Storm Nowcasting Research Articles

Nowcasting of Storms Using Predicted Integrated Water Vapor With a Machine Learning Technique and Satellite Brightness Temperature

Nowcasting of Storms Using Predicted Integrated Water Vapor With a Machine Learning Technique and Satellite Brightness Temperature

A Customized Multi‐Scale Deep Learning Framework for Storm Nowcasting

AbstractStorm nowcasting is critical and urgently needed. Recent advances in deep learning (DL) have shown potential for improving nowcasting accuracy and predicting general low‐intensity precipitation events. However, DL models yield poor performance on high‐impact storms due to insufficient extraction and characterization of complex multi‐scale spatiotemporal variations of storms. To tackle this challenge, we propose a novel customized multi‐scale (CM) DL framework, including a flexible attention module capturing scale variations and a customized loss function ensuring multi‐scale spatiotemporal consistency. The CM framework was applied to the storm event imagery data set (SEVIR). Representative cases indicate that the CM framework preserves the shape of storms and adequately forecasts intense storms even for longer predictions. The quantitative evaluation shows that all models applying our framework can improve skill scores by 8.5%–42.6% for 1‐hr nowcasting. This work highlights the importance of modeling multi‐scale spatiotemporal characteristics of meteorological variables when using DL.

A Deep Learning–Based Velocity Dealiasing Algorithm Derived from the WSR-88D Open Radar Product Generator

Abstract Radial velocity estimates provided by Doppler weather radar are critical measurements used by operational forecasters for the detection and monitoring of life-impacting storms. The sampling methods used to produce these measurements are inherently susceptible to aliasing, which produces ambiguous velocity values in regions with high winds and needs to be corrected using a velocity dealiasing algorithm (VDA). In the United States, the Weather Surveillance Radar-1988 Doppler (WSR-88D) Open Radar Product Generator (ORPG) is a processing environment that provides a world-class VDA; however, this algorithm is complex and can be difficult to port to other radar systems outside the WSR-88D network. In this work, a deep neural network (DNN) is used to emulate the two-dimensional WSR-88D ORPG dealiasing algorithm. It is shown that a DNN, specifically a customized U-Net, is highly effective for building VDAs that are accurate, fast, and portable to multiple radar types. To train the DNN model, a large dataset is generated containing aligned samples of folded and dealiased velocity pairs. This dataset contains samples collected from WSR-88D Level-II and Level-III archives and uses the ORPG dealiasing algorithm output as a source of truth. Using this dataset, a U-Net is trained to produce the number of folds at each point of a velocity image. Several performance metrics are presented using WSR-88D data. The algorithm is also applied to other non-WSR-88D radar systems to demonstrate portability to other hardware/software interfaces. A discussion of the broad applicability of this method is presented, including how other Level-III algorithms may benefit from this approach. Significance Statement Accurate and timely estimates of wind within storms are critically important for a number of applications, including severe storm nowcasting, maritime operational planning, aviation forecasting, and public safety coordination. Velocity aliasing is a common artifact that requires data quality control. While velocity dealiasing algorithms (VDAs) have been developed for decades, they remain a computationally complex and challenging problem. This paper presents an application of deep neural networks (DNNs) to increase the computational efficiency and portability of VDAs. A DNN is trained to emulate an operational algorithm, and performance is quantified over a large dataset. This work gives a convincing example of the benefits that deep learning can provide for radar algorithms, and future work highlighting these opportunities is discussed.

Vyhodnocení radarových produktů pravděpodobnosti výskytu krup a jejich využití v ČHMÚ

Radar measurements are one of the key data sources for detection and nowcasting of convective storms and accompanying severe phenomena. Radar algorithm estimating probability of hail occurrence was first implemented in the Czech Hydrometeorological Institute (CHMI) in 2004. Development of new radar products estimating probability of severe hail occurrence and their visualization has taken place in the CHMI in recent years. This article describes evaluation of radar products which, combined with temperature sounding measurements, determine probability of hail occurrence and provide estimates of maximum hail size. These products are operationally produced in CHMI and visualized in a new JSMeteoView2 application which serves as the main tool for visualization of radar data and other remote sensing data in combination with detailed geographical layers. The article discusses performance of radar hail products tested on a large set of data and their operational usability when estimating hail presence.

