| Popis: |
Accurately estimating how much carbon is leaving the atmosphere and being taken up by plants, by processes such as photosynthesis, is critical in order to make accurate climate forecasts. There is a large uncertainty of this atmosphere-plant carbon flux, sometimes referred to as Net Ecosystem Exchange (NEE), therefore reducing this is essential. One way of doing this is through Data Assimilation (DA), the framework by which data and models are combined together in a statistically optimal way. A key aspect of DA is that the uncertainty on the estimate of interest, e.g. NEE, is less than either the uncertainty on using either the model or observations on their own. DA can also been used to estimate model parameters, which have traditionally been estimated from expert knowledge or from small scale studies. While DA has gained much interest as a powerful tool in estimating model parameters and quantities such as NEE, there are a number of issues surrounding its use, which are not yet properly understood. These include: (1) When using DA to estimate parameters using ground observations: (a) Understanding the limitations of DA and the conditions it performs best; (b) Determining likely factors that cause variations in parameter estimates. (2) Assessing the impact of assimilating satellite observations of leaf area index to improve the model states, and whether DA is robust against unrealistic features of the satellite data. The aim of this PhD was to address and learn more about these issues. This was done by using the evergreen and deciduous versions of the Data Assimilation Linked Ecosystem (DALEC) model. The main findings are summarised in the following four paragraphs: The Ensemble Kalman Filter (EnKF) is good at estimating parameters using synthetic NEE data. It was found that between 2 and 5 years of this data was required in order for the parameters and NEE forecasts to be close to the truth. There was for the most part very little difference to the EnKF parameter estimates and NEE forecasts whether very noisy or very non-noisy observations were used, or whether 20% or 100% of the daily observations were present in the dataset. For the Metropolis algorithm, most of the runs had to be discarded as it was found that the the algorithm was not converging for the global minimum for these runs; this caused some other problems, in particular residuals between the modelled and observed NEE were autocorrelated. For these discarded runs, the parameter estimates tended to be far from the truth and the 90% posterior intervals rarely included the truth. For the remaining runs, where the converse of the above was found to be true, as dataset length increased from 1 to 5 years, the posterior parameter distribution coincided with the truth to a greater extent. Using the Metropolis algorithm and assimilating three years daily NEE observations (with around 60% data coverage) and around 10 LAI observations during this period, it was found that parameter estimates were sensitive to the initial value of the labile carbon store. Moreover, the parameters were close to their true values if the true initial value of the labile C pool was used. It was also found that when these initial conditions were treated as parameters, although the modal value of the corresponding marginal posterior distributions were far from the truth, every other aspect of the model (parameters and trajectories of the model states) agreed well with the truth. This supported the common approach by many of the DA community that treating initial conditions as parameters is preferable than keeping them fixed (using site inventory data or from model spin-up). The novelty of this part of the thesis was for the first time an emulator was applied to a DA scheme. Finally, the EnKF was used to estimate the LAI and NEE model states, using a fixed parameter set and LAI data from the MODIS satellite sensor. It was found that processing the MODIS LAI in order to correct for unrealistic features of the dataset, such as excessive temporal variation and very small uncertainties, improved the fit of the modelled to observed NEE after assimilation. The improvement in the fit was significantly better for Gross Primary Production (GPP). |
