Storm-scale Numerical Weather Prediction Research Articles

A satellite lightning observation operator for storm-scale numerical weather prediction

Abstract. This study aims at simulating satellite-measured lightning observations with numerical weather prediction (NWP) system variables. A total of eight parameters, calculated with the AROME-France NWP system variables, were selected from a literature review to be used as proxies for satellite lightning observations. Two different proxy types emerged from this literature review: microphysical and dynamical proxies. Here, we investigate which ones are best related to satellite lightning and calibrate an empirical relationship between the best parameters and lightning data. To obtain those relationships, we fit machine learning regression models to our data. In this study, pseudo flash extent accumulation (FEA) observations are used because no actual geostationary lightning observations are available yet over France, and non-geostationary satellite lightning data represent a sample that is too small for our study. The performances of each proxy and machine learning regression model are evaluated by computing fractions skill scores (FSSs) with respect to observed FEA and proxy-based FEA. The present study suggests that microphysical proxies are more suited than the dynamical ones to model satellite lightning observations with the AROME-France NWP system. The performances of multivariate regression models are also evaluated by combining several proxies after a feature selection based on a principal component analysis and a proxy correlation study, but no proxy combination yielded better results than microphysical proxies alone. Finally, different accumulation periods of the FEA had little influence, i.e. similar FSS, on the regression model's ability to reproduce the observed FEA. In future studies, the microphysical-based relationship will be used as an observation operator to perform satellite lightning data assimilation in storm-scale NWP systems and applied to NWP forecasts to simulate satellite lightning data.

Non-Gaussian Probability Densities of Convection Initiation and Development Investigated Using a Particle Filter with a Storm-Scale Numerical Weather Prediction Model

Abstract Non-Gaussian probability density functions (PDFs) in convection initiation (CI) and development were investigated using a particle filter with a storm-scale numerical prediction model and an adaptive observation error estimator (NHM-RPF). An observing system simulation experiment (OSSE) was conducted with a 90-min assimilation period and 1000 particles at a 2-km grid spacing. Pseudosurface observations of potential temperature (PT), winds, water vapor (QV), and pseudoradar observations of rainwater (QR) in the lower troposphere were created in a nature run that simulated a well-developed cumulonimbus. The results of the OSSE (PF) show a significant improvement in comparison to ensemble simulations without any observations. The Gaussianity of the PDFs for PF in the CI area was evaluated using the Bayesian information criterion to compare goodness-of-fit of Gaussian, two-Gaussian mixture, and histogram models. The PDFs are strongly non-Gaussian when NHM-RPF produces diverse particles over the CI period. The non-Gaussian PDF of the updraft is followed by the upper-bounded PDF of the relative humidity, which produces non-Gaussian PDFs of QV and PT. The PDFs of the cloud water and QR are strongly non-Gaussian throughout the experimental period. We conclude that the non-Gaussianity of the CI originated from the non-Gaussianity of the updraft. In addition, we show that the adaptive observation error estimator significantly contributes to the stability of PF and the robustness to many observations.

Hydrometeor Mixing Ratio Retrievals for Storm-Scale Radar Data Assimilation: Utility of Current Relations and Potential Benefits of Polarimetry

Abstract The assimilation of radar data into storm-scale numerical weather prediction models has been shown to be beneficial for successfully modeling convective storms. Because of the difficulty of directly assimilating reflectivity (Z), hydrometeor mixing ratios, and sometimes rainfall rate, are often retrieved from Z observations using retrieval relations, and are assimilated as state variables. The most limiting (although widely employed) cases of these relations are derived, and their assumptions and limitations are discussed. To investigate the utility of these retrieval relations for liquid water content (LWC) and ice water content (IWC) in rain and hail as well as the potential for improvement using polarimetric variables, two models with spectral bin microphysics coupled with a polarimetric radar operator are used: a one-dimensional melting hail model and the two-dimensional Hebrew University Cloud Model. The relationship between LWC and Z in pure rain varies spatially and temporally, with biases clearly seen using the normalized number concentration. Retrievals using Z perform the poorest while specific attenuation and specific differential phase shift (KDP) perform much better. Within rain–hail mixtures, separate estimation of LWC and IWC is necessary. Prohibitively large errors in the retrieved LWC may result when using Z. The quantity KDP can be used to effectively retrieve the LWC and to isolate the contribution of IWC to Z. It is found that the relationship between Z and IWC is a function of radar wavelength, maximum hail diameter, and principally the height below the melting layer, which must be accounted for in order to achieve accurate retrievals.