Pixel-CRN: A New Machine Learning Approach for Convective Storm Nowcasting

The short-term convective storm forecasting (i.e., nowcasting) mainly rely on weather radar which can resolve the 3D structure of convective storms. With the rapid development of numerical models, modern models can produce 3D reanalysis data which gives atmospheric background information of convective storms. Current deep learning nowcasting models use only 2D radar images for nowcasting and often require massive historical data for training. But it may not be operationally feasible to collect long-term radar data to train a new model. Hence, how to establish a nowcasting model using only a small dataset has become an important issue. In addition, existing models do not effectively use the state-of-the-art model reanalysis data, which is a shortcoming of these models. To tackle these problems, this article develops a pixel-wise convolutional recurrent neural network (Pixel-CRN) for precipitation nowcasting. It has three key designs: (1) Through a concise pixel-wise sampling and oversampling technique, Pixel-CRN can be trained using only a small dataset. (2) In spatial learning, Pixel-CRN embeds a spatial convolution subnet into the recurrent unit, which can input raw 3D radar and model reanalysis data, thus valuable atmospheric background information can be learned to assist nowcasting. (3) In spatiotemporal learning, according to the information bottleneck principle, Pixel-CRN builds a heterogeneous encoder-decoder structure to squeeze multi-channel 3D input data into latent space, and recurrently generates 30 and 60 min nowcasts. Compared with existing deep learning nowcasting methods, the experimental results show that Pixel-CRN can provide skillful results with a rather small training dataset.

Combined assimilation of radar and lightning data for the short-term forecast of severe convection system

In this study, a lightning data assimilation (LDA) method adjusting dynamical fields was examined in the rapid cycle with Weather Research and Forecasting (WRF) model and WRF Data Assimilation (WRFDA). This method retrieves pseudo vertical velocity profiles from total lightning observations and promotes wind convergence over lightning regions. This newly developed LDA scheme is compared with radar data assimilation (RDA) that has been routinely applied in severe storm nowcasting models. Depending on whether the radar or lightning data is assimilated, four experiments were designed to evaluate the positive impacts of dynamical adjustment from LDA. Using a typical severe mesoscale convection system occurred in Beijing, we found that the effect of RDA and LDA are different. The assimilation of radar radial velocity mainly improves the forecast in a longer time, while assimilation of pseudo-vertical-velocity from lightning data mainly corrects intensity and location of the forecasted precipitation. When both radar and lightning data are assimilated, the small-scale wind convergence are promoted and contributes to the intensified updrafts. Thermal and water vapor fields are also adjusted indirectly. Consequently, the convective precipitation is significantly improved and the positive impacts from the combined assimilation scheme persisted over a longer period (at least 3 h). It can be concluded that a combined assimilation of convective data from multiple sources such as radar and lightning data enhance prominently the accuracy of short-term convection forecasts.

Three-Dimensional Gridded Radar Echo Extrapolation for Convective Storm Nowcasting Based on 3D-ConvLSTM Model

Radar echo extrapolation has been widely developed in previous studies for precipitation and storm nowcasting. However, most studies have focused on two-dimensional radar images, and extrapolation of multi-altitude radar images, which can provide more informative and visual forecasts about weather systems in realistic space, has been less explored. Thus, this paper proposes a 3D-convolutional long short-term memory (ConvLSTM)-based model to perform three-dimensional gridded radar echo extrapolation for severe storm nowcasting. First, a 3D-convolutional neural network (CNN) is used to extract the 3D spatial features of each input grid radar volume. Then, 3D-ConvLSTM layers are leveraged to model the spatial–temporal relationship between the extracted 3D features and recursively generate the 3D hidden states correlated to the future. Nowcasting results are obtained after applying another 3D-CNN to up-sample the generated 3D hidden states. Comparative experiments were conducted on a public National Center for Atmospheric Research Data Archive dataset with a 3D optical flow method and other deep-learning-based models. Quantitative evaluations demonstrate that the proposed 3D-ConvLSTM-based model achieves better overall and longer-term performance for storms with reflectivity values above 35 and 45 dBZ. In addition, case studies qualitatively demonstrate that the proposed model predicts more realistic storm evolution and can facilitate early warning regarding impending severe storms.

