|
The terrestrial carbon levels are determined by the inputs and outputs
of carbon to and from ecosystems. Carbon is input to ecosystems from
atmospheric CO2 via photosynthesis and becomes tied up in plant matter
before eventually returning to the atmosphere via respiration or
burning. There is mounting evidence of the importance of burning. Fire
accounts for around 10% of carbon losses from ecosystems, and has a much
larger indirect impact by reducing global forest area by 50%; and fire
is thought to drive much of the interannual variation in the net flux of
carbon between the atmosphere and ecosystems. To date, a lack of data
has imposed a severe limit on the ability to produce reliable models of
fire at global scales, especially because parameterizing terms to
describe interannual variation requires long-term data covering multiple
years.
Dr Cyril Crevoisier has recently derived a new, simple way to
extract reliable fire severity information from infra-red satellite
data. Two novel aspects of this new technique make it particularly
valuable. First, unlike previous methods, the technique does not rely on
Bayesian parameter estimation techniques constrained by strong regional
priors – rather, the estimates follow simply and directly from the
observational data within each separate grid-cell, making the effective
sample size much greater than was possible previously. Second, global
infra-red data from are available from 1979 to present: measurement
periods of this length are rare for any satellite data.
The first aim of the collaboration is to extend Cyril’s technique,
which has only been applied to 5 years of data so far, to 25-year
section of past infra-red data. This represents a significant, but
manageable, computational challenge in data retrieval, storage,
filtering, and error checking. The resultant gridded, 25-year, monthly
time-scale map of fire fluxes will be an interesting scientific result
in itself, and a valuable resource for a variety of global analyses.
The second aim is to combine the new fire data with global climate data
for the same period to parameterize a new global model of fire. This
model will predict the magnitude of ecosystem carbon loss due to fire
given the weather over previous months. The model will be parameterized
using likelihood and / or Bayesian methods in conjunction with MCMC
sampling. The model will be placed within one of the current generation
of Earth system models (the GFDL model, one of the 14 Earth system
models included in the IPCC 2001 report), and will hopefully be adopted
by other Earth system models, all of which currently use fire models
that are much less constrained by data than this new model will be. An
additional, long-term aim of the collaboration is to build analysis
capability in advance of the new sources of infra-red satellite data
that are set to become available in the near future. In principle, these
more detailed infra-red data could be used to measure CO2 global fluxes
directly.
Collaborators
|