Feasibility study into accurately modelling sulphur dioxide flux in the lower atmosphere.
MSc(R) thesis, University of Glasgow.
Full text available as:
Atmospheric pollutants are of concern for both their effects on human health and on plants and crops. Since the 1960s monitoring networks have been created, linked to international protocols regulating emissions of pollutants such as sulphur and nitrogen and also to validation studies of large-scale atmospheric transport models. One such monitoring site in the UK is at Auchencorth Moss, close to Edinburgh, where routine half-hourly measurements of sulphur dioxide are made. The time series shows a large amount of variation, and it is of interest to explore any trend in the pollutant level along with any presence of seasonal and diurnal cycles and to draw comparisons with pollutant transport model predictions. However, before carrying out such analysis, it is necessary to investigate the sources of variation. This thesis will consider the nature of the calculation of the sulphur dioxide flux, based on three simultaneous concentration measurements corrected for stability height. The need to calculate a slope estimate based on three points led to some difficulties and these were looked at to see whether these were creating difficulty when it came to modelling the fluxes. It was concluded that there were a high proportion of fluxes calculated using slope estimates with high R2 values and so any difficulty might lie in the actual data themselves rather than any technicalities in the calculations used to define the flux. From there, each variable involved in the calculation of the flux was studied, using approaches such as signal-to-noise ratios and sensitivity analysis. From these it was seen where most variation was occurring. Signal-to-noise ratio techniques did not work very well with the very low data measurements collected, which was disappointing but the values collected were generally very low suggesting a large level of noise in the data. Sensitivity analysis helped to show where most of the variation lay. Using a sampling based method it was shown that most of the variation lay in the gas concentrations themselves rather than any of the other variables involved in the calculation of the flux. This led to the conclusion that the gas concentrations rather than anything else were contributing to the difficulty of modelling sulphur dioxide fluxes. This suggested that there might be a possibility that there was no problem in the data collection approach or calculations of a flux, but perhaps the data itself was too variable to be modelled.
Chaos theory offers a different approach to the analysis of time-series and this thesis explores the use of the Lyapunov exponent to investigate chaotic behaviour over different aggregated timescales. The chaos definition used was the popular “Sensitivity based on initial conditions” approach favoured by most people in this field. Looking at how quickly two data points placed very closely together could diverge after a certain time period would show whether any predictions made would be highly susceptible to any variation would be a very useful finding. Using three different techniques gave disappointing results however. The techniques all produced results which were sometimes conflicting with each other and none of which gave any convincing argument for, or against, the existence of chaos. This led to two potential conclusions. One being that the data were very noisy, but predictable underneath this, or methods of estimating chaotic behaviour can be flawed. This thesis also looks at how Extreme Value Analysis can be used on very noisy environmental time series and how useful it can be in explaining the behaviour of the larger values measured. In this study there were some large peaks in each of the years when looking at a time series analysis. These values were studied separately from the data using Generalised Extreme Value theory and the General Pareto Distribution. The Pareto distribution approach was concluded to give the better insight into the data. This was shown to model the extreme values reasonably well though both options could be taken as valid from these approaches. Finally the measured and modelled data (collected from a Europe-wide model) were compared and analysed to see how well they compare and what techniques from each of the previous analyses can be used to bring them closer together. These tended to show that the two data sets (modelled and measured) did not match up particularly well. Techniques such as a Bland-Altman analysis and many comparison diagnostic tests were analysed to see whether there were differences between the two. Even when some findings from earlier chapters were applied to the data such as applying a minimum R2 to any slope estimates did not help.
Actions (login required)