Quantitative Precipitation Nowcasting Research Articles

Key factors for quantitative precipitation nowcasting using ground weather radar data based on deep learning

Abstract. Quantitative precipitation nowcasting (QPN) can help to reduce the enormous socioeconomic damage caused by extreme weather. The QPN has been a challenging topic due to rapid atmospheric variability. Recent QPN studies have proposed data-driven models using deep learning (DL) and ground weather radar. Previous studies have primarily focused on developing DL models, but other factors for DL-QPN have not been thoroughly investigated. This study examined four critical factors in DL-QPN, focusing on their impact on forecasting performance. These factors are the deep learning model (U-Net, as well as a convolutional long short-term memory, or ConvLSTM), input past sequence length (1, 2, or 3 h), loss function (mean squared error, MSE, or balanced MSE, BMSE), and ensemble aggregation. A total of 24 schemes were designed to measure the effects of each factor using weather radar data from South Korea with a maximum lead time of 2 h. A long-term evaluation was conducted for the summers of 2020–2022 from an operational perspective, and a heavy rainfall event was analyzed to examine an extreme case. In both evaluations, U-Net outperformed ConvLSTM in overall accuracy metrics. For the critical success index (CSI), MSE loss yielded better results for both models in the weak intensity range (≤ 5 mm h−1), whereas BMSE loss was more effective for heavier precipitation. There was a small trend where a longer input time (3 h) gave better results in terms of MSE and BMSE, but this effect was less significant than other factors. The ensemble by averaging results of using MSE and BMSE losses provided balanced performance across all aspects, suggesting a potential strategy to improve skill scores when implemented with optimal weights for each member. All DL-QPN schemes exhibited problems with underestimation and overestimation when trained by MSE and BMSE losses, respectively. All DL models produced blurry results as the lead time increased, while the non-DL model retained detail in prediction. With a comprehensive comparison of these crucial factors, this study offers a modeling strategy for future DL-QPN work using weather radar data.

A Deep Learning-Based Algorithm for Identifying Precipitation Clouds Using Fengyun-4A Satellite Observation Data.

Rapid and accurate identification of precipitation clouds from satellite observations is essential for the research of quantitative precipitation estimation and precipitation nowcasting. In this study, we proposed a novel Convolutional Neural Network (CNN)-based algorithm for precipitation cloud identification (PCINet) in the daytime, nighttime, and nychthemeron. High spatiotemporal and multi-spectral information from the Fengyun-4A (FY-4A) satellite is utilized as the inputs, and a multi-scale structure and skip connection constraint strategy are presented in the framework of the algorithm to improve the precipitation cloud identification. Moreover, the effectiveness of visible/near-infrared spectral information in improving daytime precipitation cloud identification is explored. To evaluate this algorithm, we compare it with five other deep learning models used for image segmentation and perform qualitative and quantitative analyses of long-time series using data from 2021. In addition, two heavy precipitation events are selected to analyze the spatial distribution of precipitation cloud identification. Statistics and visualization of the experiment results show that the proposed model outperforms the baseline models in this task, and adding visible/near-infrared spectral information in the daytime can effectively improve model performance. More importantly, the proposed model can provide accurate and near-real-time results, which has important application in observing precipitation clouds.

Precipitation nowcasting using ground radar data and simpler yet better video prediction deep learning

