|Publication Type||Journal Article|
|Year of Publication||2005|
|Authors||Barrett, D, Hill, MJ, Assoc Prof Hutley, LB, Beringer, J, Xu, JH, Cook, GD, Carter, JO, Williams, RJ|
|Journal||Australian Journal of Botany|
|Pagination||689 - 714|
|Keywords||carbon-dioxide, chain monte-carlo, northern australia, parameter-estimation, seasonal patterns, sensing data assimilation, surface-temperature, terrestrial biosphere, tropical savanna, vegetation cover|
A 'multiple-constraints' model-data assimilation scheme using a diverse range of data types offers the prospect of improved predictions of carbon and water budgets at regional scales. Global savannas, occupying more than 12% of total land area, are an economically and ecologically important biome but are relatively poorly covered by observations. In Australia, savannas are particularly poorly sampled across their extent, despite their amenity to ground-based measurement ( largely intact vegetation, low relief and accessible canopies). In this paper, we describe the theoretical and practical requirements of integrating three types of data ( ground-based observations, measurements of CO2/H2O fluxes and remote-sensing data) into a terrestrial carbon, water and energy budget model by using simulated observations for a hypothetical site of given climatic and vegetation conditions. The simulated data mimic specific errors, biases and uncertainties inherent in real data. Retrieval of model parameters and initial conditions by the assimilation scheme, using only one data type, led to poor representation of modelled plant-canopy production and ecosystem respiration fluxes because of errors and bias inherent in the underlying data. By combining two or more types of data, parameter retrieval was improved; however, the full compliment of data types was necessary before all measurement errors and biases in data were minimised. This demonstration illustrates the potential of these techniques to improve the performance of ecosystem biophysical models by examining consistency among datasets and thereby reducing uncertainty in model parameters and predictions. Furthermore, by using existing available data, it is possible to design field campaigns with a specified network design for sampling to maximise uncertainty reduction, given available funding. Application of these techniques will not only help fill knowledge gaps in the carbon and water dynamics of savannas but will result in better information for decision support systems to solve natural-resource management problems in this biome worldwide.