The statistical characterization of precipitation across multiple scales in urban regions is crucial  to support flood prediction and management, rain harvesting, irrigation of parks and green areas,  and infrastructure design. In this talk, I will present results of an extensive analysis of the  precipitation statistical properties in the metropolitan area of Phoenix, Arizona. In this desert  region of southwestern U.S., the precipitation regime has two seasons: summer (July September), which is dominated by isolated short-duration monsoonal thunderstorms, and winter  (November-March), during which cold fronts lead to widespread storms lasting for a few days. A  network of more than 300 high-resolution rain gages, with a density of 1 gage every 95 km2 over  an area of 29,600 km2, is used for the analyses. I will first show how orography controls seasonal  and annual precipitation, as well as the timing of the diurnal cycle of summer storms. I will then  present evidence that the statistical distribution of daily precipitation extremes is different in the  two seasons and that a mixed model of peak-over-threshold series can be used to simulate the  frequency of annual extremes, which are used for infrastructure design. Finally, I will show how  a state-of-the-art stochastic model based on multivariate autoregressive model can be used to  effectively generate high-resolution space-time precipitation fields that preserves intermittency  and spatiotemporal correlation structure of summer and winter storms.