H23K-01 INVITED
Hydro-NEXRAD: A Community Resource for Future Research on Improving Rainfall-Rainfall Estimation and Hydrologic Applications
Hydro-NEXRAD is a software system with web-based user interface for obtaining historical customized NEXRAD- based radar-rainfall maps (products) from some 40 WSR-88D radars covering mainly the central and eastern U.S. These products have increased spatial and temporal resolution in comparison to the operational products available from the National Weather Service. Hydrologists can request customized products by selecting various algorithmic modules and parameter values and projected on a grid of choice. The output is formatted for ingest by geographic information systems and mapping software. The authors discuss the system architecture, the database extent and the possibilities of including additional algorithms in the future versions. They illustrate the utility of the software with several applications where side-by-side comparisons of various products allow studies of uncertainty propagation and sensitivity analysis. One of the comparisons presented involves the NWS products obtained with the Precipitation Processing System. Since one of the features of Hydro-NEXRAD is repeatability of the results, the system promotes systematic studies of new algorithms, comparisons with rain gauge data and error modeling, and uncertainty propagation in hydrologic applications. The authors also discuss future extensions of the system including a real-time version being developed in collaboration with Unidata of UCAR.
H23K-02 INVITED
Use of Historical Radar Rainfall Estimates to Develop Design Storms in Los Angeles.
A database of 15-minute historical gage adjusted radar-rainfall estimates was used to evaluate the geometric properties of storms in the City of Los Angeles, CA. The database includes selected months containing significant rainfall during the period 1996-2007. For each time step, areas of contiguous rainfall were identified as individual storm cells. An idealized ellipse was fit to each storm cell and the properties of the ellipse (e.g., size, shape, orientation, velocity and other parameters) were recorded. To accurately account for the range of storm cell sizes, capture a large number of storm cells in a climatologically similar area, assess the variability of storm movement, and minimize the impact of edge effects (i.e., incomplete coverage of cells entering and leaving), a study area substantially larger than the City of Los Angeles was used. The study area extends from city center to 30 miles north to the crest of San Gabriel Mountains, 45 miles east to Ontario, 60 miles south to Santa Catalina Island, and 70 miles west to Oxnard, an area of about10,000 square miles. Radar data for this area over 30 months in the study yields many thousands of storm cells for analysis. Storms were separated into classes by origin, direction and speed of movement. Preliminary investigations considers three types: Arctic origin (west-northwest), Pacific origin (southwest) and Tropical origin (south or stationary). Radar data (for 1996-2007) and upper air maps (1948-2006) are used to identify the direction and speed of significant precipitation events. Typical duration and temporal patterns of Los Angeles historical storms were described by season and storm type. Time of maximum intensity loading variation were determined for a selection of historic storms Depth-Areal Reduction Factors (DARF) for cloudbursts were developedfrom the radar data. These data curves are fit to equations showing the relationships between DARF, area and central intensity. Separate DARF curves are developed for 6X (6 events per year), 4X, 3X, 2X, 1, 2, 5 and 10 year recurrence, and durations from 5 minutes to 7-days. A comparison is made between DARF derived in these analyses with NOAA Atlas 12 DARF, the USACE Sierra Madre Storm and other DARF developed for the interior Southwest. Orographic increases in DDF are related to the Los Angeles County Flood Control District Hydrology Manual 24-hr 50-yr Precipitation maps, elevation from USGS topographic maps and Mean Annual Precipitation maps.