ABSTRACT Skillful quantitative precipitation nowcasting (QPN) is important for predicting precipitation in the upcoming few hours and thus avoiding significant socioeconomic damage. Recent QPN studies have actively adopted deep learning (DL) to generate precipitation maps using sequences of ground radar data. Although high skill scores in forecasting precipitation areas of weak intensity (~1 mm/h) have been achieved, the horizontal movement of precipitation areas could not be accurately simulated, exhibiting poor forecasting skills for stronger intensities. For lead times up to 120 min, this study suggests using an improved radar-based QPN model that utilizes a state-of-the-art DL model termed simpler yet better video prediction (SimVP). An independent evaluation using ground radar data in South Korea from June to September 2022 demonstrated that the proposed model outperformed the existing DL models in terms of critical score index (CSI) with a lead time of 120 min (0.46, 0.23, and 0.09 for 1, 5, and 10 mm/h thresholds, respectively). Three case analyses were conducted to reflect various precipitation conditions: heavy rainfall, typhoons, and fast-moving narrow convection events. The proposed SimVP-based QPN model yielded robust performance for all cases, producing a comparable or highest CSI at the lead time of 120 min with a 1 mm/h threshold (0.49, 0.69, and 0.29 for heavy rainfall, typhoon, and narrow convection, respectively). Qualitative evaluation of the model indicated better results in terms of displacement movement and reduced underestimation than other models under the high variability of precipitation patterns of the three cases. A comparison of model complexity among DL-QPN models was conducted, taking into consideration operational applications across various study areas and environments. The proposed approach is expected to provide a new baseline for DL-based QPN, and the improved prediction using the proposed model can lead to reduced socioeconomic damage incurred as a result of short-term intense precipitation.

Accurate and Clear Quantitative Precipitation Nowcasting Based on a Deep Learning Model with Consecutive Attention and Rain-Map Discrimination

Abstract Deep learning models are developed for high-resolution quantitative precipitation nowcasting (QPN) in Taiwan up to 3 h ahead. Many recent works aim to accurately predict relatively rare high-rainfall events with the help of deep learning. This rarity is often addressed by formulations that reweight the rare events. However, these formulations often carry a side effect of producing blurry rain-map nowcasts that overpredict in low-rainfall regions. Such nowcasts are visually less trustworthy and practically less useful for forecasters. We fix the trust issue by introducing a discriminator that encourages the model to generate realistic rain maps without sacrificing the predictive accuracy of rainfall extremes. Moreover, with consecutive attention across different hours, we extend the nowcasting time frame from typically 1 to 3 h to further address the needs for socioeconomic weather-dependent decision-making. By combining the discriminator and the attention techniques, the proposed model based on the convolutional recurrent neural network is trained with a dataset containing radar reflectivity and rain rates at a granularity of 10 min and predicts the hourly accumulated rainfall in the next three hours. Model performance is evaluated from both statistical and case-study perspectives. Statistical verification shows that the new model outperforms the current operational QPN techniques. Case studies further show that the model can capture the motion of rainbands in a frontal case and also provide an effective warning of urban-area torrential rainfall in an afternoon-thunderstorm case, implying that deep learning has great potential and is useful in 0–3-h nowcasting.

Near-Realtime Quantitative Precipitation Estimation and Prediction (RealPEP)

Flash floods in small- to medium-sized catchments and intense precipitation over cities caused by severe local storms pose increasing threats to our society. For the timely prediction of such events, the value of high-resolution and high-quality QPE and corresponding forecasts cannot be overrated. Seamless predictions harmonizing nowcasting and numerical weather prediction (NWP) across forecast lead times from minutes to days would greatly help to improve the value and efficiency of warnings. Organized by the Research Unit on Near-Realtime Precipitation Estimation and Prediction (RealPEP, www2.meteo.uni-bonn.de/realpep) and supported by the Project on Seamless Integrated Forecasting System (SINFONY, www.dwd .de/DE/forschung/forschungsprogramme/sinfony_iafe/sinfony_node.html) of the German Meteorological Service (DWD), an international 3-day online conference was held from 5 to 7 October 2020, dedicated to Precipitation and Flash-Flood Predictions from Minutes to Days (https://indico .scc.kit.edu/event/883/). Most speakers agreed to have their presentations recorded, which we uploaded to YouTube for further distribution (see, e.g., on the conference homepage, https:// indico.scc.kit.edu/event/883/page/588-recorded-talks). The speakers were both invited experts in the respective research fields and researchers from the RealPEP and SINFONY projects. Talks and discussions could be followed on video stream. Interaction between the about 250 participants was enabled by entering written questions and comments via a dedicated tool, which allowed for voting and thus also ranking questions. Registered participants could enter chat rooms from where they could be moved to the speaker room for posing the questions directly to the speakers and the auditorium. On the last day of the conference podium discussions with selected speakers summarized talks and discussions and elaborated on overarching problems, ideas, and developments in the fields of quantitative precipitation estimation (QPE), quantitative precipitation nowcasting (QPN), quantitative precipitation forecasting (QPF), flash-flood prediction (FFP), and their organization into seamless prediction systems, which also constituted the topics of the five sessions during the conference. We report here in particular on the outcomes of the panel discussions.

