EUMETSATEUMETSAT20020044, March , March 2244thth Earth Observation Dep.tEarth Observation Dep.t
AutomaticFire Detection andCharacterizationby MSG/SEVIRI
A. Bartoloni, E. Cisbani, E. Zappitelli(Telespazio/Rome)
B. Greco(ESA/ESRIN)
Monitoring the summer of 2003 forest fires of Spain and Portugal, part I
EUMETSATEUMETSAT20020044, March , March 2244thth Earth Observation Dep.tEarth Observation Dep.tTOC / CreditTOC / Credit
Approach Assumptions Fire Equation
Performance on real data Validated Data Comparison Fire Evolution Characterization False Alarm / Efficiency
Implementation SEVIRI data Processor Flow Chart
Conclusions User Needs Prospect
Credit:
- ESA/CDMC: Prototype
processor implementation and
analysis of real SEVIRI data
(this presentation)
- ESA/FiresMed: Simulated data
analysis
- FIRES/RIT-NASA: Original
ideas (for GOES)
EUMETSATEUMETSAT20020044, March , March 2244thth Earth Observation Dep.tEarth Observation Dep.tRationaleRationale
Use of existing Satellite Sensors for fire detection and characterization
Need of high spatial sensitivity
Smart processing Use Polar sensors data
Geo-stationarysensors
Polarsensors
Need of frequent revisit Geo-stationary
Optimal Sensors(non existing)
EUMETSATEUMETSAT20020044, March , March 2244thth Earth Observation Dep.tEarth Observation Dep.tApproachApproach
Quasi-stationary parameters from polar sensors
Fast changing quantities from geostationary sensor/SEVIRI
Improved spatial sensitivity by multi band, sub-pixel exploitation
Contrast analysis for fire temporal changes enhancement
Procedure based on a simple Radiative Transfer Model
Looking for thermal ground changes
SWIR/MIR/TIR atmospheric windows
Ground emissivity
Background/Fire Temperatures/Size, Atmospheric water vapor content
Include atmospheric transmissivity, ground emissivity, solar term
EUMETSATEUMETSAT20020044, March , March 2244thth Earth Observation Dep.tEarth Observation Dep.t
(t+t)
- =
SolAtm)()()()( bckfire TBTBftRttRR
Effective PixelFraction Change
Fire temperature
Background temperature
Atmospheric and Solar changes neglected (except SWIR)
One equation for each atmospheric window in IR/TIR
(t)
Fire Background
Fire EquationFire Equation
The technique is sensible to fire variations
EUMETSATEUMETSAT20020044, March , March 2244thth Earth Observation Dep.tEarth Observation Dep.t
SEVIRI
MODIS
Overall Excellent Image Quality
Geolocation pretty good Frame coregistration second order
effect in detection
Saturation effects on very large fires (see images)
SWIR
ch 3
MIR ch 4
TIR chs 7,9,10
Input Data: SEVIRI RadiancesInput Data: SEVIRI Radiances
EUMETSATEUMETSAT20020044, March , March 2244thth Earth Observation Dep.tEarth Observation Dep.tImplementation: ProcessorImplementation: Processor
SEVIRIRadiances
Sea/Land and plain Cloud Masking
Estimate Bck. Temperature
Estimate TPW
Compute Acquisition Geometry
Evaluate Solar Terms
Minimize Fire System of Equations
GroundEmissivity
Detected Fire Characteristics
Cuts on: - Minimization Residue - Fire Temperature - Minimum Fire Size - Minimum Power
Processor: Pixel Based, fully automated
and parallelizzable
EUMETSATEUMETSAT20020044, March , March 2244thth Earth Observation Dep.tEarth Observation Dep.tData for Processor CharacterizationData for Processor Characterization
Heavy Cloudy
SEVIRI/MSG Data:
sparse temporal coverage
due to acquisition station
malfunctioning
Commissioning data!
Raw Validated Data:
MODIS: ch 21/22 + ch 31
BIRD: MIR 3.4-4.2 + TIR
8.5-9.3 m
Test Case:
Portugal Fires beginning
of August 2003
No Ground Truth
analyzed !
