In order to explore the behaviour of floods, droughts, heatwaves, and cold snaps, across the Northern Mediterranean region (35°-45°N, -10°-30°E) with a selection of hemispheric circulation predictors (drawn from the NCEP/NCAR reanalyses), two differing models have been used. Orthogonal Spatial Regression (OSR) is an inversion of a dendroclimatology technique that relies on spatial variability, with Principal Components Analysis (PCA) at its core. Radial Basis Function Artificial Neural Networking (RBF ANN) is a machine-learning pattern matching approach, capable of non-linearity. The two have been applied as direct downscaling methods to the STARDEX (STAtistical and Regional dynamical Downscaling of EXtremes for European regions) indices of extremes, across the target area. Analysis suggests that there may be significant departures between regional and seasonal contrasts in extreme behaviour, and those evident for mean climate. In addition, where warming occurs, extreme high temperatures generally show a trend of greater magnitude than the mean. Modelled links between circulation predictors and extreme climate are consistent with these results, statistically significant, largely linear, and are (in many cases) stronger for extremes than the mean. Distinct circulation regimes have been identified, as described by groups of predictors (representative of Atlantic influence, for instance), each with effects that are relevant to a particular region, season, and type of Mediterranean extreme climate.
This thesis also explores direct relationships between extreme events (quantified by the indices of extremes) and socio-economic indicators (i.e., agricultural yield, energy consumption, and excess mortality). OSR and ANN are applied again, as econometric upscaling models, between climate indices and socio-economic indicators, to provide the final link in a chain of potential predictability. Long-term (i.e., decadal) trends in the socio-economic indicators considered are consistent with non-climatic influences. However, regional variations in sensitivity to extreme climate have been identified (on a seasonal basis) that demonstrate both the advantages of using upscaling technique, and the use of station-scale predictors over spatially aggregated data. The model functions (e.g. linear, gaussian, or quadratic) most useful for modeling relationships are seen to vary between regions and seasons. Furthermore, regions that display strong trends in extreme behaviour and significant links to sensitive sectors of activity have been highlighted. This study suggests that Mediterranean climate extremes are changing over time, and that policy concerning the socio-economic impacts of those changes must be specified with regional concerns in mind.