Start Date
1-7-2006 12:00 AM
Abstract
WETLAND-DNDC, a process-based model that integrates soil, vegetation andhydrology, was developed to represent unique biogeochemical properties and processesof wetland ecosystems. Recent studies have demonstrated great potential of the model byapplying it to different wetland sites, from Minnesota to Florida. However, to have abroad application of the model to more systems and at larger scales, a full uncertaintyanalysis is needed to understand its behaviors and determine its critical variables andparameters. With a wetland in Florida as the test site, we used the Monte Carlo method torun WETLAND-DNDC, simulating carbon (C) and nitrogen pools/fluxes driven by acombination of various input parameters (e.g., climate variables, soil properties, forestcharacteristics, hydrological conditions) and determining impacts of varying parametervalues on C dynamics in wetland ecosystems. Input variables included daily maximumair temperature, total organic carbon content, soil saturated conductivity, available carbonin plants, water use efficiency, and water table fluctuation. The Latin hypercube samplingmethod was used to determine values of each input variable by taking ten samples withinits range, one from each of the ten equal probability strata. The model predictionsincluded annual net primary productivity (NPP), annual net ecosystem exchange (NEE),and annual methane flux (CH4). Error propagation from input parameters to outputvariables in the modeling process was examined and uncertainties associated with thecritical parameters were quantified. The results indicated that the most critical uncertaintysources differed for NPP, NEE and CH4. For example, the accuracy of climate and soilfertility data needs to be improved to reduce uncertainty in NPP or NEE predictions,whereas water table data were the most critical to CH4 predictions. Uncertaintyinformation is of great interest to both scientists and policy makers to assess thereliability of model predictions.
Uncertainty analysis of WETLAND-DNDC: Errors and critical parameters for predicting carbon dynamics in wetland ecosystems
WETLAND-DNDC, a process-based model that integrates soil, vegetation andhydrology, was developed to represent unique biogeochemical properties and processesof wetland ecosystems. Recent studies have demonstrated great potential of the model byapplying it to different wetland sites, from Minnesota to Florida. However, to have abroad application of the model to more systems and at larger scales, a full uncertaintyanalysis is needed to understand its behaviors and determine its critical variables andparameters. With a wetland in Florida as the test site, we used the Monte Carlo method torun WETLAND-DNDC, simulating carbon (C) and nitrogen pools/fluxes driven by acombination of various input parameters (e.g., climate variables, soil properties, forestcharacteristics, hydrological conditions) and determining impacts of varying parametervalues on C dynamics in wetland ecosystems. Input variables included daily maximumair temperature, total organic carbon content, soil saturated conductivity, available carbonin plants, water use efficiency, and water table fluctuation. The Latin hypercube samplingmethod was used to determine values of each input variable by taking ten samples withinits range, one from each of the ten equal probability strata. The model predictionsincluded annual net primary productivity (NPP), annual net ecosystem exchange (NEE),and annual methane flux (CH4). Error propagation from input parameters to outputvariables in the modeling process was examined and uncertainties associated with thecritical parameters were quantified. The results indicated that the most critical uncertaintysources differed for NPP, NEE and CH4. For example, the accuracy of climate and soilfertility data needs to be improved to reduce uncertainty in NPP or NEE predictions,whereas water table data were the most critical to CH4 predictions. Uncertaintyinformation is of great interest to both scientists and policy makers to assess thereliability of model predictions.
