Pressures on water resources in much of England are set to rise owing to factors such as climate change and population growth. This is particularly the case in south-east UK, with its high population density, relatively low number of reservoirs and heavy reliance on groundwater supplies. Investigating the sustainability of water supply in this region requires a better understanding the rainfall distribution both temporally and spatially. Statistical downscaling methods were commonly adopted to simulate rainfall including under climate change scenarios. When projecting the downscaling model into future, it is often assumed that the statistical relationships observed in the historical fitting period remain stationary under the climate scenarios. However, this assumption is questionable as many downscaling models are trained on only around 30 years of data. The objectives of the thesis are therefore to 1) develop long-term statistical downscaling rainfall models to characterise the spatial and temporal distribution of rainfall for south-east UK; 2) identify the non-stationarity in statistical downscaling models; and 3) project the rainfall under climate change scenarios and assess the uncertainties in rainfall projections. To address these challenges, regression models of monthly rainfall for south-east UK were developed. Conditioned on 50 gauged sites, the model infilled and simulated the historic record from 1855-2011 in both space and time. The long record length allows more insight into the variability of rainfall and potentially a stronger basis for risk assessment than is generally possible. It is shown that, although localised biases exist in both space and time, the model results are generally consistent with the observed record including for a range of inter-annual droughts and spatial statistics. The non-stationarity was then assessed using two approaches, including visualising the non-stationarity by plotting the time series of regression coefficient estimates derived by using a moving window of 30 years, and testing whether the decadal scale change in the coefficient values is statistically significant. The results illustrate the existence of significant non-stationarity in the model, which could not be removed by adding additional available input variables to the regression. The models fitted from five 30-year control periods and climate data from five GCMs were used to generate rainfall projections in 2021-2050. Both the uncertainty introduced by non-stationarity and the GCMs was visualised. The results show that in some months the source of uncertainty introduced by non-stationarity can lead to significant uncertainty in the projections. The uncertainties in projections were then estimated using ensembles.