H21E-0802
Modeling Streamflow Using Gauge-Only Versus Radar-derived Rainfall
Rainfall in Florida is very dynamic in nature and is the greatest determining factor in hydrologic modeling studies. These studies traditionally have used gauge-only rainfall estimates despite their limitations and the advances in multi-sensor precipitation estimates. The convenience and familiarity with gauge data are a major factor leading to their continued use; however, there also have been questions about the statistical consistency and quality of radar-derived precipitation data. We previously reported on an intercomparison between gauge-only Thiessen polygon data with the gridded 4 × 4 km Florida State University (FSU) version of the National Weather Service (NWS) Multi-sensor Precipitation Estimator (MPE) scheme over several Florida basins. We showed that gauge-density within a basin is highly correlated with the magnitude of rainfall differences between the two datasets. The study also showed that seasonal characteristics of rainfall are an important factor leading to differences. The current paper evaluates the impacts of these input differences on streamflow by using a specialized, fully-distributed hydrologic model—the Watershed Assessment Model (WAM). Although WAM can model various water quality parameters, we focus on the streamflow produced by the different rainfall inputs. By describing differences in streamflow, we provide results that modelers can easily relate to--the impact of higher-resolution MPE rainfall data on their model's bottom-line. We have modeled the Suwannee River basin in North Florida between 1996 and 2005. Hourly rain gauge data used as input to the FSU MPE scheme were obtained from the National Climatic Data Center (NCDC) and the Suwannee River Water Management District (SRWMD). This combination provides the most reliably-dense gauge network possible. All of the rain gauge data were quality-controlled by FSU. Quality-controlled radar data were obtained from the NWS's Southeast River Forecast Center (SERFC). The FSU 4 × 4 km MPE dataset was developed through years of collaboration between Florida State University, the Florida Department of Environmental Protection (FDEP), and the National Weather Service. Its characteristics have been described in our previous publications. This paper will compare simulated flows using the gridded FSU MPE rainfall inputs with simulated flows using gauge-only Thiessen polygon rainfall inputs. Both versions of simulated flow also are compared to measured streamflow observations. The Suwannee River basin contains abundant streamflow data, allowing comparisons to be made at various points within the basin. This allows us to determine how streamflow at different locations in the basin responds to rainfall scenarios that vary from intense and isolated to widespread and light. Rain gauge density in the SRWMD has progressively increased by more than five times during the past 10 years. Our paper also will describe the impact of gauge density on the quality of simulated streamflow. Finally, the paper will describe oossible thresholds of gauge density, as well as the pros and cons encountered in modeling with the two distinctly different rainfall inputs.
H21E-0803
Sensitivity Studies of the Radar-Rainfall Error Models
It is well acknowledged that there are large uncertainties associated with the operational quantitative precipitation estimates produced by the U.S. national network of WSR-88D radars. These errors are due to the measurement principles, parameter estimation, and not fully understood physical processes. Comprehensive quantitative evaluation of these uncertainties is still at an early stage. The authors proposed an empirically-based model in which the relation between true rainfall (RA) and radar-rainfall (RR) could be described as the product of a deterministic distortion function and a random component. However, how different values of the parameters in the radar-rainfall algorithms used to create these products impact the model results still remains an open question. In this study, the authors investigate the effects of different exponents in the Z-R relation (Marshall- Palmer, NEXRAD, and tropical) and of an anomalous propagation (AP) removal algorithm. Additionally, they generalize the model to describe the radar-rainfall uncertainties in the additive form. This approach is fully empirically based and rain gauge estimates are considered as an approximation of the true rainfall. The proposed results are based on a large sample (six years) of data from the Oklahoma City radar (KTLX) and processed through the Hydro-NEXRAD software system. The radar data are complemented with the corresponding rain gauge observations from the Oklahoma Mesonet, and the Agricultural Research Service Micronet.
