding with periods of low lunar illumination. Fire detection frequency are suppressed when lunar illumination is high. Under the full moon the brightness contrast between fires and other land surfaces is reduced. The detection of fires under high lunar illumination is reduced further by the standard practice of reducing the gain setting on the OLS PMT during the full moon period.

Figure 9. Chain of data transfers to developed to provide 3+ hour old OLS data to subscribers, including organizations engaged in operational fire monitoring.

LIST OF IMAGES FOR FIGURES.