Optical flow models as an open benchmark for radar-based precipitation nowcasting (rainymotion v0.1)

Abstract. Quantitative precipitation nowcasting (QPN) has become an essential technique in various application contexts, such as early warning or urban sewage control. A common heuristic prediction approach is to track the motion of precipitation features from a sequence of weather radar images and then to displace the precipitation field to the imminent future (minutes to hours) based on that motion, assuming that the intensity of the features remains constant (“Lagrangian persistence”). In that context, “optical flow” has become one of the most popular tracking techniques. Yet the present landscape of computational QPN models still struggles with producing open software implementations. Focusing on this gap, we have developed and extensively benchmarked a stack of models based on different optical flow algorithms for the tracking step and a set of parsimonious extrapolation procedures based on image warping and advection. We demonstrate that these models provide skillful predictions comparable with or even superior to state-of-the-art operational software. Our software library (“rainymotion”) for precipitation nowcasting is written in the Python programming language and openly available at GitHub (https://github.com/hydrogo/rainymotion, Ayzel et al., 2019). That way, the library may serve as a tool for providing fast, free, and transparent solutions that could serve as a benchmark for further model development and hypothesis testing – a benchmark that is far more advanced than the conventional benchmark of Eulerian persistence commonly used in QPN verification experiments.

Retraction: Effects of the Forecasting Methods, Precipitation Character, and Satellite Resolution on the Predictability of Short-Term Quantitative Precipitation Nowcasting (QPN) from a Geostationary Satellite.

Retraction: Effects of the Forecasting Methods, Precipitation Character, and Satellite Resolution on the Predictability of Short-Term Quantitative Precipitation Nowcasting (QPN) from a Geostationary Satellite.

The Implementation of the Ice-Phase Microphysical Process into a Four-Dimensional Variational Doppler Radar Analysis System (VDRAS) and Its Impact on Parameter Retrieval and Quantitative Precipitation Nowcasting

Abstract The microphysical process of a cloud-scale model used by a four-dimensional Variational Doppler Radar Analysis System (VDRAS) is extended from its original warm rain parameterization scheme to a cold rain process containing ice and snow. The development of the adjoint equations for the additional control variables related to ice physics is accomplished by utilizing the existing four-dimensional variational (4DVar) minimization framework employed by VDRAS. Experiments are conducted to examine the accuracy of the new 4DVar system with the ice physics scheme implemented and to explore the impact of the ice-phase process on numerical simulations, parameter retrievals, and the model’s quantitative precipitation nowcasting (QPN) capability. It is shown that the ice-phase microphysical process can significantly alter the kinematic and thermodynamic structure of deep convection and provide a better description of the contents of the hydrometeors. During the 4DVar minimization, using the VDRAS-predicted freezing level after the previous assimilation cycle to replace the true but unknown 0°C line is found to be a feasible approach for separating the rain and snow and, at the same time, allowing the 4DVar minimization algorithm to converge to an optimal solution. A real case study from intensive observation period 8 of the 2008 Southwest Monsoon Experiment shows that, with the added ice-phase process, VDRAS is more capable of capturing the actual evolution of the reflectivity field than the original scheme. The model’s QPN skill is also improved significantly. Thus, the benefits of adding the ice-phase process into a 4DVar radar data assimilation system on the convective-scale weather analysis and forecast are demonstrated.

Effects of the Forecasting Methods, Precipitation Character, and Satellite Resolution on the Predictability of Short-Term Quantitative Precipitation Nowcasting (QPN) from a Geostationary Satellite.

