A probabilistic A probabilistic model of fire model of fire ignition and ignition and spreading for the spreading for the Xingu Headwaters Xingu Headwaters Rafaella Almeida Silvestrini – UFMG Britaldo Silveira Soares Filho – UFMG Ane Alencar – IPAM, UFL Hermann Oliveira Rodrigues - UFMG Daniel Curtis Nepstad – Moore Foundation/ IPAM Paulo Brando – IPAM, UFL 13/10/05 Mauro Maciel
24
Embed
A probabilistic model of fire ignition and spreading for the Xingu Headwaters Rafaella Almeida Silvestrini – UFMG Britaldo Silveira Soares Filho – UFMG.
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
A probabilistic A probabilistic model of fire model of fire ignition and ignition and spreading for the spreading for the Xingu HeadwatersXingu Headwaters
The frequency of fire have been increasing dramatically (Cochrane et al., 1999; Alencar et al., 2006);
Forest Fires are a major Brazilian contribution to global warming. (FEARNSIDE, 2002);
Reduce forest ability to retain water, exacerbating flooding, erosion and seasonal water shortages. As a consequence, it increases forest flammability (Cochrane et al., 2003);
Lead the Forest remnants to a cycle of self-destruction (Nepstad et al., 1998)
Aerosols released by forest fires disrupt normal hydrological processes and reduce rainfall;
Affects human health. So, fire modeling is important not only to understand fire
drivers in the Amazon but also to check what may happen in the future.
Structure of the Model
1. Simulate the occurrence of hot pixels along the Amazon forest fringe;
Hot pixels: pixels captured by NOOA-12 night.
2. Based on the hot pixels, simulate fire spreading at the Xingu Headwaters.
Most of these fires are understory fires.
3. Feedback of fire in the quantity of fuel
(biomass) and climate at the Xingu Headwaters.
CARLUC (CARbon and Land-Use Change Model)
Step 1: Modeling Hot Pixels
Part 1: Modeling the hot pixels
1. Spatial Modeling: anthropogenic and biophysical variables Spatial ProbabilityWeights of evidence – Annual Maps
where α is a constant that depends on the month of the year.
Given a high VPD value, hot pixels will concentrate in area of higher anthropogenic activity.
months αFebruary and March 0.2other months 0.4
Modeling the hot pixels
Example: August, 2004
0.4 * +0.6
*
=
Modeling the hot pixels
Simulating the hot pixelsThe simulation is done using 10 steps, each
one analyzing 10% of the cells, in each one:
1. There is a hot pixel if:{Rand(dist)}- γ < Final Prob < {rand(dist)} +
γ, where: rand(dist) = random number from the “dist”
probability; γ = constantFurthermore, x% of the cells are allocated as null
cells.
Parameter February - March May - SeptemberOctober - December,
January, April
γ 0.05 0.1 0.4x 90% 85% 90%
distBeta(0.985, 0.1) truncada em 0.5
Beta(0.985, 0.1) truncada em 0.5
Weibull(13,0.6)
Control the number of hot pixels Control the spatial distribution
2. The probabilty map is adjusted: Hot pixels neighboring cells have their probability increased by (1+ Moran Autocorrelation Index for month i)
Modeling the hot pixels
Observed hot pixels – August, 2004
Simulated hot pixels – August, 2004
Modeling the hot pixels
Validation (2002, 2004, 2005) 1. ROC Statistic:
above 0.85 in all months of all years.
2. Fuzzy Similarity: Match of 60~70% within a 10 km x 10 km resolution for the dry months.
3. Time series of observed and simulated hot pixels:
The number of simulated hot pixels was similar than the observed ones and the temporal distribution of the simulated events followed the seasonal tendency. Considering the whole year, the model predicted, in average, 15% more or less than the observed.
Modelagem dos focos de calor
Part 2:Modeling fire spreading
Data
Fire Scars in Xingu region, mapped by Ane Alencar. Most of them are not detected by satellite – understory fires. Is there a relation between these scars and the Hot Pixels?
Relation between hot pixels and understory fire Are the Hot Pixels and fire scars spatially
dependent?Ripley’s K12 Function
Once the observed function is above the confidence envelope (dashed lines), there are evidences (with 99% of confidence) that hot pixels and forest fires are positively spatially dependent.
Probability of spreading
1. Probability based on: Distance to a hot Pixel Altitude Water bodies Forest / non forest
(<=4Km from the forest)
2. Probability based on the inner VPD For fires outside the forest this Probability is set to 1
Modeling the hot pixels
Cost Map = fire effort to spread
Weihts of Evidence Probability Map
Fuzzy Function
Monthly Maps
The Spreading Model
Cost Prob. Map
For each month
Inner VPD
Fuzzy
Functi
on
Multiply
Hot Pixels
Spreading (30)
Cellular Automata
Wind Map
Adjust the Probability
Spread Map
(initial map)
Calculate the number of neighbors in fire
Spread Map
Neighborhood
(initia
l map)
June, July, August
Example: June, July and August 2005
ValidationNumber of Simulated Scars = 1.023 Observed Scars
Considering the Fuzzy similarity we have achieved a match of 50% in a window of 9 km.
Conclusions Satisfactory validation results concerning the hot pixels model,
considering both spatial and temporal distributions. The probability map (hot pixels) is well fitted, so it can be useful
on monitoring the occurrence of these events at the Amazon. Anthropogenic variables, such as distance to town centers,
roads, protected areas, etc – play an important role when talking about fire in the Amazon, where fire is not only a natural event.
Spreading model also showed good results once its spatial and temporal distribution were satisfactory.
Besides obtaining the extension of the areas that can be burned, the spreading model is important because given a hot pixel it is possible to be attentive to the areas that are more prone to the spreading of the hot pixels, mainly the understory fires, that the monitoring satellites can not easily capture.
Next step: finish CARLUC calibration so that the coupled model (fire, CARLUC) can give an estimative of the forest feedback on fire on the next 50 years.
Cochrane, M.A. 2003. Fire science for rainforests. Nature 42: 913-919. Marengo, J. et al. The drought of Amazonia in 2005. Journal of Climate,
2006 (Submetido). FEARNSIDE, Philip M.. Fogo e emissão de gases de efeito estufa dos
ecossistemas florestais da Amazônia brasileira. Estud. av., São Paulo, v. 16, n. 44, 2002. Disponível em: <http://www.scielo.br/scielo.php?script=sci_arttext&pid=S0103-40142002000100007&lng=en&nrm=iso>. Acesso em: 08 June 2007. Pré-publicação.
SISMANOGLU, R.A.; SETZER, A.W. Risco de fogo da vegetação na América do Sul: comparação de três versões na estiagem de 2004. Anais XII Simpósio Brasileiro de Sensoriamento Remoto [online], Goiânia, GO, p. 3349-3355, 16-21/Abr/2005. [acesso em 24 de maio de 2007]. Disponível em: http://www.cptec.inpe.br/queimadas/documentos/pub_queimadas.pdf
Soares Filho, B.S. 2006. Dinamica project. Disponível em <http://www.csr.ufmg.br/dinamica> Acesso em 3 set. 2006.
Bonham-Carter, G., 1994: Geographic information systems for geoscientists: modelling with GIS. Pergamon, 398 pp.