| Understanding the global carbon (C) cycle is currently in the main focus of Earth system science. Major uncertainties in our current knowledge are due to an incomplete understanding of carbon cycling within terrestrial ecosystems. The existing model structures that seek to explain the terrestrial part of the C cycle have been developed partly based on a priori assumptions about the underlying processes. The latter have been partly derived from laboratory experiments. Unfortunately, we cannot say yet with full certainty if these model structures are of general validity or if they dependent on the conditions in which they were built. In order to minimize the need of theoretical assumptions about the underlying processes in the terrestrial carbon exchanges, I will use a reverse engineering method through which the structures of the models will only emerge from the data themselves. The method I have chosen is called Gene Expression Programming and it is a machine learning algorithm being based on the following steps: it starts with a collection of proposed solutions to the problem (model structures), it trains these over a number of generations and evolves them through different genetic operators and it selects the fittest specimen in the solution based on previously indicated criteria. The advantage of using this type of method is that it can start from already existing model structures and then it can combine sub-expressions of these structures and build other more complex and exhaustive models, or just generate entirely new ones. After applying this technique I will validate the new results over the ones that are already in the literature and hopefully I will be able to draw conclusions about the Carbon fluxes in the terrestrial biosphere based only on the actual, existing data on them. |
|