Global circulation models (GCM) provide state-of-the-art information about the behaviour of the climate under different emission scenarios via catalogues of simulations, yet they are not calibrated for extreme events. We consider the latter by filtering exceedances at a single location using the conditional spatial extremes model to model left-censored zero-inflated precipitation data. We use Gaussian Markov random field residual processes to ensure the extreme model can be fitted to a large number of measurement sites and consider estimation of both the margins and the dependence structure parameters under the Bayesian paradigm, using MCMC methods through data augmentation for estimation. Using the joint model, we compare extremal properties of rainfall fields of statistically downscaled GCM output from the Pacific Climate Impact Consortium with measurements from station data in British Columbia and Washington, finding important discrepancies between the two.