space applications institute global vegetation monitoring unit operational implementation of burnt...
TRANSCRIPT
Space Applications InstituteGlobal Vegetation Monitoring Unit
Operational Implementation of Operational Implementation of Burnt Area Algorithms for Burnt Area Algorithms for
GBA2000:GBA2000:Initial Results for Australia and Initial Results for Australia and
AfricaAfricaKevin J. Tansey & Jean-Marie Grégoire
Global Vegetation Monitoring Unit (GVM)
Joint Research Centre (JRC)TP-440, I-21020, Ispra (VA), Italy
Tel: +39 0332 785769E-mail: [email protected]
GOFC-FireSatellite Product Validation Workshop
Space Applications InstituteGlobal Vegetation Monitoring Unit
Pre-processing ObjectivePre-processing Objective
o To produce a ‘clean’, daily, global dataset
o Possible cleaning measures:•BRDF correction
•Extreme view zenith masked
•Maximum NDVI composite masked
•Cloud masked
•Cloud shadow masked
•Non-burnable area masked
Space Applications InstituteGlobal Vegetation Monitoring Unit
Maximum off-nadir observation angle = 50.5
Extreme Viewing Zenith Extreme Viewing Zenith MaskMaskSaint, G., 1998, The VEGETATION Programme
Maximum viewing zenith angle 60
Viewing zenith angle threshold <= x
Space Applications InstituteGlobal Vegetation Monitoring Unit
Max. NDVI ProblemsMax. NDVI Problems
o Boreal forest fire scar detectiono Speckle in cloud and cloud shadowso BRDF correction algorithms could be
usedo NDVI comp. areas could be masked
Space Applications InstituteGlobal Vegetation Monitoring Unit
Cloud MaskingCloud Maskingo Several approaches available:
• Compositing of cloud-free data
• Thresholding of SPOT VGT radiance values
• Use the status map provided with S products
o Thresholding of b0 (blue) and mir (swir)• Values suggested by Kempeneers et al.
Vegetation 2000 Meeting, 3-6 April 2000, Lake Maggiore
http://vegetation.cnes.fr:8080/vgtprep/vgt2000/kempeneers.html
• Different thresholds are tested for S products to capture all of the cloud
Space Applications InstituteGlobal Vegetation Monitoring Unit
Cloud Shadow MaskingCloud Shadow Maskingo Burnt areas and shadow have similar
spectral propertieso Calculate cloud -> shadow vector
•Using available sun and viewing angles•Using the latitude and longitude of each
pixelo Assume a cloud height (e.g. 10km)
•Variable input parameter (test on small areas)
•Or, assume a maximum tropopause heighto Mask pixels between cloud and shadow
Cloud pixel
Intermediate shadow pixels
Shadow pixelCloud – shadow vector
Space Applications InstituteGlobal Vegetation Monitoring Unit
CLOUD + CLOUD SHADOW
Space Applications InstituteGlobal Vegetation Monitoring Unit
Non-Burnable Area MaskNon-Burnable Area Masko Utilize global land cover products
• IGBP DIS landcover datasetLoveland, T.R. et al., 2000, IJRS, Vol. 21,
1303-1330
http://edcdaac.usgs.gov/glcc/glcc.html
•Uni. of Maryland global landcover dataset
Hansen, M.C. et al., 2000, IJRS, Vol. 21, 1331-1364
http://glcf.umiacs.umd.edu/data.html
o Mask urban, snow/ice & bare soil classes
Space Applications InstituteGlobal Vegetation Monitoring Unit
Compositing CriteriaCompositing Criteriao 10-day min-NIR composites
No Masking Applied Masking Applied
Australia’s Cape York
Space Applications InstituteGlobal Vegetation Monitoring Unit
Compositing Compositing ImprovementsImprovements
o ‘Salt and pepper’ speckle removed
o Permanently cloudy areas removed
o Cloud shadow masked
o Burnt areas more clearly delimited
Space Applications InstituteGlobal Vegetation Monitoring Unit
Compositing Compositing ImprovementsImprovements
o 1 month min-NIR composites
No Masking Applied Masking Applied
Mozambique
Space Applications InstituteGlobal Vegetation Monitoring Unit
Compositing Compositing ImprovementsImprovements
o 1 month min-NIR composites
No Masking Applied Masking Applied
Mozambique
Space Applications InstituteGlobal Vegetation Monitoring Unit
Operational Operational ImplementationImplementation
o Series of discrete moduleso Programs designed to have maximum
functionality and flexibilityo Well documented and repeatableo Provide feedback on results to
respective partnerso Test algorithms over other regions
easilyo Example of NRI algorithm in C. Africa
Boschetti, L. et al., (submitted), A multitemporal change-detection algorithm for the monitoring of burnt areas with SPOT-Vegetation data, MultiTemp 2001 Workshop, 13-14 September 2001, Trento, Italy
Burnt Area Map (November 1999 - March 2000)
November 1999 December 1999 January 2000 February 2000 March 2000
Algorithm defined by NRI, UK
Comoe National Park:Ivory CoastBurnt Dec/Jan
National Parks in Benin, Niger and Burkina FassoBurnt Nov/Dec/Jan
Kainji Lake National Park: NigeriaBurnt Dec/Jan
Garamba National Park: ZaireBurnt Jan/Feb
Murchison Falls National Park:UgandaBurnt Jan/Feb
Gambella, Jikoa, Tedo National Parks and CHA’s: EthiopiaBurnt Jan/Feb/March
Burning in National Park & Controlled Hunting Areas (CHA’s)November 1999 – March 2000
Algorithm defined by NRI, UK