Is the past a guide to the future? Evaluating the assumption of climate stationarity in statistical downscaling