GNSS Storm Nowcasting Demonstrator for Bulgaria

Global Navigation Satellite System (GNSS) is an established atmospheric monitoring technique delivering water vapour data in near-real time with a latency of 90 min for operational Numerical Weather Prediction in Europe within the GNSS water vapour service (E-GVAP). The advancement of GNSS processing made the quality of real-time GNSS tropospheric products comparable to near-real-time solutions. In addition, they can be provided with a temporal resolution of 5 min and latency of 10 min, suitable for severe weather nowcasting. This paper exploits the added value of sub-hourly real-time GNSS tropospheric products for the nowcasting of convective storms in Bulgaria. A convective Storm Demonstrator (Storm Demo) is build using real-time GNSS tropospheric products and Instability Indices to derive site-specific threshold values in support of public weather and hail suppression services. The Storm Demo targets the development of service featuring GNSS products for two regions with hail suppression operations in Bulgaria, where thunderstorms and hail events occur between May and September, with a peak in July. The Storm Demo real-time Precise Point Positioning processing is conducted with the G-Nut software with a temporal resolution of 15 min for 12 ground-based GNSS stations in Bulgaria. Real-time data evaluation is done using reprocessed products and the achieved precision is below 9 mm, which is within the nowcasting requirements of the World Meteorologic Organisation. For the period May–September 2021, the seasonal classification function for thunderstorm nowcasting is computed and evaluated. The probability of thunderstorm detection is 83%, with a false alarm ration of 38%. The added value of the high temporal resolution of the GNSS tropospheric gradients is investigated for a storm case on 24–30 August 2021. Real-time tropospheric products and classification functions are integrated and updated in real-time on a publicly accessible geoportal.

Developing Deep Learning Models for Storm Nowcasting

Storm nowcasting relies on reasonably fast sampled radar data, and deep learning (DL) can be used to harness this vast amount of data. Despite all the publications on this topic over the past five years, there are still <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">ad hoc</i> assumptions and a lack of standardization. This work addresses aspects that have not yet been analyzed on the development of DL models for nowcasting systems, such as the effects of different history lengths or using non-convex metrics during the training phase. For example, we show that even if the loss function is varied, it does not significantly influence the predictions, and that the number of predicted frames has a significant impact. We used the experiments’ results to propose different models and compare their performance against other DL models. The results show that the proposed models outperform, in many aspects, the existing implementations.

A 3-D Storm Motion Estimation Method Based on Point Cloud Learning and Doppler Weather Radar Data

In recent years, deep learning techniques have been developed in the field of storm nowcasting, and they primarily focus on 2D radar product processing. Deep learning’s application to obtaining motion fields is limited by the lack of a motion field ground truth dataset for training. In this paper, we propose a method for storm motion estimation based on a point cloud deep learning network, which can perform 3D motion field estimation on two consecutive frames of weather radar volumetric data. To address the lack of ground truth data, we propose a synthetic data generation method based on Perlin noise and also generate a dataset to train the network. Our model’s architecture is based on FlowNet 3D. In this sense, we import an attention architecture to improve its ability to embed motion features. We also redesign the loss function to make it suitable for our task. In the experiment section, we first evaluate the performance of the trained model on our synthetic dataset. The experimental results demonstrate the following: first, that the trained network has the ability to estimate 3D motion from radar point cloud data; second, that it is a robust solution for radar data with different resolutions and observation ranges; and finally, that for the examined task, our proposed architecture has a 5-15% lower end point error than the original. Then, we directly apply the model trained on synthetic data to real cases, demonstrating the feasibility of our trained model on real data. Although motion field is a low-level product when considering a storm’s evolutionary process, it has the potential to serve a higher-level purpose, such as lightning or hail nowcasting. The point cloud approach provides a new perspective on weather radar data processing.

