ctcd fire activities
DESCRIPTION
CTCD Fire Activities. P. Lewis, L. Rebelo, I. Woodward, P. Bowyer, B. H eung, M. Wooster, D. Roy. Fire Workpackage. Aim: Provide improved estimates and model of global C-release from fires Identification of existing Burn-Affected Area Datasets Calibration and Testing of SDGVM Fire Module - PowerPoint PPT PresentationTRANSCRIPT
CTCD Fire Activities
P. Lewis, L. Rebelo, I. Woodward, P. Bowyer, B. Heung, M. Wooster, D. Roy
Fire Workpackage
Aim:– Provide improved estimates and model of
global C-release from fires Identification of existing Burn-Affected Area
Datasets Calibration and Testing of SDGVM Fire Module End-to-end testing via Satellite C-emission
estimates Generation and Testing of Burn-Affected Area
Datasets and Associated Products
Mapping of day of burn
Degree of burning (~= cc*f)
Name Data Type Spatial Extent Time Period Spatial Resolution
MODIS Thermal Anomalies product Active Fire Global From 2000 to present 1km x 1km
WFW Active Fire Global From 1996 to 2001 0.5 x 0.5 degree
WFA Active Fire Global Nov. 1995 to June 2004 with process on going
1km x 1km
Web fire mapper Active Fire Global 1km x 1km
TRMM VIRS Monthly Fire Product Active Fire Regional Jan 1998 to Aug 2004 0.5 x 0.5 degree
CIMSS Active Fire Regional May to Oct for 1995 to 1997 4 km x 4 km
AVHRR fire atlas (Australia) Active Fire Regional 1993 1km x 1km
AVHRR fire atlas (South America) Active Fire Regional 1993 1km x 1km
AVHRR fire atlas (Africa) Active Fire Regional June 1992 to June 1994 1km x 1km
GLOBSCAR Burned Area Global 2000 1km x 1km
GBA2000 Burned Area Global 2000 1km x 1km
MODIS Burned Area Product Burned Area Global 2000+ 500 m x 500 m
GLOBCARBON Burned Area Global 1998-2007 10km +
GBA 82-99 1982 to 1999 8km x 8km
Canadian Forest Service Other Regional 1959 to 1999 All fire > 200 ha
Mouillot’s Database Other Global 1990 to 2000 1 x 1 degree
Issues
– Many EO datasets single year only Though increasing production of longer time
series datasets
– Active fire detection underestimates fire activity
– Non-geo-located products double count fires at swath edges
– Burn-affected area mapping needs to account for BRDF effects
– General lack of ‘validation’
Calibration and Testing of SDGVM Fire Module
regression models based on the simulated SDGVM result
– plant function types, temperature, surface soil content and precipitation
– Currently using Global Burn Area (GBA) and World Fire Atlas (WFA) data
fitted to estimate the number of fire occurs in a 1 degree pixel.
Calibration and Testing of SDGVM Fire Module
Moved to 2-step model:– Logistic model of Fire Occurance– Model to estimate number of fires
Model testing– Canadian large fire data base– SDGVM run to simulate a fraction of the area burn in Canada
between 1959 and1999. – data ½ degree resolution. – Initial analysis:
SDGVM fire estimated burnt area is a factor ~3 greater than the LFDB result.
Also shows less variation does not pick up the extreme years the time-series from 1958 to 2000 for the SDGVM and the LFDB show
little correlation. Current efforts are to understand the possible reasons and hence how to improve the SDGVM prediction.
Demonstrates requirement for further work on model development and requirement for observations
End-to-end testing via Satellite C-emission estimates
Wooster producing C-emission estimate from Fire Radiative Energy
FRE from Meteosat Seviri (2004+)– And Boreal region MODIS (2000+)?– Diurnal activity from Seviri
Allows end-to-end testing of models– And estimation of other terms when
combined with satellite burn affected area
Generation and Testing of Burn-Affected Area Datasets and Associated Products
Working with David Roy in development and testing MODIS burn-affected area product
Testing alternative methods
Examining derived products in S. Africa– Fire return frequency– Seasonality
(#fires in 5 years)
Monthly area burned as a proportion of the annual total
Seasonality of burning 2004
‘Degree of burning’ 2004
Degree of burning
Fire Frequency
40% of the land surface burned, with 6% (area of approximately 131,420km ° ) burning during each of the five annual fire seasons.
