EC
EEE EEE a
* Step 1: Clouds are automatically detected within the image
and a mask is generated. Clouds provide three potential
sources of interference. (1) Clouds can cover regions con-
taining active fires and reduce the likelihood of location.
This will provide useful ancillary information post process-
ing. (2) The cloud mask is used as input for the lightning
detection process. (3) Moonlight reflecting off cloud edges
causes glint which may be mistaken as active VNIR emis-
sions.
» Step 2: The visible band of the OLS data is filtered to gen-
erate a raster representation ofthe light sources in the
image.
* . Step 3: The mask from Step 1 and the raster image from
Step 2 are used to identify the lightning sources that are
then eliminated from the raster source.
* . Step 4: The light sources resulting from Step 3 are geolo-
cated. The result of this step is added to enhance the stable
lights image for future dates.
Step 5: The stable lights database generated from previous
dates is used to mask out inhabited areas from the current
pass (results of step 4) and the remaining lights are labeled
as intermittent lights, or fire.
+ Step 6: Aggregation of the remaining light pixels results in
the reporting of a centroid and a size (i.e., number of
affected pixels).
While portions of this process were put in place by NGDC, full
automation has been achieved through the INFER project. Both
the cloud and lightning detection steps were previously pro-
cessed manually; automation of these steps has substantially cut
processing time.
With the fire algorithm in place, the user can query the data in a
variety of ways. First, the user can request information on cur-
rent fires at either a national or regional level. These can be
viewed relative to a variety of ancillary data such as topography,
vegetative greenness, or land cover type. Second, a historical
query could be made through which a user could request an ani-
mated series of fire images over a specified time period, two
dates could be compared to determine relative fire levels, or fire
images could be compared to historical data bases. Third, the
subproduct of the fire location process, the stable lights filter,
could be queried. For instance, an emergency response manager
might be interested in a comparison of city lights before and
after a disaster to determine area and extent of damage.
Additional search algorithms in the INFER system currently
include land cover and burn scar. Scenario two will focus on the
local needs pertaining to a specific fire event. Pattern and texture
searches on higher resolution data will allow users to query on
values at risk, e.g. man made structures, access roads, or land
cover type.
Usage scenario
We describe here a scenario of how INFER technology might
have been used during a period of high fire activity in the North-
ern Rockies in August of 1994 (Bradshaw and Andrews in
press). A Multi-Agency Coordinating Group (MAC Group) was
activated due to extreme burning conditions that posed serious
threat to life and property throughout much of western Montana
and northern Idaho. The role of a MAC group is to evaluate the
fire situation and set priorities for use of limited fire suppression
resources (e.g. crews, equipment, and aircraft); they are not
involved in the management of individual fires. We expect that
INFER products would have supplemented, not replaced, infor-
mation available from other sources.
During each day’s MAC group meeting, members would have
used INFER to view the current fire activity for the entire U.S.
(satellite imagery in raw pixel form with state boundaries for ref-
erence). Verifying that most of the activity is in their area of
responsibility and that there won’t be additional competition for
resources, they would concentrate on (zoomed into) the area
they had predefined as their area of concern. Administrative
boundaries (e.g. Forest Service, National Park Service, State)
would be added for reference and a query made as to the number
of fires under each administrative jurisdiction. These numbers
can aid in the assessment of overall fire activity even though the
actual values may not agree with those from other sources due to
limitations of imagery and delays in reporting and compiling.
The number of fires as identified by satellite would be readily
available for comparison to previous days.
Ongoing fires could be examined with respect to fire potential
through models available through the Wildland Fire Assessment
System (WFAS) (Andrews and others in press). In setting priori-
ties, the MAC group could query for those fires in areas identi-
fied as extreme fire danger.
Using INFER, the MAC group would be able to zoom in to look
at higher resolution data for areas of special interest. There was
concern about Little Wolf Fire because of its location with
respect to a new development. High resolution data on fuel,
topography, roads, and structures would be used in conjunction
with weather forecasts and fire growth simulation models to
project fire growth, leading to a conclusion that the fire would
pose significant threats to life and property only in the event of
strong west or northwest winds.
The MAC group would choose some images to show at a brief-
ing later in the day, where the audience included members of the
press as well as agency personnel. In addition to showing the
area of high activity in this part of the country, additional images
would show how the season had progressed to date and how it
compared to other severe fire seasons in recent memory.
EXTENSION OF TECHNOLOGY
Although this paper has discussed the access, search, and distri-
bution of image data via the Internet in the context of wildland
fire, the applicability of content-based search is much broader.
As the use of the Internet for commercial purposes increases and
the resolution of satellite imagery becomes finer, e.g. one to two
meter, it is possible to imagine a wide array of application areas.
Emergency managers might search on oil spills or flood areas,
transportation managers on road or bridge damage, and resource
managers on vegetation health or weed infestations. The key to
these scenarios is ready access to data in usable form and within
the individual time constraints. In other words the technology
for data use must keep up with or exceed that of data collection.
196
International Archives of Photogrammetry and Remote Sensing. Vol. XXXI, Part B2. Vienna 1996
Th
Br:
Fir
int
cil,
Cas
sivi
Sat
Coi
Elv
ber!
pre:
Op:
edit
Elv
Eth
fror
Pho
Nay
the
Nev
Li |
ofL
tral
199
Pen
Con
Rio
ing.