Advancing Radar Nowcasting Through Deep Transfer Learning

Deep learning is emerging as a powerful tool in scientific applications, such as radar-based convective storm nowcasting. However, it is still a challenge to extend the application of a well-trained deep learning nowcasting model, which demands to incorporate the learned knowledge at a certain location to other locations characterized by different precipitation features. This article designs a transfer learning framework to tackle this problem. A convolutional neural network (CNN)-based nowcasting method is utilized as the benchmark, based on which two transfer learning models are constructed through fine-tune and maximum mean discrepancy (MMD) minimization. The base CNN model is trained using radar data in the source study domain near Beijing, China, whereas the transferred models are applied to the target domain near Guangzhou, China, with only a small amount of data in the target area. The influence of a varying number of target data samples on the nowcasting performance is quantified. The experimental results demonstrate that the deep transfer learning models can improve the nowcasting skills.

Enhance Low Level Temperature and Moisture Profiles Through Combining NUCAPS, ABI Observations, and RTMA Analysis

AbstractThermodynamic information from low levels in the atmosphere is crucial for operational weather forecasts and meteorological researchers. The NOAA Unique Combined Atmospheric Processing System (NUCAPS) sounding products have been proven beneficial to fill the data gap between synoptic radiosonde observations (RAOBs). However, compared with the upper troposphere, the accuracy of NUCAPS soundings in the low levels still needs improvement. In this study, a deep neural network (DNN) is applied to fuse multiple data sources to enhance the NUCAPS temperature and moisture profiles in the lower atmosphere. The network is developed by combining satellite observations, including NUCAPS sounding retrievals and high resolution geostationary satellite observations from the Advanced Baseline Imager, and surface analysis from the Real‐Time Mesoscale Analysis (RTMA) as inputs, while collocated soundings from ECMWF re‐analysis version 5 are used as the benchmark for the training. The performance of the model is evaluated by using the independent testing data set, data from a different year, as well as collocated RAOBs, showing improvement to the temperature and moisture profiles by reducing the root‐mean‐squared‐error (RMSE) by more than 30% in the lower atmosphere (from 700 hPa to surface) in both clear sky and partially cloudy conditions. A convective event from June 18, 2017 is presented to illustrate the application of the enhanced low level soundings on high impact weather events. The enhanced soundings from fused data capture the large surface‐based convective available potential energy structures in the preconvection environment, which is very useful for severe storm nowcasting and forecasting applications.

Radar reflectivity signatures and possible lead times of warnings for very large hail in Poland based on data from 2007-2015

Abstract Hail involving very large hailstones (maximum diameter ≥ 5 cm), is a rare but very hazardous phenomenon in Poland, and can be forecast using reflectivity signatures. Every year, Poland experiences from one to over a dozen storms with such large hailstones. Despite the current recommendations regarding polarimetric techniques used in hail risk monitoring, Poland does not have a fully polarimetric radar network. Therefore it is essential to check hail detection capabilities using only reflectivity techniques based on individual radar systems involving hail detection algorithms such as Waldvogel et al. (1979) or Vertically Integrated Liquid thresholds connected with manual signature analysis to get better warning decisions. This study is aimed to determine the reflectivity features, thresholds and lead times for nowcasting of severe storms with very large hailstones in Poland, using data from the Polish radar system and from the European Severe Weather Database for the period 2007–2015. Most incidents involving very large hailstones were linked to supercell storms with distinctive reflectivity signatures, however, some storms with extremely large hailstones presented very poorly developed signatures. These signatures enabled the prediction of hail involving very large hailstones approximately 29 minutes before it fell. The Lemon (1980) criterion and WER were found to be the best hail predictors for Polish radar system conditions.

