HR: 0800h
AN: B41B-0108    [Abstracts]
TI: Predicting the Global Distribution of Fire Emissions: Can Supervised Classifiers Inform Physical Models?
AU: * Newman, D J
EM: newman@uci.edu
AF: UC Irvine, Dept Earth System Science, Irvine, CA 92612 United States
AU: Randerson, J T
EM: jranders@uci.edu
AF: UC Irvine, Dept Earth System Science, Irvine, CA 92612 United States
AB: An accurate prediction of fire emissions is an important piece of the climate variability puzzle. Accurately predicting emissions depends on an accurate prediction of ignition events, fuel moisture levels, fire spreadrates, and combustion completeness of different fuel types. Current global vegetation models predict fire ignition and spread using simplified models that depend primarily on fuel load and litter moisture. We used data-mining and machine-learning techniques to find relationships between potential drivers of fire and the occurrence of fire (as measured by the TRMM-VIRS satellite). We compared the fire prediction accuracy of physical models and supervised classifiers trained on 1998-2002 data. We finally discuss how relevant drivers can be incorporated into fire models to improve accuracy of predicted fire emissions.
DE: 1615 Biogeochemical processes (4805)
DE: 0400 Biogeosciences
SC: Biogeosciences [B]
MN: 2004 AGU Fall Meeting