Statewide real-time quantitative precipitation estimation using weather radar and NWP model analysis: Algorithm description and product evaluation

Highlights

• An automated QPE system generates a statewide real-time rainfall product for flood forecasting in Iowa.

• The system processes NWP and radar data using algorithms accounting for precipitation microphysics and QPE uncertainties.

• The system generates a rainfall map covering the entire state of Iowa at a resolution of 0.5 km, updated every 5 min.

• The evaluation demonstrates that a new algorithm implementation significantly improves the rainfall estimation accuracy.

Abstract

This study describes an automated system that generates a statewide real-time quantitative precipitation estimation (QPE) product for flood forecasting in Iowa. The QPE system comprises, real-time data acquisition, processing, and product visualization subsystems. Combined with information retrieved from numerical weather prediction, the system processes data from multiple radars using various algorithms accounting for precipitation microphysics and radar remote sensing uncertainties. The system generates a composite rainfall map covering the entire state of Iowa at a resolution of 0.5 km, updated every 5 min. With the help of the system's flexible modular configuration, we have recently added a new polarimetric algorithm based on specific attenuation. Independent evaluations based on comparisons with rain gauge data and hydrologic model prediction of streamflow demonstrate that the new implementation significantly improves the rainfall estimation accuracy. The new QPE product shows performance comparable to the Multi-Radar Multi-Sensor product that contains a rain gauge correction.

Introduction

Using data from the U.S. Weather Surveillance Radar-1988 Doppler (WSR-88D) network, the Iowa Flood Center (IFC) has provided a statewide real-time rainfall product since the IFC's establishment in 2009. This product generation was motivated by the need for real-time flood prediction in Iowa, which has repeatedly experienced devastating floods at various scales in recent decades (e.g., Smith et al., 2013; Vennapusa and White, 2015; Seo et al., 2018). Our goal to forecast “everywhere” and cover all Iowa communities (e.g., regardless of catchment scale) has led to the use of a distributed hydrologic model, which requires spatially variable rainfall inputs (Krajewski et al., 2017).

The IFC quantitative precipitation estimation (QPE) framework was initially built on the real-time Hydro-NEXRAD application (Krajewski et al., 2011a, 2013; Kruger et al., 2011; Seo et al., 2011). Most scientific algorithms in our QPE system have evolved according to the WSR-88D's hardware and polarimetric upgrades (e.g., Istok et al., 2009). The IFC QPE system acquires real-time data from seven WSR-88Ds (KARX in La Crosse, Wisconsin; KDMX in Des Moines, Iowa; KDVN in Davenport, Iowa; KEAX in Kansas City, Missouri; KFSD in Sioux Falls, South Dakota; KMPX in Minneapolis, Minnesota; and KOAX in Omaha, Nebraska), as shown in Fig. 1. The system then generates a composite rain rate map covering the entire domain (Fig. 1), with temporal and spatial resolutions of 5 min and 0.5 km, respectively, using a variety of processing algorithms, as documented in Seo et al. (2011, 2015) and Seo and Krajewski (2015).

As of early 2019, we had added a state-of-the-art polarimetric algorithm known as “specific attenuation” (e.g., Ryzhkov et al., 2014; Wang et al., 2019) to our QPE procedures to fully benefit from the WSR-88D's dual-polarization (DP) capability. This new algorithm required several new elements; for example, one that retrieves temperature soundings from the numerical weather prediction (NWP) model analyses to identify the melting layer (ML) location. Before this new implementation, the use of DP in the IFC system was limited to basic data quality control (e.g., removal of non-meteorological returns), and the system's main estimator was a single polarization–based algorithm using a reflectivity-rain rate (Z-R) relation. The new implementation using the specific attenuation method promises to be the most significant milestone in our system's 10-year history, as the method has demonstrated meaningful improvements in QPE accuracy (Seo et al., 2020b).

Therefore, we take this opportunity to document the architecture and capabilities of our fully automated QPE system, including algorithm updates and new developments as well as the way it complements the outdated descriptions presented in Seo et al. (2011). To validate the attainable improvement in QPE and subsequent hydrologic prediction, we generated the statewide QPE products using the latest and prior algorithms for a three-year period (2016–2018) and evaluate the performance of each one using rain and stream gauge observations. We also compared the performance of our QPE products with that of U.S. national QPE products (e.g., Zhang et al., 2016; Cunha et al., 2013) that have been widely used for meteorological and hydrological applications.

In Section 2, we describe the architecture of our QPE system by specifying the three main subsystems associated with real-time data acquisition, NWP analysis and radar data processing, and final product visualization. Section 3 provides the algorithm details of module elements in the NWP, individual radar, and composite data processing. Section 4 evaluates the QPE products generated by our algorithms using rain gauge data and hydrologic simulations. In Section 5, we summarize the algorithm features and main findings from the product evaluation. Finally, we discuss the improvements gained by implementing the new polarimetric algorithm and potential future developments.