FIMMA
Fire Identification, Mapping and Monitoring from AVHRR
I.
Csiszar
CIRA@NOAA/NESDIS/ORA
FIMMA
status
Current system:
fully automated on NT and LINUX platforms including improved navigation
Automatic
fire monitoring from AVHRR
Input data
Preprocessing
Fire detection
Data dissemination
Input
data
AVHRR level-1b files
from the WWB Polar Data Server
HRPT: High Resolution
Picture Transmission
direct-readout
LAC: Local Area Coverage
recorded on board the satellite
Preprocessing
Radiometric
calibration
vis/NIR:
time-dependent coefficients from vicarious calibration
IR:
on-flight coefficients
(Atmospheric
and angular corrections)
(not done for fire detection)
Geo-referencing
currently:
interpolation between anchor points as provided within the level-1b files (from
NESDIS operational navigation) -> error can be up to 5 km -> further
correction is needed for fire detection
future:
automatic landmark-based geo-referencing
Automatic
landmark-based geo-referencing
Commercial
software: PCI/GeoComp-n
successfully
used at the Canada Center for Remote Sensing for operational fire monitoring
Main
processing steps:
data input:
conversion
of level-1b files into PCI .pix database format
ephemeris
update
current TBUS
for good first guess navigation
Calibration
standard
procedure as in current operational processing
geocoding
ground
control points (GCPs) are found automatically from a image chip database - no
manual selection of GCPs is needed
geocoded
scenes in pre-defined projection
Automatic
nighttime geocoding
Nighttime fire detection (?)
Nighttime
geocoding:
criteria for valid
geocoding need to be relaxed
number of GCPs found
minimum GCP span
maximum RMS
good fake channel data
-> more GCPs found
stable hot spots (e.g.
Weirton Iron Works etc.)
Algorithm:
low saturation level of
ch. 3 on pre-KLM satellites
higher saturation level,
but no daytime ch. 3b data on KLM satellites -> need to develop
near-terminator and nighttime algorithms and nighttime geocoding
Validation:
no quantitative
validation yet
Automatic processing
In
NT/MS-DOS environment (Perl scripts and EASI/PACE routines)
Batch
processes for
acquisition
of latest TBUS bulletins (TBUS)
acquisition
of latest level-1b files and preparing for geocoding (Level1b)
submitting
files for geocoding (Geocode)
exporting
geocoded scenes into binary format
In
LINUX environment (shell scripts)
Cron jobs
for
running fire
detection
forwarding
results
The fire detection algorithm: physical basis
Rapid increase of mid-IR
(3.5-4.0 um) radiance with fire temperature
B(8, T)=C1/ [85(eC2/8T -1)]
AVHRR channel 3(b): 3.7
micron
Problems in AVHRR-based fire detection
Obscuration
clouds
thick smoke
(wet)
False
signals
hot/bright
surfaces: heat + solar reflection
sun glint
(direct reflection) from water bodies
cloud edges:
solar reflection + emission from underlying hot surface
Current Algorithm: Fire Identification, Monitoring and Mapping from
AVHRR (FIMMA)
Based on NASA
Goddard Space Flight Center two-step day-time scheme
(Justice et al. 1996)
Potential fire
pixels Original Modified
T3 >316K; T4 >280K T3 >316K; T4 >280K
T3-T4 >12K
if glint angle<40o then
R1<30% and R2<30%
Valid background T3<316; T4>280 T3<316; T4>280
Calculate mean and F for {)T},{T3},{T4}
)T=T3-T4
Classification
)T' ={)T}+2F({)T}) )T> )T' ; T4> {T4} )T> )T'; T>{T4} or 5K
T'3 ={T3}+ 2F({T3}) T3>T'3
or 325K
Remaining issues
Algorithm-related:
imperfect
cloud detection
commission
errors
omission
errors
Sensor-related:
low
saturation level of ch. 3 on pre-KLM satellites
higher
saturation level, but no daytime ch. 3b data on KLM satellites -> need to
develop near-terminator and nighttime algorithms and nighttime geocoding
Validation:
no quantitative validation yet
limited
ground truth data
Data dissemination
Experimental/semi-operational
ASCII files
with latitude, longitude of fire locations
being moved into operations by OSDPD/SSD
Data
are being integrated into the Hazard Mapping System (HMS)