Technical Abstract: Urban growth in the southwestern U.S. influences strongly storm water runoff by creating impervious surfaces. Hydrologic models used to compute runoff from watersheds require an estimate of the area and location of pervious surfaces as input data. In semi-arid regions, this information is provided currently by delineating these areas manually, which is labor-intensive, costly and potentially inaccurate. The goal of this study was to use high-resolution Quickbird imagery to determine pervious and impervious areas by using a semi-automated approach that is capable of processing much bigger areas with repeatable quality. The study site was a small, recently developed watershed in Sierra Vista, Arizona, characterized by a gated community with xeriscape landscaping. Commonly used pervious area classifications were unsuccessful in this watershed because they relied heavily on vegetation as indicator for perviousness. We chose instead an object-based classifier to distinguish between gravel mulch in backyards and asphalt on the streets. The KINEROS 2 hydrologic model was used to compute runoff values based on the hand-digitized versus semi-automated Quickbird estimates of perviousness. Modeled runoff values were compared to each other and to runoff measured with a flume at the watershed outlet. Results show the value of remote sensing data, processed with semi-automated methods, to produce input layers for hydrologic models. This method enables hydrologic modeling of larger areas with reduced effort and known accuracy.