Flooding is a major concern in many urbanized areas due to increased runoff generated by impervious areas and the close proximity of homes and businesses to flood prone areas. A major goal of hydrologic modeling is to predict when and where flooding will occur. This research presents hydrologic modeling results for Brays Bayou, a highly urbanized 240km2 watershed in Houston, TX, as well as a 10km2 subwatershed of Brays called Harris Gully. Simulations for Brays were conducted at a 120m resolution, while simulations for Harris Gully were run at a 40m resolution. All simulations were run using a fully distributed physics-based hydrologic model called Vflo™. Vflo™ utilizes geographic information and multi-sensor precipitation input to simulate rainfall runoff from catchments. Watershed response to precipitation is determined by the spatial and temporal distribution of rainfall, soils, land cover, and the hydraulics of detention and stormwater conveyance. Understanding the limits to prediction accuracy concerns both the rainfall input and the hydrologic and hydraulic characteristics of an urban or rural catchment.

Prediction accuracy for Brays Bayou was assessed through rainfall event reconstruction using archived Level II NEXRAD radar data along with rain and stream gauge data. Operationally, the radar rainfall is bias corrected and used for real-time flood forecasting in the urbanized Brays Bayou. Ten rainfall events spanning a 4 year period were chosen for simulation and analyzed at Main St. in Brays. Seven of the ten simulated events showed good accuracy in peak discharge and peak timing with an average error of 8% and 1%, respectively. Three of the ten events were considerably over predicted by the model. For two of the over-predicted events, analysis of radar and gauge input revealed that rainfall was significantly over-estimated by radar compared to rain gauge. The remaining over-simulated event showed little discrepancy between radar and gauge rainfall totals. It is believed that the over-simulation may have occurred due to recently constructed detention ponds. These detention effects are evident in the observed hydrograph indicated by the slow recession limb and the ‘shaved' peak, which were not present in other events. The storm total map also provides evidence that detention may have played a significant role, as much of the rainfall occurred upstream of the detention ponds. As a result, these three events were excluded from the above statistics for peak and peak timing.

Prediction accuracy for Harris Gully was assessed in much the same manner as Brays Bayou. Eleven events with an observed stage of 2.3 to 6.7ft (0.7-2.0m) were chosen for simulation. High flow events were not considered for analysis due to known backwater effects as Harris Gully discharges into Brays. Results showed that accurate simulations were obtainable for most events. Eight of the eleven events were simulated within 1.3ft (0.4m) of the observed peak and within 10min of the observed peak time. Of the three events not accurately simulated, two showed an over prediction, while the other showed an under prediction in peak stage. Storm totals from the three events showed reasonable agreement between radar estimates and rain gauge. However, little can be said about the spatial accuracy of the radar estimates as there is only one gauge in the watershed available for comparison. Thus, it is unknown whether this point location was truly representative of the spatial accuracy of radar estimates for other portions of the basin. Even though the watershed is relatively small (10km2), the spatial accuracy of radar is important as many of the events showed considerable variability in the distribution of rainfall.

Simulation accuracy for both watersheds was improved when a more detailed hydraulic network was included in the model. Channel cells comprised 9% of the grid cells in the Brays Bayou model and 16% in the Harris Gully model, as opposed to 3.6% and 1.9% in previous versions, respectively. Results showed a better model was obtained when using a model with a more detailed channel network representing the hydraulics of the stormwater conveyance infrastructure. The largest improvement in predictability was found in the rising limb and peak of the predicted hydrograph compared to the observed discharge. The dense network of channel cells more accurately represented the actual conveyance of stormwater in the watersheds and resulted in more hydraulically accurate models.