Identification of enhanced rainfall rates using the near-storm environment for radar precipitation estimates

Reliable and timely flash flood warnings are critically dependent on the accuracy of real-time rainfall estimates. Precipitation is not only the most vital input for basin-scale accumulation algorithms such as the Flash Flood Monitoring and Prediction (FFMP) program used operationally by the U.S. National Weather Service, but it is the primary forcing for hydrologic models at all scales. Quantitative precipitation estimates (QPE) from radar are widely used for such a purpose due to their high spatial and temporal resolution compared to rain gauges and satellite-based algorithms. However, converting the native radar variables into an instantaneous rain rate is fraught with uncertainties.;One of those uncertainties is the varying relationship of radar observables to rain rate for different regions and storm types due to variations in drop size distributions. Many unique reflectivity-to-rain rate (Z-R) functions have been proposed in the literature over the past 70 years for single-polarization radars, and it is becoming apparent that various rain rate functions will also be needed in different environments for dual-polarization radars as well. The challenge then becomes identifying the environments in real-time such that the appropriate rain rate function can be applied. This study addresses the challenge of identifying environments conducive for tropical rain rates, or rain rates that are enhanced by highly productive warm rain processes. Rain rates in tropical environments tend to be underestimated by other operational Z-R functions and have often been associated with historic flash flooding events, so delineating them in real-time can greatly improve not only the radar-based QPE accuracy, but the level of certainty by forecasters for issuing flash flood warnings as well.;Six consecutive months of hourly data from the 2010 warm season were used to train ensembles of statistical classification models such that probabilities of warm rain enhancement of rain rate can be derived. The predictors for the ensemble were retrieved from the 20-km Rapid Update Cycle (RUC) model analyses and were chosen to provide a general description of the thermodynamic environment from the which the rainfall developed. Those environmental predictors were trained against two different predictands: bias of rain rates for the convective Z-R function vs. collocated, quality controlled rain gauges, and the vertical gradient of radar reflectivity between the freezing level and the lowest elevation observed by the radar. The resulting probabilities from the trained ensembles were then used to delineate where tropical rain rates would be assigned in a gridded QPE product, and the resulting hourly accumulations were verified against independent rain gauges.;Overall, the probability-based precipitation type delineation scheme improved hourly rainfall accumulations for three independent cases tested when compared to both the legacy rainfall product from the National Mosaic and Multisensor Quantitative Precipitation Estimation (NMQ) project and the operational NWS rainfall product (Stage II), but neither the gauge-based nor VPR-based ensembles emerged as a clearly superior predictor than the other for all cases tested. However, spatial similarities between the two probability fields and similar results from variable importance analysis suggest that both methods are attempting to delineate the same environment. This implies that the systematic underestimation of radar-based QPE and the enhancement of reflectivity in the warm layer from warm rain hydrometeor growth are related or at the very least are associated with the same type of environment. Initial analysis of polarimetric variables, particularly differential reflectivity, in areas of high and low probabilities also support a connection between rain rate underestimation and tropical airmasses.