H23K-03 INVITED
Evaluation of Hydrologic Prediction Accuracy Using Gauge-Adjusted Radar Precipitation Input
Operational hydrologic prediction in real-time at the event scale depends on having accurate representation of rainfall over watershed areas. A major limitation to predictions in gauged and ungauged basins is the lack of precipitation observations that are accurate or representative. Radar rainfall that has been enhanced for accuracy with rain gauge observations represents a significant advance in hydrologic prediction. This study is motivated by research in operational hydrology where predictions are needed at stream locations in headwater basins and river basins in rural and urban settings. Setup of the physics-based distributed hydrologic model, Vflo, is accomplished using geospatial data to derive readily obtainable physical parameters and used with input derived from NEXRAD radar and a rain gauge network. Evaluation of the accuracy in streamflow predictions is accomplished with several types of rainfall products as input derived from radar and gauge data. While gauge- adjusted radar and gauge-only products agree well after quality enhancement procedures are applied, the hydrologic predictions derived from the precipitation products reveal that improved accuracy is obtained by enhancing radar accuracy through bias correction and quality control procedures. Hydrologic prediction accuracy is known to be affected by both the forcing products derived from radar/gauge observations and from model uncertainty. The achievable accuracy for various radar input datasets derived from adjustment of continuous radar and gauge data is examined in this presentation to gain an understanding of inherent uncertainties associated with data quality, sampling error and bias correction procedures applied to radar. Through the framework of the Distributed Modeling Intercomparison Project (DMIP2), hydrologic prediction accuracy obtained through comparison of the hydrograph volume and peak discharge produced from two radar products and a gauge-only product over an extended period reveals prediction uncertainty associated with gauge network density. Uncertainty in rainfall derived from a multi-radar mosaic is compared with hydrologic prediction uncertainty. Distributed hydrologic simulation accuracy for a ten year period of hourly rainfall is presented for the 1200 km2 Blue River to test the uncertainty and accuracy of streamflow produced by a physics-based model in relation to the rainfall input product.
H23K-04 INVITED
Comparing Distribution Functions of Rain Rates from Gauge, NEXRAD, and TRMM Radar Observations
The distribution of rain rate is of great interest in many fields. For example, hydrological applications such as flood forecasting depend on an accurate representation of the rainfall--driven by rain rate--that does not infiltrate the soil. However, efforts to evaluate quantitative instantaneous rain rate estimates and rain rate forecasts, as opposed to rainfall amounts, are surprisingly rare. In this presentation we will compare probability distribution functions (pdfs) of radar rain rate estimates derived using different gauge adjustment schemes. Then we will compare the gauge adjusted ground radar (NEXRAD) pdfs with those of the TRMM spaceborne radar observations. The ground reference rain rate products used to generate the pdfs are from the NASA TRMM ground validation site in central Florida and from the new NOAA/NSSL experimental radar products (Q2) of high- resolution (1 km, 5-min) instantaneous rain rate mosaic available over the entire continental U.S.
H23K-05
Using Independent NCDC Rain Gauges to Analyze Precipitation Values from the OneRain Corporation Algorithm and the National Weather Service Procedure
Two widely used procedures for optimally combining radar- and gauge-derived rainfall are those of the OneRain Corporation and the National Weather Service (NWS). The NWS procedure, called the Multi-sensor Precipitation Estimator (MPE), produces an hourly product on the 4×4 km Hydrologic Rainfall Analysis Project (HRAP) grid. MPE is used operationally by local NWS offices and NWS River Forecast Centers (RFCs). Florida State University (FSU) has employed the MPE scheme with NWS hourly digital precipitation arrays (DPAs) to create an hourly historical precipitation database for the Florida Department of Environmental Protection (FDEP) for the period 1996-2006. The OneRain procedure is proprietary and has not been described well in the literature. However, it produces a product at 15 min intervals on a 2×2 km Cartesian grid. Florida's Water Management Districts as well as other government agencies and private firms use the OneRain product. Although their methodologies and their temporal and spatial resolutions differ, each dataset is being used to make water management and regulatory decisions. Thus, it is useful to evaluate the two procedures against independent data. This paper will evaluate the two procedures against daily co-op gauges from the National Climatic Data Center (NCDC) that were not used in creating either the MPE or OneRain products. The area of the South Florida Water Management District is investigated during the 2004-2005 calendar years. Both radar-derived products are summed over 24 h periods based on the daily recording time of each gauge. The OneRain precipitation values then are placed onto the same 4×4 km HRAP grid containing the MPE data. Finally, the 4×4 km MPE and OneRain values are compared with any NCDC gauges located within the HRAP grid cells. Results of daily precipitation comparisons will be presented for all gauges combined over the two year time period, over individual years, the cold and warm seasons, and over individual months. Individual gauge sites also will be evaluated. Intervals of precipitation are analyzed to see how each scheme handles light, moderate, and heavy rainfall events. Finally, case studies describe how each scheme estimates particular rainfall events, including land-falling tropical cyclones. In summary, this paper will describe which procedure compares best with the NCDC independent gauges, and whether the OneRain and MPE products can be used interchangeably.