A quantitative comparison of precipitation forecasts between the storm-scale numerical weather prediction model and auto-nowcast system in Jiangsu, China

Extrapolation techniques and storm-scale Numerical Weather Prediction (NWP) models are two primary approaches for short-term precipitation forecasts. The primary objective of this study is to verify precipitation forecasts and compare the performances of two nowcasting schemes: a Beijing Auto-Nowcast system (BJ-ANC) based on extrapolation techniques and a storm-scale NWP model called the Advanced Regional Prediction System (ARPS). The verification and comparison takes into account six heavy precipitation events that occurred in the summer of 2014 and 2015 in Jiangsu, China. The forecast performances of the two schemes were evaluated for the next 6h at 1-h intervals using gridpoint-based measures of critical success index, bias, index of agreement, root mean square error, and using an object-based verification method called Structure-Amplitude-Location (SAL) score. Regarding gridpoint-based measures, BJ-ANC outperforms ARPS at first, but then the forecast accuracy decreases rapidly with lead time and performs worse than ARPS after 4–5h of the initial forecast. Regarding the object-based verification method, most forecasts produced by BJ-ANC focus on the center of the diagram at the 1-h lead time and indicate high-quality forecasts. As the lead time increases, BJ-ANC overestimates precipitation amount and produces widespread precipitation, especially at a 6-h lead time. The ARPS model overestimates precipitation at all lead times, particularly at first.

Evaluation and Correction of Quantitative Precipitation Forecast by Storm-Scale NWP Model in Jiangsu, China

With the development of high-performance computer systems and data assimilation techniques, storm-scale numerical weather prediction (NWP) models are gradually used for short-term deterministic forecasts. The primary objective of this study is to evaluate and correct precipitation forecasts of a storm-scale NWP model called the advanced regional prediction system (ARPS). The evaluation and correction consider five heavy precipitation events that occurred in the summer of 2015 in Jiangsu, China. The performances of the original and corrected ARPS precipitation forecasts are evaluated as a function of lead time using standard measurements and a spatial verification method called Structure-Amplitude-Location (SAL). In general, the ARPS could not produce optimal forecasts for very short lead times, and the forecast accuracy improves with increasing lead time. The ARPS overestimates precipitation for all lead times, which is confirmed by large bias in many forecasts in the first and second quadrant of the diagram of SAL, especially at the 1 h lead time. The amplitude correction is performed by matching percentile values of the ARPS precipitation forecasts and observations for each lead time. Amplitude correction significantly improved the ARPS precipitation forecasts in terms of the considered performance indices of standard measures and A-component and S-component of SAL.

Day-Ahead Hail Prediction Integrating Machine Learning with Storm-Scale Numerical Weather Models

Hail causes billions of dollars in losses by damaging buildings, vehicles, and crops. Improving the spatial and temporal accuracy of hail forecasts would allow people to mitigate hail damage. We have developed an approach to forecasting hail that identifies potential hail storms in storm-scale numerical weather prediction models and matches them with observed hailstorms. Machine learning models, including random forests, gradient boosting trees, and linear regression, are used to predict the expected hail size from each forecast storm. The individual hail size forecasts are merged with a spatial neighborhood ensemble probability technique to produce a consensus probability of hail at least 25.4 mm in diameter. The system was evaluated during the 2014 National Oceanic and Atmospheric Administration Hazardous Weather Testbed Experimental Forecast Program and compared with a physics-based hail size model. The machine-learning-based technique shows advantages in producing smaller size errors and more reliable probability forecasts. The machine learning approaches correctly predicted the location and extent of a significant hail event in eastern Nebraska and a marginal severe hail event in Colorado.