Higher fire frequencies identified in savanna and grassland ecosystems, with shrublands and deciduous broadleaf forests burning less frequently.
Fire return intervals indicate that locations which burn every year do so at the same time each year.
These areas also have a distinct spatial pattern and are predominantly located in the northern section of Angola, southern Zaire and northern Zambia, as well as in a belt along the Namibia/Angola/Botswana borders.
Spatial extent
Between 27% and 32% of the study area has burned during each of the five years of observation. This equates to an area of approximately 610,000 to 690,000km2 .
The distribution of burning within each of the main vegetation types is similar from year to year, with a much larger proportion of deciduous broadleaf forests, woody savannas and savannas burning each year in comparison to shrublands and grasslands.
Summary #1
Fire models (e.g. SDGVM) based on understanding of ecology and fire interactions– Very limited datasets previously available
for testing– EO provides potential for much greater
spatial sampling and analysis– FRE provides potential for end-to-end
testing of model and C-release
Summary #2
Many EO datasets generated– Active fire detection underestimates activity
and depends on time of observation– New generation of burn-affected area
products under generation provide most high quality information But need furter testing/validation
– Rich source of information available for analysis
– But over limited time period
Spare slides
Active Fire Datasets: Global
MODIS Thermal Anomolies (NASA)– 1 km resolution 2000+– 2x daily (morning/afternoon)– High confidence of detection if fire observed– Also MODIS Rapid Response System
World Fire Web (GVM/JRC)– 0.5o resolution AVHRR 1996-2001– Errors of commision & omission– Different processing methods used at different receiving stations– Frame overlap issues– Discontinued
World Fire Atlas (ESA)– 1995-2004+ night time (A)ATSR – Frame overlap issues– Revisit period ~3 days
Active Fire Datasets: Regional
TRMM VIRS Monthly Fire Product– 0.5o resolution, 1998-2004+– 38oS to 28oN– 2+ observations/day– Moderate detection capability with higher probability of
detection in non-forest land cover classes GOES-8 ABBA Fire Product
– 4km x 4km, 1994-1997– 4x/day– Coverage S. America
AVHRR Fire Atlas (ESA ESRIN) – S. Hemisphere, day time AVHRR, 1993 (1992-1994 Africa)– High confidence detections only
Burn Affected Area (Global)
GLOBSCAR (ESA ESRIN)– 1 km, year 2000, monthly or annual– Daytime ATSR-2 data (3 day repeat) 10:30 am– 2 algorithms: combination gives low error of commission– Particular underdetection in United States (open shrubland and grasslands), Australia (open
shrublands), Zimbabwe (croplands) and Brazil (broadleaf evergreen forest) GBA-2000 (JRC/GVM)
– 1 km, year 2000 SPOT VGT– Regional algorithms used– R2 comparisons with TM data from 0.4 (Mozambique) to 0.99 (Botswana)– False detections in sub-Saharan Africa include false detections due to flooding of non-
permanent water features as well as due to the presence of hot dark rocks. (but small proportion)
– Only burned areas of at least 400ha in size output MODIS Burned Area product (NASA)
– David Roy will discuss– 500m resolution day of burn, monthly product 2000+– Africa testing: 99.7% correct detections, and lowest in Mozambique (74.3%) (overall R2 0.8)
GLOBCARBON– Steve Plummer will discuss– ERS-2 / ATSR-2, ENVISAT / AATSR, and SPOT /VEGETATION. ENVISAT / MERIS– global monthly maps of burnt areas for the period 1998-2007 in 10 km, 0.25° and– 0.5° resolution– based on the experience of both GLOBSCAR and GBA-2000.– CTCD testing dataset
Burn Affected Area (Regional)
Canadian Forest Service (Large fire database)– 1959-1999+, fires > 200ha– Small proportion of fires but 97% of total area
burned– the date (year, month, day, start date, detect
date), location (latitude, longitude, Province), cause, size and ecozone of each fire detection.
Mouillot’s Database – 20th Century fire, 1o resolution– Reconstructed from various data sources
(incomplete) uses ATSR for recent fires