GNSS-Based Machine Learning Storm Nowcasting

Nowcasting of severe weather events and summer storms, in particular, are intensively studied as they have great potential for large economic and societal losses. Use of Global Navigation Satellite Systems (GNSS) observations for weather nowcasting has been investigated in various regions. However, combining the vertically integrated water vapour (IWV) with vertical profiles of wet refractivity derived from GNSS tomography has not been exploited for short-range forecasts of storms. In this study, we introduce a methodology to use the synergy of IWV and tomography-based vertical profiles to predict 0–2 h of storms using a machine learning approach for Poland. Moreover, we present an analysis of the importance of features that take part in the prediction process. The accuracy of the model reached over 87%, and the precision of prediction was about 30%. The results show that wet refractivity below 6 km and IWV on the west of the storm are among the significant parameters with potential for predicting storm location. The analysis of IWV demonstrated a correlation between IWV changes and storm occurrence.

Convolutional Neural Network for Convective Storm Nowcasting Using 3-D Doppler Weather Radar Data

Convective storms are one of the severe weather hazards found during the warm season. Doppler weather radar is the only operational instrument that can frequently sample the detailed structure of convective storm which has a small spatial scale and short lifetime. For the challenging task of short-term convective storm forecasting, 3-D radar images contain information about the processes in convective storm. However, effectively extracting such information from multisource raw data has been problematic due to a lack of methodology and computation limitations. Recent advancements in deep learning techniques and graphics processing units now make it possible. This article investigates the feasibility and performance of an end-to-end deep learning nowcasting method. The nowcasting problem was transformed into a classification problem first, and then, a deep learning method that uses a convolutional neural network was presented to make predictions. On the first layer of CNN, a cross-channel 3D convolution was proposed to fuse 3D raw data. The CNN method eliminates the handcrafted feature engineering, i.e., the process of using domain knowledge of the data to manually design features. Operationally produced historical data of the Beijing-Tianjin-Hebei region in China was used to train the nowcasting system and evaluate its performance; 3737332 samples were collected in the training data set. The experimental results show that the deep learning method improves nowcasting skills compared with traditional machine learning methods.

Severe Convective Weather Nowcasting System in Heilongjiang Meteorological Bureau and Its Preliminary Performance Evaluation

In this paper, we present the severe convective weather nowcasting system in Heilongjiang Meteorological Bureau and its preliminary performance evaluation, which has been in operational use since 2018. We believe that a good workflow design and intelligently integrated system plays an important role in the weather forecast service. Based on the current data environment, the system builds a unified data environment. Both the functions of automatic nowcasting system and Very-short-range Interactive Prediction System are presented too. The results show that this system plays a key role in the production of rapidly updated, convective storm nowcasting for end users.

Enhanced object-based tracking algorithm for convective rain storms and cells

This paper proposes a new object-based storm tracking algorithm, based upon TITAN (Thunderstorm Identification, Tracking, Analysis and Nowcasting). TITAN is a widely-used convective storm tracking algorithm but has limitations in handling small-scale yet high-intensity storm entities due to its single-threshold identification approach. It also has difficulties to effectively track fast-moving storms because of the employed matching approach that largely relies on the overlapping areas between successive storm entities. To address these deficiencies, a number of modifications are proposed and tested in this paper. These include a two-stage multi-threshold storm identification, a new formulation for characterizing storm's physical features, and an enhanced matching technique in synergy with an optical-flow storm field tracker, as well as, according to these modifications, a more complex merging and splitting scheme. High-resolution (5-min and 529-m) radar reflectivity data for 18 storm events over Belgium are used to calibrate and evaluate the algorithm. The performance of the proposed algorithm is compared with that of the original TITAN. The results suggest that the proposed algorithm can better isolate and match convective rainfall entities, as well as to provide more reliable and detailed motion estimates. Furthermore, the improvement is found to be more significant for higher rainfall intensities. The new algorithm has the potential to serve as a basis for further applications, such as storm nowcasting and long-term stochastic spatial and temporal rainfall generation.