Improvement of forecast skill for severe weather by merging radar-based extrapolation and storm-scale NWP corrected forecast

The primary objective of this study is to improve the performance of deterministic high resolution rainfall forecasts caused by severe storms by merging an extrapolation radar-based scheme with a storm-scale Numerical Weather Prediction (NWP) model. Effectiveness of Multi-scale Tracking and Forecasting Radar Echoes (MTaRE) model was compared with that of a storm-scale NWP model named Advanced Regional Prediction System (ARPS) for forecasting a violent tornado event that developed over parts of western and much of central Oklahoma on May 24, 2011. Then the bias corrections were performed to improve the forecast accuracy of ARPS forecasts. Finally, the corrected ARPS forecast and radar-based extrapolation were optimally merged by using a hyperbolic tangent weight scheme. The comparison of forecast skill between MTaRE and ARPS in high spatial resolution of 0.01°×0.01° and high temporal resolution of 5min showed that MTaRE outperformed ARPS in terms of index of agreement and mean absolute error (MAE). MTaRE had a better Critical Success Index (CSI) for less than 20-min lead times and was comparable to ARPS for 20- to 50-min lead times, while ARPS had a better CSI for more than 50-min lead times. Bias correction significantly improved ARPS forecasts in terms of MAE and index of agreement, although the CSI of corrected ARPS forecasts was similar to that of the uncorrected ARPS forecasts. Moreover, optimally merging results using hyperbolic tangent weight scheme further improved the forecast accuracy and became more stable.

Ensemble Probabilistic Forecasts of a Tornadic Mesoscale Convective System from Ensemble Kalman Filter Analyses Using WSR-88D and CASA Radar Data

AbstractThis study examines the ability of a storm-scale numerical weather prediction (NWP) model to predict precipitation and mesovortices within a tornadic mesoscale convective system that occurred over Oklahoma on 8–9 May 2007, when the model is initialized from ensemble Kalman filter (EnKF) analyses including data from four Engineering Research Center for Collaborative Adaptive Sensing of the Atmosphere (CASA) X-band and five Weather Surveillance Radar-1988 Doppler (WSR-88D) S-band radars. Ensemble forecasts are performed and probabilistic forecast products generated, focusing on prediction of radar reflectivity (a proxy of quantitative precipitation) and mesovortices (an indication of tornado potential).Assimilating data from both the CASA and WSR-88D radars for the ensemble and using a mixed-microphysics ensemble during data assimilation produces the best probabilistic mesovortex forecast. The use of multiple microphysics schemes within the ensemble aims to address at least partially the model physics uncertainty and effectively plays a role of flow-dependent inflation (in precipitation regions) during EnKF data assimilation. The ensemble predicts with high probability (approximately 0.65) the near-surface mesovortex associated with the first of three reported tornadoes. Though a bias toward stronger precipitation is noted in the ensemble forecasts, all experiments produce skillful probabilistic forecasts of radar reflectivity on a 0–3-h time scale as evaluated by multiple probabilistic verification metrics. These results suggest that both the inclusion of CASA radar data and use of a mixed-microphysics ensemble during EnKF data assimilation positively impact the skill of 2–3-h ensemble forecasts of mesovortices, despite having little impact on the quality of precipitation forecasts (analyzed in terms of predicted radar reflectivity), and are important steps toward an operational EnKF-based ensemble analysis and probabilistic forecast system to support convective-scale warn-on-forecast operations.

The Impact of Size Sorting on the Polarimetric Radar Variables

AbstractDifferential sedimentation of precipitation occurs because heavier hydrometeors fall faster than lighter ones. Updrafts and vertical wind shear can maintain this otherwise transient size sorting, resulting in prolonged regions of ongoing particle sorting in storms. This study quantifies the impact of size sorting on the S-band polarimetric radar variables (radar reflectivity factor at horizontal polarization ZH, differential reflectivity ZDR, specific differential phase KDP, and the copolar cross-correlation coefficient ρhv). These variables are calculated from output of two idealized bin models: a one-dimensional model of pure raindrop fallout and a two-dimensional rain shaft encountering vertical wind shear. Additionally, errors in the radar variables as simulated by single-, double-, and triple-moment bulk microphysics parameterizations are quantified for the same size sorting scenarios.Size sorting produces regions of sparsely concentrated large drops with a lack of smaller drops, causing ZDR enhancements as large as 1 dB in areas of decreased ZH, often along a ZH gradient. Such areas of enhanced ZDR are offset from those of high ZH and KDP. Illustrative examples of polarimetric radar observations in a variety of precipitation regimes demonstrate the widespread occurrence of size sorting and are consistent with the bin model simulations. Single-moment schemes are incapable of size sorting, leading to large underestimations in ZDR (>2 dB) compared to the bin model solution. Double-moment schemes with a fixed spectral shape parameter produce excessive size sorting by incorrectly increasing the number of large raindrops, overestimating ZDR by 2–3 dB. Three-moment schemes with variable shape parameters better capture the narrowing drop size distribution resulting from size sorting but can underestimate ZDR and overestimate KDP by as much as 20%. Implications for polarimetric radar data assimilation into storm-scale numerical weather prediction models are discussed.

