Search

A hybrid ensemble-variational (EnVAR) data assimilation and forecast system, namely WoF-Hybrid, was initially developed at the National Severe Storms Laboratory (NSSL) supported by the NOAA Warn-on-Forecast project (WoF). It is a real-time, deterministic, weather adaptive, dual-resolution analysis and forecasting system in parallel with the WoF ensemble data assimilation and forecast system (WoFS), which provides an on-demand, short-term, rapidly updating, convection-allowing ensemble forecasts to the end users. Some new components for the WoF-Hybrid system are developed and tested in this dissertation to address a few scientific and technical challenges in assimilating the conventional, radar and satellite derived observations to improve the short-term numerical weather prediction (NWP) of convective-scale severe and high-impact weather events. First of all, a novel component is introduced into the WoF-Hybrid system, which extends the observation forward operators to include cloud water path (CWP) and layered precipitable water (LPW). The impact of assimilating observations from several platforms on the estimation of initial states and the short-term forecasts using the WoF-Hybrid system is firstly investigated. The findings of this research reveal that (1) assimilating CWP improves the analysis of cloud properties, downward shortwave radiation over cloudy areas, surface temperature, and atmospheric instability, (2) assimilating the LPW product intensifies the moisture gradient at the surface and saturates the moisture fields surrounding ongoing convection, thereby enhancing the convection initiation (CI), and (3) low-level vertical wind shear is also improved through flow-dependent error covariances derived from the WoFS ensemble. Furthermore, a diagnostic analysis is conducted to explain why additional satellite-derived observations can improve the positioning and intensity forecast of storm cells, illustrated with three different hazardous severe weather events occurring in 2017. In the next part, the dissertation proceeds to further develop and evaluate an updated version of the variational component of the WoF-Hybrid system (named as WoF-Var hereafter) by incorporating weather-dependent static background error covariances derived from the WoFS ensemble forecast products. The impact of weather-dependent background error structure on WoF-Var analyses is evaluated using a single observation experiment, which reveals the heterogeneous nature of the analysis increment for the meridional wind field under different weather situations. Two severe weather events that occurred on 5 May 2020 and 10 Dec 2021 are used to evaluate the impact of the weather-dependent error covariances on the WoF-var analysis and forecast system in detail. Diagnostic analysis of both events indicates that the utilization of weather-dependent errors in the WoF-Var leads to improved analyses of in-storm structure, particularly for the three-dimensional wind, temperature, and humidity fields, resulting in significantly better subsequent 0- to 6-hour forecasts. The last part of the dissertation investigates the impact of weather-dependent error covariances on convective-scale NWP in the framework of WoF-Hybrid. In this, the flow-dependent ensemble error covariances are incorporated by using WoFS ensemble forecasts. Quantitative evaluations, aggregated over five selected severe and high-impact weather events, demonstrate that the experiments using weather-dependent covariances, instead of homogeneous covariances, in the hybrid system performs better in 0- to 6-hour reflectivity and rotational strength forecasts, particularly for strong convection with either higher reflectivity or higher updraft helicity. A more in-depth analysis of each event indicates that the adoption of weather-dependent covariances typically results in more skillful severe weather predictions regarding the areal coverage of reflectivity, the initiation and movement of storms, as well as the position and intensity of rotation tracks, even for the event that performed the worst out of the five selected events. The findings suggest that the weather-dependent error covariances within a hybrid system have added values to short-term convective-scale NWP.

Recently, an ensemble of the three-dimensional variational data assimilation (En3DA) system has been developed based on Advanced Regional Prediction System (ARPS) three-dimensional variational (3DVar) data assimilation (DA) system. However, the DA system is basically designed for radar data assimilation and only compatible with the ARPS model. In this study, the DA system is extended to include the interface that link the DA system with the broadly used Weather Research and Forecasting (WRF) model and to also incorporate additional satellite retrieval data into the DA process. A major goal here is to evaluate whether simultaneous assimilation of satellite retrieval data and radar data through an En3DA approach has positive impacts on initializing and forecasting convective-scale weather. The first part of this thesis describes the details about the WRF interface and algorithm of assimilating selected satellite data in the En3DA system. To better drive the WRF model, a post-processing program is introduced after DA. The forward operator and its adjoint for satellite data assimilation are embedded into the En3DA system. This DA system and the WRF model are then tested with an idealized tornadic supercell case and a real data tornadic supercell case. More precisely, an idealized supercell storm initialized by the sounding from the tornadic event occurred on 20 May 1977 Del City, Oklahoma, and the real data case is the 26 May 2016 Solomon, Kansas EF-4 tornadic thunderstorm case. The idealized case evaluates the assimilation of simulated radar and satellite retrieval data under different storm environments. Radar radial velocity, reflectivity, satellite derived cloud water path (CWP) and total precipitable water (TPW) data are produced from the simulated supercell storm, and then assimilated into the WRF model. Two types of experiments are performed. Radar and satellite CWP data are assimilated firstly under a perfect storm environment. Additional TPW data is assimilated under a dry biased environment in the second type of experiments. The first set of experiments indicates that incorporating satellite CWP data into radar data assimilation leads to a much faster initiation of supercell storms. Assimilating CWP data primarily improves the analyses of non-precipitating hydrometeor variables such as cloud and ice, as well as temperature. Results from the second set of experiments demonstrate the importance of the storm environment. With the biased storm environment, assimilation of CWP and radar data can enhance the analyses, but the forecast skill rapidly decreases in 1-hour forecast. Experiments show that the TPW data has a large effectiveness in the first 30-min free forecast, but gives little impact thereafter. The reason is that the TPW is a vertical integration of water vapor content and cannot provide information of the moisture vertical structure which is the key to the storm development. A similar experimental design is applied to a tornadic supercell storm event on 25-26 May 2016. Four sets of experiments are performed. The first one does not assimilate any data (NoDA), the second one assimilates CWP data alone (CWP), the third one assimilates radar data alone (RAD), and the final one assimilates both radar and satellite data (RADCWP). It is found that best DA and forecast results are provided in RADCWP experiment. In the DA results, the structures of the supercell storm including cold pool, wind fields and reflectivity pattern are reasonably analyzed in the RADCWP. The scope and intensity of strong precipitation area (indicated by reflectivity field) and cold pool area look more reasonable. The positive impacts of both data types are also supported by 1 hour free forecasts after DA cycling. Biases and RMS errors are smallest as well as skill scores are best for the RADCWP experiment when both data are used. In general, the idealized case with inaccurate moisture environment indicates that the correctness of storm environment is a significant component to the accuracy of the convective scale NWP. The combination of radar data and satellite data within the En3DA method results in better analyses and forecasts than that when only using radar data for the idealized. In the real data case, the positive impact of both radar and CWP data is demonstrated. However, this is only one real data case. Further investigations of the influence of radar and CWP data in storm-scale DA and convective NWP, especially the impact of more satellite data and product such as TPW, will be conducted in the future studies.