H21E-0804
A Validation Study of the NWS/MPE Precipitation Products Using a Dense Rain Gauge Network in South Louisiana
This study focuses on validation of the multi-sensor precipitation products developed by the operational multi- sensor precipitation estimation (MPE) algorithm of the National Weather Service (NWS) River Forecast Centers (RFC). MPE data are acquired through the Stage IV archives at the National Center for Environmental Prediction (NCEP). MPE data, which is based upon the merging of data from WSR-88D radar, surface rain gauge, and occasionally geo-stationary satellite data, is provided at hourly temporal resolution and over a national Hydrologic Rainfall Analysis Project (HRAP) grid which has a nominal size of 4 square kilometers. Operational hydrologic forecasting applications now require higher spatial and temporal resolution which is provided by radar data. To help determine the validity of radar data in south Louisiana, a study was performed on MPE data for a three-year period (2004-2006) using 13 independently operated rain gauges located within an area of ~30 km2. The close proximity of gauge sites to each other allows for multiple gauges to be located within the same HRAP pixel. As a result, two pixels contained four gauges each, and one pixel contained two gauges. This co-location of multiple gauges within an HRAP pixel allows for a reasonably accurate estimation of the MPE errors over different scales such as hourly, daily, and monthly temporal resolution. In this context, the errors are defined as the deviation of MPE estimates from the corresponding average of gauge measurements in each pixel. The self dependence of these errors is assessed by analyzing their temporal and spatial auto-correlations. The MPE products are mainly intended for operational hydrologic forecasting. Therefore, the study will examine the impact of MPE uncertainties on runoff simulations in a mid-size experimental watershed in south Louisiana. The physically- based hydrologic model (Gridded Surface Subsurface Hydrologic Analysis, GSSHA) is driven by two sets of rainfall forcing: MPE products and data from a dense rain gauge network over the watershed. Differences in the simulated hydrographs are assessed in terms of prediction accuracy of runoff peaks and volumes. Investigating the need for validation of multi-sensor estimates can facilitate the improvement of radar-gauge merging algorithms and further enhance the accuracy of operational hydrologic forecasting.
H21E-0805
Enhanced NEXRAD Radar-based Flood Warning System with Hydraulic Prediction Feature: Floodplain Map Library (FPML)
Houston is facing flood problems of a serious nature. Until more permanent solutions are found accurate and timely, early warning flood systems are vitally needed to provide the early warnings that public and private entities are demanding. The current Rice University/TMC Flood Alert System (FAS2) began to utilize higher-resolutioned Level II NEXRAD radar data (1 x 1 km) that is calibrated against local rain gauges by the end of 2004, with the real-time hydrologic model (RTHEC-1) to provide important data for predicting flood levels along Brays Bayou. The finer resolution of Level II radar rainfall data provides significantly greater details with respect to the spatial variability of rainfall. FAS2 has been tested for more than 30 events including three recent events in 2006 season with excellent performance. It has been found from 2006 season that the average difference in peak flows is 8.76%; the average difference in terms of volumes is 13.70%. The floodplain map library (FPML) as a new hydraulic prediction tool has been developed based on the radar- based FAS2 and is being integrated into FAS2 to provide inundations maps in near real time. The development of FPML includes three stages: designing rainfall based on historical rainfall data over the watershed, delineating 99 maps based on design rainfalls, and designing algorithm to link real-time NEXRAD radar rainfall to appropriate maps. The enhance system can be a prototype for other flood-prone areas along the Gulf coast, and will improve emergency personnel's ability to initiate evacuation strategies at many levels.
H21E-0806
Radar Rainfall Estimates for Extreme Flood Events
Analyses of radar rainfall estimates from the WSR-88D radar network are presented for a sample of major flood events in the US. The framework for radar rainfall analysis is the HydroNEXRAD system and analyses highlight the utility of radar rainfall estimates for a diverse range of flood hydrology applications. The sample of flood events focuses on regions of the US with large flood potential, including the Edwards Plateau of Texas, the central Appalachians and urban watersheds of the eastern US. The basin scales of interest range from less than 10 sq. km. to more than 10,000 sq. km. We examine errors in radar rainfall estimates from the perspective of extreme flood-producing storms. Analyses for the Edwards Plateau of Texas focus on major storm events in June 1997 and July 2002. In the Delaware River basin, analyses center on a sequence of record and near-record flood events in September 2003, April 2004 and June 2005.
H21E-0807
On the Remapping of Radar Estimates onto Cartesian Coordinates
A precise method for remapping of radar estimates onto Cartesian grids of different sizes was introduced. The method is straightforward and the algorithm developed for this study can be used to perform averaging over grids of any size or shape. Comparison of rainfall estimates using this method and estimates computed using a simple averaging method typically used by hydrologists reveals that there can be significant differences. Differences are expected to be much larger if precise remapping is compared to simpler methods such as the nearest neighbor. The method is particularly useful if products from more that one radar are to be merged. Using this method it is not necessary to interpolate estimates from two radars over Cartesian grid before merging is performed. In addition, gauge-radar comparisons and bias computation can be more consistent when precise remapping is used. Another advantage of precise remapping is that it will be very easy to account for the variability of the power within the radar bin using this scheme because the contribution from each radar bin is defined geometrically in a precise way. Grids constructed this way are the closest representation of radar observation of the atmosphere in Cartesian coordinates. The same approach was used to remap rainfall estimates from one radar onto the bin of another radar of lower resolution. Precise remapping was used in comparing estimates of the higher-resolution NCAR's S-band radar (S-Pol) and a WSR-88D radar. Detailed comparison of fields of instantaneous rain rate reveals that the correlation between the estimates of the two systems varies spatially. The correlation is highest at the mid-point between the two radars and decreases with distance from this point.