H23K-06
Autocorrelation and Error Structure of Rainfall Derived from NEXRAD in Central and South Florida
Motivation for this study comes from the South Florida Water Management District (District) who is responsible for managing water resources in 16-counties over a 46,439-square kilometer (17,930 square-mile) area. Near-real- time rainfall data are used in operation of approximately 3,000 kilometers (~1,800 miles) of canals, 22 major pump stations and 200 water control structures. The spatial extent of the District extends from Orlando to Key West and from the Gulf Coast to the Atlantic Ocean and contains major water features including Lake Okeechobee and the Everglades wetlands. Rainfall is a key factor in the water management decisions made by the District in real-time and through studies that rely on archival rainfall data derived from radar and rain gauge observations. Rainfall measurements are obtained from a combination of four NEXRAD radars and a rain gauge network that comprises 280 active rain gauge stations located in the more populated areas. Four NEXRAD (Next Generation Weather Radar) sites operated by the National Weather Service cover the region. Rain gauges are used for frequency analysis and for adjustment of the radar rainfall products. An optimization study of the rain gauge network is accomplished by removing gauges in areas of excess coverage, and by adding or moving rain gauges to gain a more even spatial distribution over the District. Rainfall fields measured at daily and hourly timesteps exhibit autocorrelation which can affect the network design subject to optimality constraints. This presentation will describe the autocorrelation and error structure found in rainfall measurements derived from rain gage and NEXRAD data. The data used in the analysis includes rain gage data and the NEXRAD rainfall data that was collected during 1995-2005 at 2 x 2 km resolution. A set of clusters of rain gages and a regular array of analysis blocks that were 20 x 20 km in size for the NEXRAD data were used to account for variability of the rainfall processes and local rainfall patterns. The spatial autocorrelations of the rain gage and NEXRAD rainfall were identified using a semivariogram approach at daily timescale. The model fitting to the semivariograms were performed on data from 1998-2005. The spatial autocorrelations from rain gage and NEXRAD rainfall data sets were compared and evaluated.
H23K-07
NEXRAD Reflectivity Derived Rainfall Estimates Comparison With Measurement from Co- located Rain Gauges Newfel Mazari, Hongjie Xie, Hatim Sharif, Jon Zetler
Rain gauges were the main means to measure rainfall until the introduction of radar technology in hydrology and meteorological studies. Radar can cover areas inaccessible with conventional rain gauges networks. The application of NEXRAD products in the estimation of quantity and spatial distribution of precipitation has recently improved with the offering of Multi-sensor Precipitation Estimator (MPE) products. However the accuracy of radar precipitation estimates remains mostly unclear. We used a set of two tipping bucket rain gauges for comparison with radar rainfall estimates derived from level II reflectivity. A power law Z = 300 R1.4 was used in this study. We used non-zero rainfall events pairs (radar, gauge), with a threshold of 0.31 mm corresponding to a reflectivity of 15 dBZ. The results indicated that the best time delay between radar reflectivity measurement and the rain gauges measurement of rainfall results was 7 minutes. The time delay is due to the fact that the radar observes rainfall aloft (in our case around 1,200 meters), while the rain gage records rainfall only at ground level. Data were collected from September 2006 to March 2007. Statistical analysis resulted in an R-square of 0.85 for one rain gauge and 0.80 for the other, with a P-value of 0.0001. Furthermore, the radar underestimated rain rates by 23 % and overestimated rain rate by 11% for rain gauge one and two, respectively. Our analysis indicates that the difference between the rain gauges is mostly due to hardware inconsistencies.
H23K-08
Characterizing Rain Gage – Radar (NEXRAD) Data Relationships Using Inductive Modeling
The use of radar (NEXRAD) estimated rainfall data for providing information about the extreme rainfall amounts resulting from storms, hurricanes and tropical depressions is common today. Often corrections are applied to the RADAR-based rainfall data-based on what was actually measured on the ground by rain gages. Understanding and modeling the relationships between RADAR and rain gage data are essential tasks to confirm the accuracy and reliability of the former surrogate method of rainfall measurement. Conventional regression models in many situations are found to be incapable of capturing these highly variant non-linear spatial and temporal relationships. This study aims to understand and model the relationships between RADAR (NEXRAD) estimated rainfall data and the data measured by conventional rain gages. This study proposes to investigate the use of emerging computational inductive modeling techniques and to develop optimal functional approximation methods for this purpose. The raw and transformed RADAR rainfall data and rain gage data will also be analyzed to understand spatio-temporal associations. The study areas selected from upper and lower Kissimmee basins of south Florida form the test-bed for the proposed approaches and ensure the testing of the validity and operational applicability of these approaches.