Presenter/Author Information

G. R. Larocque
D. Paré
R. Boutin
V. Lacerte

Keywords

carbon cycle, soil organic matter, stochasticity, boreal forest

Start Date

1-7-2006 12:00 AM

Abstract

The majority of process-based models of thecarbon cycle in forest ecosystems aredeterministic. Very few components have beenimplemented in these models to represent theuncertainty that may result from natural variation,model structure and parameter estimates. Thereare many sources of natural variation in thecarbon cycle of forest ecosystems. The mainsources of variation occur in the soil organicmatter (SOM) in terms of quantity and quality,both of which vary according to vegetation type,climatic conditions, soil characteristics (textureand structure) and carbon fluxes. For instance, thelitterfall rate and periodicity influencesignificantly the carbon input in the soil organicand mineral horizons. The litter carbon andnutrient contents affect both the SOM turnoverand nutrient cycling rates. While a proportion ofthe natural variation observed may be explainedby the differences in species composition,climatic conditions or soil characteristics, theamplitude of natural variation can nevertheless beimportant within a forest ecosystem due to theimportance of extreme small-scale naturalvariations in soil characteristics and microclimaticconditions. Models can theoretically capture thelast type of variation by using many variables inthe description of the processes. However, the useof many variables may be impractical. The morevariables and parameters a model contains, themore likely its capacity of application to simulatethe carbon cycle for different forest ecosystemtypes will decrease. Thus, there has to be acompromise between the number of variables thatmust be included in a model and its intended use.On the other hand, the interactions among sitevariables also contribute to creating thestochasticity observed in forest ecosystems. Thesefacts highlight the necessity of integratingstochasticity components in carbon cycle modelsto better deal with uncertainty.The amplitude of uncertainty in model predictionsmay be important and have an effect on thedegree to which a model is sensitive to relativelysmall variations in the inputs. As the developmentof forest management policies relies more andmore on the use of models, it is essential thatpolicy makers have good estimates of the level ofuncertainty in the predictions. Several approachesbased on Monte Carlo simulations can be used toquantify uncertainty. However, for a complexdynamic model that contains several statevariables and fluxes, the application of MonteCarlo methods can be cumbersome.For the present study, we discuss uncertainty andsensitivity issues by applying the Monte Carlomethod to a soil carbon cycle model developedfor balsam fir (Abies balsamea (L.) Mill.) andblack spruce (Picea mariana (Mill.) B.S.P.) forestecosystems in the boreal forest. The structure ofthe model is based on the presence of litter andSOM pools. Litter pools consist of plant material(foliage, twigs, understory species) and fine roots.The SOM in the organic and mineral horizons issubdivided into active, slow and passive poolsthat differ in mineralization rate. The effect oftemperature on the mineralization rate wasmodelled using results from incubationexperiments. Gaussian random distributions werecomputed on key parameters of the model.Compared with predictions using a deterministicversion of the model, the introduction ofstochasticity may generate fluctuations withoutmodifying appreciably the overall patterns ofprediction.

Share

COinS
 
Jul 1st, 12:00 AM

Using the Monte Carlo Method to Quantify Uncertainty in Predictions of a Soil Carbon Cycle Model in Balsam Fir (Abies balsamea (L.) Mill.) and Black Spruce (Picea mariana (Mill.) B.S.P.) Forest Ecosystems in the Boreal Forest

The majority of process-based models of thecarbon cycle in forest ecosystems aredeterministic. Very few components have beenimplemented in these models to represent theuncertainty that may result from natural variation,model structure and parameter estimates. Thereare many sources of natural variation in thecarbon cycle of forest ecosystems. The mainsources of variation occur in the soil organicmatter (SOM) in terms of quantity and quality,both of which vary according to vegetation type,climatic conditions, soil characteristics (textureand structure) and carbon fluxes. For instance, thelitterfall rate and periodicity influencesignificantly the carbon input in the soil organicand mineral horizons. The litter carbon andnutrient contents affect both the SOM turnoverand nutrient cycling rates. While a proportion ofthe natural variation observed may be explainedby the differences in species composition,climatic conditions or soil characteristics, theamplitude of natural variation can nevertheless beimportant within a forest ecosystem due to theimportance of extreme small-scale naturalvariations in soil characteristics and microclimaticconditions. Models can theoretically capture thelast type of variation by using many variables inthe description of the processes. However, the useof many variables may be impractical. The morevariables and parameters a model contains, themore likely its capacity of application to simulatethe carbon cycle for different forest ecosystemtypes will decrease. Thus, there has to be acompromise between the number of variables thatmust be included in a model and its intended use.On the other hand, the interactions among sitevariables also contribute to creating thestochasticity observed in forest ecosystems. Thesefacts highlight the necessity of integratingstochasticity components in carbon cycle modelsto better deal with uncertainty.The amplitude of uncertainty in model predictionsmay be important and have an effect on thedegree to which a model is sensitive to relativelysmall variations in the inputs. As the developmentof forest management policies relies more andmore on the use of models, it is essential thatpolicy makers have good estimates of the level ofuncertainty in the predictions. Several approachesbased on Monte Carlo simulations can be used toquantify uncertainty. However, for a complexdynamic model that contains several statevariables and fluxes, the application of MonteCarlo methods can be cumbersome.For the present study, we discuss uncertainty andsensitivity issues by applying the Monte Carlomethod to a soil carbon cycle model developedfor balsam fir (Abies balsamea (L.) Mill.) andblack spruce (Picea mariana (Mill.) B.S.P.) forestecosystems in the boreal forest. The structure ofthe model is based on the presence of litter andSOM pools. Litter pools consist of plant material(foliage, twigs, understory species) and fine roots.The SOM in the organic and mineral horizons issubdivided into active, slow and passive poolsthat differ in mineralization rate. The effect oftemperature on the mineralization rate wasmodelled using results from incubationexperiments. Gaussian random distributions werecomputed on key parameters of the model.Compared with predictions using a deterministicversion of the model, the introduction ofstochasticity may generate fluctuations withoutmodifying appreciably the overall patterns ofprediction.