A machine learning nowcasting method based on real‐time reanalysis data

AbstractDespite marked progress over the past several decades, convective storm nowcasting remains a challenge because most nowcasting systems are based on linear extrapolation of radar reflectivity without much consideration for other meteorological fields. The variational Doppler radar analysis system (VDRAS) is an advanced convective‐scale analysis system capable of providing analysis of 3‐D wind, temperature, and humidity by assimilating Doppler radar observations. Although potentially useful, it is still an open question as to how to use these fields to improve nowcasting. In this study, we present results from our first attempt at developing a support vector machine (SVM) box‐based nowcasting (SBOW) method under the machine learning framework using VDRAS analysis data. The key design points of SBOW are as follows: (1) The study domain is divided into many position‐fixed small boxes, and the nowcasting problem is transformed into one question, i.e., will a radar echo &gt; 35 dBZappear in a box in 30 min? (2) Box‐based temporal and spatial features, which include time trends and surrounding environmental information, are constructed. (3) And the box‐based constructed features are used to first train the SVM classifier, and then the trained classifier is used to make predictions. Compared with complicated and expensive expert systems, the above design of SBOW allows the system to be small, compact, straightforward, and easy to maintain and expand at low cost. The experimental results show that although no complicated tracking algorithm is used, SBOW can predict the storm movement trend and storm growth with reasonable skill.

Storm Tracking via Tree Structure Representation of Radar Data

AbstractAn algorithm for automatic storm identification, tracking, and nowcasting using tree structure representation of radar reflectivity images is proposed. The algorithm aims to track and nowcast different kinds of storm objects (stratiform regions, convective storms, and storm cells) simultaneously and to preserve their spatial relationships in the tracking and nowcasting processes. The algorithm applies a region tree structure to represent intensity regions and their spatial relationships in radar reflectivity images. Storm objects are identified by clustering regions within the region tree structure. Storm tracking is accomplished using an iterative region tree matching algorithm. Storm nowcasting applies the tree structure to the nowcasting of the internal structures of storm objects. Using eight cases with different storm types, a comparative evaluation with the enhanced Thunderstorm Identification, Tracking, Analysis, and Nowcasting (ETITAN) method and the Storm Cell Identification and Tracking (SCIT) method has shown that the proposed tree-based storm-tracking algorithm achieves better performance in storm tracking and nowcasting. The critical success index (CSI) value of storm association is 78.16% for the tree-based method, as compared with 74.88% for SCIT and 74.71% for ETITAN. The CSI value of an 18-min nowcast is 29.02% for the tree-based method, as compared with 24.98% for SCIT and 24.44% for ETITAN. The evaluation also shows that the tree-based method is able to nowcast the internal structure of storms and therefore produces small mean absolute errors (MAE).

Monitoring System for Atmospheric Water Vapor with a Ground-Based Multi-Band Radiometer: Meteorological Application of Radio Astronomy Technologies

High-resolution estimation of thermodynamic properties in the atmosphere can help to predict and mitigate meteorological disasters, such as local heavy rainfall and tornadic storms. For the purposes of short-term forecasting and nowcasting of severe storms, we propose a novel ground-based measurement system, which observes the intensity of atmospheric radiation in the microwave range. Our multi-band receiver system is designed to identify a rapid increase in water vapor before clouds are generated. At frequencies between 20 and 30 GHz, our system simultaneously measures water vapor as a broad absorption peak at 22 GHz as well as cloud liquid water. Another band at 50–60 GHz provides supplementary information from oxygen radiation to give vertical profiles of physical temperature. For the construction of this cold receiver system, novel technologies originally developed for observations of cosmic microwave background radiation were applied. The input atmospheric signal is amplified by a cold low-noise amplifier maintained below 10 K, while the spectrum of this amplified signal is measured using a signal analyzer under ambient conditions. The cryostat also contains a cold black body at 40 K to act as a calibration signal. This calibration signal is transported to each of the receivers via a wire grid. We can select either the atmospheric signal or the calibration signal by changing the orientation of this wire. Each receiver can be calibrated using this setup. Our system is designed to be compact (\(<\)1 m\(^{3}\)), with low power consumption (\(\sim \)1.5 kW). Therefore, it is easy to deploy on top of high buildings, mountains, and ship decks.