Quantitative precipitation forecasts from mesoscale models and estimates of current precipitation rate from weather radar and satellite data can provide the rainfall-rate input to models of surface runoff and river discharge. If the precipitation forecasts and estimates have resolution on the convective scale, and are reasonably accurate for areas of complex terrain, the coupled atmospheric/ surface-hydrologic system can be used for flash flood prediction, where flash floods are generally defined as floods for which the water rises from a normal level to flood level in less than six hours. Within 5 years, NWS/NCEP will likely have the computational resources to run operational models that are capable of thunderstorm-scale prediction. A number of problems first need to be addressed in order for this capability to lead to improved flash-flood prediction in complex terrain. The goals of this research deal with a subset of these problems and are being addressed by NCAR scientists T. Warner, E. Brandes, R. Bruintjes, F. Chen, C. Davis, K. Manning, J. Sun, D. Yates, G. Leavesley (USGS-Denver), and D. Matthews (BuRec-Denver). .
In order to address the above need, the following general questions require attention.
The development of improved radar precipitation estimation algorithms are based on rainfall events in two geographic areas: 1) the Walnut Creek watershed in the CASES study area which is very well instrumented with rain and stream gages and was the venue for a study in the summer of 1997 of the accuracy of radar-based estimates of precipitation and 2) the Front Range of the Rockies which contains the Front Range Alert Network, which is also well instrumented with rain and stream gages for flash-flood protection, and which was included in a 1996 study of radar estimation of rainfall. The CASES-area study will allow the development of algorithms that work well in noncomplex terrain, while the Front Range study will extrapolate those results to complex terrain.
The MM5-based modeling studies will exclusively address problems of storm-scale precipitation estimation in complex terrain, and thus will be limited to watersheds in the Front Range. The first storm being studied is associated with a flash flood that occurred on the Buffalo Creek watershed in the summer of 1996. Two types of simulations are being performed. One uses a regional domain that has a resolution that does not well-resolve storm scale dynamics, but is likely similar to the NCEP continental-U.S. area operational mesoscale model that will be employed early in the next decade. Forecasts from such models need to be capable of predicting the convective environment well, so that the storm-resolving models can be applied where convection is probable. The highest resolution computational mesh of the storm-scale model has a grid increment of 2 km. Variational retrieval of the radar data is being used to dynamically initialize this storm-scale simulation. Initial tests show that the regional model defines the environment of the storm well, with reasonable estimates being produced of the precipitation rate, however the location of the precipitation is not well predicted, as expected because no radar data were utilized in the initialization. Tests are underway with the storm-resolving model that is variationally initialized with the radar data, to determine whether the location, duration and intensity of the observed precipitation can be replicated.
The surface runoff/streamflow model that is to be used with the radar and model-based estimates of the precipitation rates is one developed by the U.S. Geological Survey (PRMS - Precipitation Runoff Modeling System). Work is underway to format the radar-based precipitation estimates for use as input to the PRMS, so that predicted stream discharge can be compared with that measured.