Analysis of a Tornadic Mesoscale Convective Vortex Based on Ensemble Kalman Filter Assimilation of CASA X-Band and WSR-88D Radar Data

Abstract One of the goals of the National Science Foundation Engineering Research Center (ERC) for Collaborative Adaptive Sensing of the Atmosphere (CASA) is to improve storm-scale numerical weather prediction (NWP) by collecting data with a dense X-band radar network that provides high-resolution low-level coverage, and by assimilating such data into NWP models. During the first spring storm season after the deployment of four radars in the CASA Integrated Project-1 (IP-1) network in southwest Oklahoma, a tornadic mesoscale convective system (MCS) was captured by CASA and surrounding Weather Surveillance Radars-1988 Doppler (WSR-88Ds) on 8–9 May 2007. The MCS moved across northwest Texas and western and central Oklahoma; two tornadoes rated as category 1 on the enhanced Fujita scale (EF-1) and one tornado of EF-0 intensity were reported during the event, just to the north of the IP-1 network. This was the first tornadic convective system observed by CASA. To quantify the impacts of CASA radar data in storm-scale NWP, a set of data assimilation experiments were performed using the Advanced Regional Prediction System (ARPS) ensemble Kalman filter (EnKF) system configured with full model physics and high-resolution terrain. Data from four CASA IP-1 radars and five WSR-88Ds were assimilated in some of the experiments. The ensemble contained 40 members, and radar data were assimilated every 5 min for 1 h. While the assimilation of WSR-88D data alone was able to produce a reasonably accurate analysis of the convective system, assimilating CASA data in addition to WSR-88D data is found to improve the representation of storm-scale circulations, particularly in the lowest few kilometers of the atmosphere, as evidenced by analyses of gust front position and comparison of simulated Vr with observations. Assimilating CASA data decreased RMS innovation of the resulting ensemble mean analyses of Z, particularly in early assimilation cycles, suggesting that the addition of CASA data allowed the EnKF system to more quickly achieve a good result. Use of multiple microphysics schemes in the forecast ensemble was found to alleviate underdispersion by increasing the ensemble spread. This work is the first assimilating real CASA data into an NWP model using EnKF.

Assimilation of Simulated Polarimetric Radar Data for a Convective Storm Using the Ensemble Kalman Filter. Part I: Observation Operators for Reflectivity and Polarimetric Variables

Abstract A radar simulator for polarimetric radar variables, including reflectivities at horizontal and vertical polarizations, the differential reflectivity, and the specific differential phase, has been developed. This simulator serves as a test bed for developing and testing forward observation operators of polarimetric radar variables that are needed when directly assimilating these variables into storm-scale numerical weather prediction (NWP) models, using either variational or ensemble-based assimilation methods. The simulator takes as input the results of high-resolution NWP model simulations with ice microphysics and produces simulated polarimetric radar data that may also contain simulated errors. It is developed based on calculations of electromagnetic wave propagation and scattering at the S band of wavelength 10.7 cm in a hydrometeor-containing atmosphere. The T-matrix method is used for the scattering calculation of raindrops and the Rayleigh scattering approximation is applied to snow and hail particles. The polarimetric variables are expressed as functions of the hydrometeor mixing ratios as well as their corresponding drop size distribution parameters and densities. The presence of wet snow and wet hail in the melting layer is accounted for by using a new, relatively simple melting model that defines the water fraction in the melting snow or hail. The effect of varying density due to the melting snow or hail is also included. Vertical cross sections and profiles of the polarimetric variables for a simulated mature multicellular squall-line system and a supercell storm show that polarimetric signatures of the bright band in the stratiform region and those associated with deep convection are well captured by the simulator.