BIRD Calibrated/Geolocated Images from BIRD Team @ DLR
EUMETSATEUMETSAT20020044, March , March 2244thth Earth Observation Dep.tEarth Observation Dep.t
Visually identify hot spots on BIRD MIR image and corresponding ones on MODIS
Define background and fire ROIs
Validation FiresValidation Fires
04/Aug/03 12:03 BIRD/MIR
Red: BIRDYellow: MODIS (4 Aug)
FireBackground
Apply the traditional sub-pixel Dozier method (MIR/TIR) to the fire ROI
Assume as firing pixels those with temperature above 350-400 K (starting plateau)
Retrieve hot spot characteristics: fire / background temperature burning area fire power location
EUMETSATEUMETSAT20020044, March , March 2244thth Earth Observation Dep.tEarth Observation Dep.tValidated Data ConsistencyValidated Data Consistency
BIRD – MODIS comparison
EUMETSATEUMETSAT20020044, March , March 2244thth Earth Observation Dep.tEarth Observation Dep.tSEVIRI Detection on 04/Aug/03 SEVIRI Detection on 04/Aug/03
10:45
12:00
11:00
12:1511:30
12:45
09:15
11:15
12:30
09:00
11:45
10:15
14:30
T B
A
Available SEVIRI frames(15 minutes apart)on 4/Aug/2003
T: Terra MODISB: BirdA: Aqua MODIS
EUMETSATEUMETSAT20020044, March , March 2244thth Earth Observation Dep.tEarth Observation Dep.tSEVIRI DetectionSEVIRI Detection
SEVIRI 04/Aug/03 12:00 upsampled
Red: BIRDYellow: MODIS (4 Aug)Green: SEVIRI (4 Aug)
BIRD 12:0304/Aug/03
downsampled
EUMETSATEUMETSAT20020044, March , March 2244thth Earth Observation Dep.tEarth Observation Dep.tPerformancePerformance
SEVIRI 04/08/03 12:00 upsampled
BIRD MIR detail Box = SEVIRI fire detection (04/Aug/03)Box Area ~ Fire Power
EUMETSATEUMETSAT20020044, March , March 2244thth Earth Observation Dep.tEarth Observation Dep.tPerformancePerformance
EUMETSATEUMETSAT20020044, March , March 2244thth Earth Observation Dep.tEarth Observation Dep.tPerformancePerformance
NOTE:-several false alarms associated to cloud boundaries (next)
-a bunch of them due to SEVIRI saturation
158 Frame Processed = 898700 pixels
Processing Time: 20 min whole Iberian Peninsula (on Athlon 1.7 GHz)
40-60 False Hits
False Alarms: 6·10-5/pixel
EUMETSATEUMETSAT20020044, March , March 2244thth Earth Observation Dep.tEarth Observation Dep.tLong lasting hot spotsLong lasting hot spots
Detected on Aug 1st by SEVIRI
Size and Power variation (SEVIRI)
generally smaller than
corresponding MODIS/BIRD
quantities
EUMETSATEUMETSAT20020044, March , March 2244thth Earth Observation Dep.tEarth Observation Dep.tLarge, 1 day fireLarge, 1 day fire
EUMETSATEUMETSAT20020044, March , March 2244thth Earth Observation Dep.tEarth Observation Dep.tA different fireA different fire
EUMETSATEUMETSAT20020044, March , March 2244thth Earth Observation Dep.tEarth Observation Dep.tLong lasting, moving fireLong lasting, moving fire
Fire moves from SW to NE
MODIS/SEVIRI/BIRD geolocations
do not fully agree
EUMETSATEUMETSAT20020044, March , March 2244thth Earth Observation Dep.tEarth Observation Dep.tFire Parameters DistributionsFire Parameters Distributions
04/Aug/2003 Fires
EUMETSATEUMETSAT20020044, March , March 2244thth Earth Observation Dep.tEarth Observation Dep.tMovieMovie
NOTE:
several frames are mission
The detection is sensible to fire condition changes
Very High radiances in TIR have been not processed
MIR VIS08 SWIR
TIR 8.7 TIR 11
EUMETSATEUMETSAT20020044, March , March 2244thth Earth Observation Dep.tEarth Observation Dep.tNoise from cloudsNoise from clouds
MIR VIS08 SWIR
TIR 8.7 TIR 11
EUMETSATEUMETSAT20020044, March , March 2244thth Earth Observation Dep.tEarth Observation Dep.t
Search MODIS images for fire outbreaks between two consecutive acquisitions:
TERRA 10:45 AQUA 14:00
24 fires identified (1-5 Aug)
17 Detected by SEVIRI (histogram)
SEVIRI/MODIS Efficiency ~ 70% some SEVIRI frames are missing
SEVIRI-MODIS Detection EfficiencySEVIRI-MODIS Detection Efficiency
EUMETSATEUMETSAT20020044, March , March 2244thth Earth Observation Dep.tEarth Observation Dep.t
Up to 2 hour before MODIS
Processor False Alarms: better than 10-4/pixel
HRV channel can be usedto improve the location
~70% MODIS efficiency
User Needs and Detection Perf.User Needs and Detection Perf.
EUMETSATEUMETSAT20020044, March , March 2244thth Earth Observation Dep.tEarth Observation Dep.t
• First results of the prototyped, fully automated, geo-stationary fire detection processor on SEVIRI commissioning data show:– comparable efficiency with MODIS (visual analysis)– reasonable false alarms rate– fire characterization capabilities
• First results of the prototyped, fully automated, geo-stationary fire detection processor on SEVIRI commissioning data show:– comparable efficiency with MODIS (visual analysis)– reasonable false alarms rate– fire characterization capabilities
Next generation of SEVIRI sensors may fulfill the tight user requirements
Next generation of SEVIRI sensors may fulfill the tight user requirements
ConclusionsConclusions
• Margin of improvements (at least):– effective cloud masking (reduce false alarms)– better integration of the SWIR channel (replace MIR on
saturated pixel)– moving toward a contextual analysis– validation on ground truth data (fire reports)
• Margin of improvements (at least):– effective cloud masking (reduce false alarms)– better integration of the SWIR channel (replace MIR on
saturated pixel)– moving toward a contextual analysis– validation on ground truth data (fire reports)