Constraint of Vegetation Photosynthesis and Respiration Model (VPRM) Parameter Uncertainty Using a Markov Chain Monte Carlo (MCMC) Technique
Abstract
Modeling the changes to the carbon cycle and their effects on the atmosphere is a key area of research for understanding climate change. The Vegetation Photosynthesis and Respiration Model (VPRM) is a light-use efficiency model that models the biogenic flux of carbon dioxide (CO2) known as Net Ecosystem Exchange (NEE). Previous studies used methods such as non-linear least squares in order to calibrate the parameters. One other method of calibrating parameters is the Metropolis-Hastings Markov Chain Monte Carlo (MCMC) technique. The MCMC technique has not been used previously due to how computationally expensive it is. The benefit of the MCMC technique is that it is a Bayesian technique that generates a probability distribution of the posterior parameters. This probability distribution can be used to quantify uncertainty in the posterior parameters.
This study compares the MCMC technique to a non-linear least squares technique to determine its viability for use in the calibration of the VPRM. Observation data from four cropland sites from the AmeriFlux eddy covariance tower network were used with both techniques to fit the model to observations. Using the parameter correlations generated from the posterior probability distributions, a series of experiments were conducted to determine the sensitivity of the optimization of VPRM to the state vector.
The analysis of this study found that the MCMC technique reduced the RMSE of the VPRM predicted flux by more than a factor of two. The technique is viable on a site-by-site scale. However, scaling up the algorithm to more sites and land use types (LUTs) would be very computationally expensive and would necessitate the use of small batches of sites and averaging the results to prove viable. Using a single LUT to cover all cropland may also be too general and splitting the cropland LUT into different types of crops may further improve the VPRM overall.
Collections
- OU - Theses [2121]
The following license files are associated with this item: