Streambed drainage was measured at the cessation of 26 ephemeral streamflow events in Rillito Creek, Tucson, Arizona from August 2000 to June 2002 using buried time domain reflectometry (TDR) probes. An unusual drainage response was identified, which was characterized by sharp drainage from saturation to near field capacity at each depth with an increased delay between depths. We simulated the drainage response using a variably saturated numerical flow model representing a two-layer system with a high permeability layer overlying a lower permeability layer. Both the observed data and the numerical simulation show a strong correlation between the drainage velocity and the temperature of the stream water. A linear combination of temperature and the no-flow period preceding flow explained about 90% of the measured variations in drainage velocity. Evaluation of this correlative relationship with the one-dimensional numerical flow model showed that the observed temperature fluctuations could not reproduce the magnitude of variation in the observed drainage velocity. Instead, the model results indicated that flow duration exerts the most control on drainage velocity, with the drainage velocity decreasing nonlinearly with increasing flow duration. These findings suggest flow duration is a primary control of water availability for plant uptake in near surface sediments of an ephemeral stream, an important finding for estimating the ecological risk of natural or engineered changes to streamflow patterns. Correlative analyses of soil moisture data, although easy and widely used, can result in erroneous conclusions of hydrologic cause—effect relationships, and demonstrating the need for joint physically-based numerical modeling and data synthesis for hypothesis testing to support quantitative risk analysis.