The Assimilation of Radar Data for Weather Prediction

Since the early years of radar meteorology, scientists have given a good deal of attention to the development of techniques for deducing the unobserved meteorological fields from Doppler radar observations. For example, Lhermitte and Atlas (1961) and Atlas (1964) described the velocity-azimuth display (VAD) method for determining a horizontally averaged wind profile from single-Doppler radial velocity observations. Lhermitte (1970) developed a method to derive the detailed horizontal wind field from dual-Doppler radial velocity observations. These early research activities were largely driven by the desire to understand the dynamical processes of convective systems in the atmosphere. In the late 1980s, however, scientists began to recognize that radar observations would also play an important role in future mesoscale and convective-scale data assimilation systems for short-term weather forecasting. This scientific vision was gained, in large part, as a result of the planned implementation of the Weather Surveillance Radar-1988 Doppler (WSR-88D) network in the United States and the rapid increase in computer power. With these increased resources, it appeared that running numerical weather prediction (NWP) models with resolutions that are able to resolve mesoscale and cloudscale features was within reach. On account of the considerable societal and economic impact of the operational storm-scale NWP and the scientific and technological advancement, Lilly (1990) argued that it was time for convective-storm research scientists to apply their knowledge and to show whether storm-scale NWP was a realistic goal. Since the WSR-88D network is the major observing system capable of sampling the four-dimensional structure of storm-scale flows, a key scientific and technical challenge is whether these observations can be used to initialize high-resolution storm prediction models.

Phase-Correcting Data Assimilation and Application to Storm-Scale Numerical Weather Prediction. Part I: Method Description and Simulation Testing

An objective method of determining and correcting phase or position errors in numerical weather prediction is described and tested in a radar data observing system simulation experiment (OSSE) addressing the forecasting of ongoing thunderstorms. Such phase or position errors are common in numerical forecasts at grid resolutions of 2–20 km (meso-γ scale). It is proposed that the process of correcting a numerical forecast field can be simplified if such errors are addressed directly. An objective method of determining the phase error in the forecast by searching for a field of shift vectors that minimizes a squared-error difference from high-resolution observations is described. Three methods of applying a phase error correction to a forecast model are detailed. The first applies the entire correction at the initial time, the second in discrete steps during an assimilation window, and the third applies the correction continuously through the model's horizontal advection process. It is shown that the phase correction method is effective in producing an analysis field that agrees with the data yet preserves the structure developed by the model. The three methods of assimilating the correction in the forecast are successful, and a long-term positive effect on the thunderstorm simulation is achieved in the simulations, even as the modeled storms go through a cycle of decline and regeneration.

Phase-Correcting Data Assimilation and Application to Storm-Scale Numerical Weather Prediction. Part II: Application to a Severe Storm Outbreak

Abstract A scheme to correct phase errors in numerical model forecasts using Doppler radar, radiosonde, profiler, and surface data is demonstrated to improve forecasts in a complex severe thunderstorm situation. The technique is designed to directly address forecast phase errors or initial position errors as part of a data assimilation strategy. In the demonstration the phase error correction is applied near the time of initial cell development and the forecast results are compared to the uncorrected forecast and forecasts made using an analysis created at the time of the observations. Forecasts are verified qualitatively for the position of thunderstorm cells and quantitatively for accumulated precipitation. It is shown that the scheme can successfully correct errors in thunderstorm locations and it has a positive influence on the subsequent forecast. The advantage of the phase correction over the control lasts for about 3 h despite storm dissipation and regeneration, and interactions among multiple storms.

The Advanced Regional Prediction System (ARPS), storm-scale numerical weather prediction and data assimilation

In this paper, we first describe the current status of the Advanced Regional Prediction System of the Center for Analysis and Prediction of Storms at the University of Oklahoma. A brief outline of future plans is also given. Two rather successful cases of explicit prediction of tornadic thunderstorms are then presented. In the first case, a series of supercell storms that produced a historical number of tornadoes was successfully predicted more than 8 hours in advance, to within tens of kilometers in space with initiation timing errors of less than 2 hours. The general behavior and evolution of the predicted thunderstorms agree very well with radar observations. In the second case, reflectivity and radial velocity observations from Doppler radars were assimilated into the model at 15-minute intervals. The ensuing forecast, covering a period of several hours, accurately reproduced the intensification and evolution of a tornadic supercell that in reality spawned two tornadoes over a major metropolitan area. These results make us optimistic that a model system such as the ARPS will be able to deterministically predict future severe convective events with significant lead time. The paper also includes a brief description of a new 3DVAR system developed in the ARPS framework. The goal is to combine several steps of Doppler radar retrieval with the analysis of other data types into a single 3-D variational framework and later to incorporate the ARPS adjoint to establish a true 4DVAR data assimilation system that is suitable for directly assimilating a wide variety of observations for flows ranging from synoptic down to the small nonhydrostatic scales.