This thesis describes the development of statistical relationships between the characteristics of point and areal daily rainfall variability using measures of spatial correlation. These relationships allow estimates of 'true' areal average precipitation to be made, using a limited number of available stations, with an assessment of the uncertainty surrounding those estimates. Relationships are developed between the dry-day probabilities of point and areal-mean rainfall, and for the parameters of the gamma distribution of wet-day amounts for point and areal rainfall, using daily station data from the UK, Zimbabwe and China.
The application of these relationships to climate model evaluation is demonstrated using three General Circulation Models (HadCM3, CGCM3 and PCM). Estimates are made of 'true' gridbox average values of dry-day probability, mean wet-day amount (mean daily intensity), the gamma distribution parameters for wet-day amounts and the 95th percentile values of wet-day amounts for grid boxes from the three GCMs over the UK and South Africa.
The relationships can also be used to make estimates of the daily variability in point rainfall from an areal average series simulated by a climate model if the values of spatial variability are known. For future climate, it may not be valid to assume that spatial correlation will be stationary, because there is evidence that suggests that rainfall in a warmer climate may become more convective, and thus potentially become more localised. Investigation using rainfall simulations for Europe from the Regional Climate Model (RCM) HadRM3H found that these simulations do indicate a shift towards a greater proportion of convective rainfall in future climate (2070-2100) under SRES scenario A2. Investigation of the level of spatial correlation between RCM grid boxes, however, indicated an increase in sub-GCM-grid-scale spatial correlation, rather than the decrease that might be expected. An alteration to the spatial correlation of rainfall under warmer conditions cannot be ruled out, however, because the model's spatial resolution might limit its ability to represent the spatial characteristics of convective rainfall realistically.
A spatial analogue approach is used to make an estimate of the spatial correlation that might be experienced under future climate. A region is selected (The Netherlands) that, according to the RCM simulations, experiences a similar proportion of convective rainfall in its recent climate as is projected for the south-east UK in summer in 2070-2100 under SRES scenario A2. The values of spatial correlation for stations from this region are used as an estimate of the future levels of spatial correlation that might accompany the changed proportion of convective rainfall for the south-east UK. Observed UK spatial correlation values, and the possible future values estimated via this spatial analogue are both applied to the summer GCM grid-box precipitation simulated for this future period to estimate the characteristics of rainfall at points within that grid box. Even when spatial correlation is unchanged in future climate, the changes in dry-day probability, meanwet day amount (mean daily intensity), the parameters of the gamma distribution and the 95th percentile values of wet-day amounts change by different factors depending on whether the areal or estimated point values are considered. For the mean intensity, even the direction of change differs. Applying the changes in spatial correlation causes the magnitude of the changes to be greater still, demonstrating that failure to take into account potential decreases in spatial correlation (even if those decreases are relatively small) when downscaling or disaggregating from grid-box projections could lead to an under-estimation of the temporal variability at points within the grid box.