![Hier klicken, um den Treffer aus der Auswahl zu entfernen](images/unchecked.gif) |
Titel |
Can we reliably estimate managed forest carbon dynamics using remotely sensed data? |
VerfasserIn |
Thomas Luke Smallman, Jean-François Exbrayat, A. Anthony Bloom, Mathew Williams |
Konferenz |
EGU General Assembly 2015
|
Medientyp |
Artikel
|
Sprache |
Englisch
|
Digitales Dokument |
PDF |
Erschienen |
In: GRA - Volume 17 (2015) |
Datensatznummer |
250106600
|
Publikation (Nr.) |
EGU/EGU2015-6277.pdf |
|
|
|
Zusammenfassung |
Forests are an important part of the global carbon cycle, serving as both a large store of
carbon and currently as a net sink of CO2. Forest biomass varies significantly in time and
space, linked to climate, soils, natural disturbance and human impacts. This variation
means that the global distribution of forest biomass and their dynamics are poorly
quantified.
Terrestrial ecosystem models (TEMs) are rarely evaluated for their predictions of forest
carbon stocks and dynamics, due to a lack of knowledge on site specific factors such as
disturbance dates and / or managed interventions. In this regard, managed forests present a
valuable opportunity for model calibration and improvement. Spatially explicit datasets of
planting dates, species and yield classification, in combination with remote sensing data and
an appropriate data assimilation (DA) framework can reduce prediction uncertainty and
error.
We use a Baysian approach to calibrate the data assimilation linked ecosystem carbon
(DALEC) model using a Metropolis Hastings-Markov Chain Monte Carlo (MH-MCMC)
framework. Forest management information is incorporated into the data assimilation
framework as part of ecological and dynamic constraints (EDCs). The key advantage here
is that DALEC simulates a full carbon balance, not just the living biomass, and
that both parameter and prediction uncertainties are estimated as part of the DA
analysis.
DALEC has been calibrated at two managed forests, in the USA (Pinus taeda; Duke
Forest) and UK (Picea sitchensis; Griffin Forest). At each site DALEC is calibrated twice
(exp1 & exp2). Both calibrations (exp1 & exp2) assimilated MODIS LAI and HWSD
estimates of soil carbon stored in soil organic matter, in addition to common management
information and prior knowledge included in parameter priors and the EDCs. Calibration
exp1 also utilises multiple site level estimates of carbon storage in multiple pools. By
comparing simulations we determine the impact of site-level observations on uncertainty and
error on predictions, and which observations are key to constraining ecosystem
processes.
Preliminary simulations indicate that DALEC calibration exp1 accurately simulated the
assimilated observations for forest and soil carbon stock estimates including, critically for
forestry, standing wood stocks (R2 = 0.92, bias = -4.46 MgC ha-1, RMSE = 5.80 MgC
ha-1). The results from exp1 indicate the model is able to find parameters that are both
consistent with EDC and observations. In the absence of site-level stock observations (exp2)
DALEC accurately estimates foliage and fine root pools, while the median estimate of above
ground litter and wood stocks (R2 = 0.92, bias = -48.30 MgC ha-1, RMSE = 50.30 MgC
ha-1) are over- and underestimated respectively, site-level observations are within
model uncertainty. These results indicate that we can estimate managed forests
dynamics using remotely sensed data, particularly as remotely sensed above ground
biomass maps become available to provide constraint to correct biases in woody
accumulation. |
|
|
|
|
|