I am currently studying my PhD under the supervision of Mat Williams and Chris Merchant. My research interests include process based modelling, environmetrics, geostatistics, computer vision and remote sensing.
The aim of my PhD is to investigate and quantify error in spatial implementations of ecosystem models. In general, ecosystem models require a set of rate parameters to control the flow of carbon and water from the vegetative surface to the atmosphere, and a set of meteorological driving variables from which estimates of the state vector are derived. The state vector may include various pools of carbon in the form of woody biomass and leaf area, and so we also require some initial land surface conception from which to start the model.The modelling situation is complicated by the difficulty in measuring and setting parameters, and finding adequate data to drive the model. On one hand, parameters may be difficult or impossible to measure in practice, particularly if the rates of the processes they represent are very slow. On the other hand, sourcing adequate data to drive the model over the required spatio-temporal extent may be difficult, expensive, and plagued by missing observations resultant from sensor failure etc. In general we rely on some optimisation procedure to infer appropriate parameter sets, and use interpolation techniques to gap-fill meteorological drivers. As such, parameters and drivers are subject to uncertainty, which in the case of meteorological drivers is rarely assessed in practice.
One solution is the use of geostatistics as an interpolation and scaling tool to produce optimal estimates of meteorological and vegetative surfaces. Geostatistical simulation also provides a means of estimating error statistics, via monte carlo methods. Propagating these errors through our ecosystem models via ensemble runs allow us to establish the spatial accuracy of our estimates, and highlight areas where our existing models perform poorly. Advanced geostatistical techniques can also integrate remote sensing data into surface prediction, allowing us to exploit spatially sparse, but very accurate ground measurements, with spatially exhaustive but uncalibrated Earth observation data.
Through the use of the above methods I was able to investigate and partition the sources of uncertainty for regional estimates for carbon budgets. Such quantification is essential if sensible decisions are to be made in terms of emission offsetting and carbon trading.
Check out my used textbooks and technical references at amazon marketplace