The prediction of the short-term quantitative precipitation nowcasting (QPN) from consecutive gestational satellite images has important implications for hydro-meteorological modeling and forecasting. However, the systematic analysis of the predictability of QPN is limited. The objective of this study is to evaluate effects of the forecasting model, precipitation character, and satellite resolution on the predictability of QPN usingimages of a Chinese geostationary meteorological satellite Fengyun-2F (FY-2F) which covered all intensive observation since its launch despite of only a total of approximately 10 days. In the first step, three methods were compared to evaluate the performance of the QPN methods: a pixel-based QPN using the maximum correlation method (PMC); the Horn-Schunck optical-flow scheme (PHS); and the Pyramid Lucas-Kanade Optical Flow method (PPLK), which is newly proposed here. Subsequently, the effect of the precipitation systems was indicated by 2338 imageries of 8 precipitation periods. Then, the resolution dependence was demonstrated by analyzing the QPN with six spatial resolutions (0.1atial, 0.3a, 0.4atial rand 0.6). The results show that the PPLK improves the predictability of QPN with better performance than the other comparison methods. The predictability of the QPN is significantly determined by the precipitation system, and a coarse spatial resolution of the satellite reduces the predictability of QPN.

A new methodology for pixel-quantitative precipitation nowcasting using a pyramid Lucas Kanade optical flow approach

Short-term high-resolution Quantitative Precipitation Nowcasting (QPN) has important implications for navigation, flood forecasting, and other hydrological and meteorological concerns. This study proposes a new algorithm called Pixel-based QPN using the Pyramid Lucas-Kanade Optical Flow method (PPLK), which comprises three steps: employing a Pyramid Lucas-Kanade Optical Flow method (PLKOF) to estimate precipitation advection, projecting rainy clouds by considering the advection and evolution pixel by pixel, and interpolating QPN imagery based on the space-time continuum of cloud patches. The PPLK methodology was evaluated with 2338 images from the geostationary meteorological satellite Fengyun-2F (FY-2F) of China and compared with two other advection-based methods, i.e., the maximum correlation method and the Horn-Schunck Optical Flow scheme. The data sample covered all intensive observations since the launch of FY-2F, despite covering a total of only approximately 10 days. The results show that the PPLK performed better than the algorithms used for comparison, demonstrating less time expenditure, more effective cloud tracking, and improved QPN accuracy. (C) 2015 Elsevier B.V. All rights reserved.

Improving quantitative precipitation nowcasting with a local ensemble transform Kalman filter radar data assimilation system: observing system simulation experiments

This study develops a Doppler radar data assimilation system, which couples the local ensemble transform Kalman filter with the Weather Research and Forecasting model. The benefits of this system to quantitative precipitation nowcasting (QPN) are evaluated with observing system simulation experiments on Typhoon Morakot (2009), which brought record-breaking rainfall and extensive damage to central and southern Taiwan. The results indicate that the assimilation of radial velocity and reflectivity observations improves the three-dimensional winds and rain-mixing ratio most significantly because of the direct relations in the observation operator. The patterns of spiral rainbands become more consistent between different ensemble members after radar data assimilation. The rainfall intensity and distribution during the 6-hour deterministic nowcast are also improved, especially for the first 3 hours. The nowcasts with and without radar data assimilation have similar evolution trends driven by synoptic-scale conditions. Furthermore, we carry out a series of sensitivity experiments to develop proper assimilation strategies, in which a mixed localisation method is proposed for the first time and found to give further QPN improvement in this typhoon case.

Improving the Model Convective Storm Quantitative Precipitation Nowcasting by Assimilating State Variables Retrieved from Multiple-Doppler Radar Observations

Abstract This research combines an advanced multiple-Doppler radar synthesis technique with the thermodynamic retrieval method, originally proposed by Gal-Chen, and a moisture/temperature adjustment scheme, and formulates a sequential procedure. The focus is on applying this procedure to improve the model quantitative precipitation nowcasting (QPN) skill in the convective scale up to 3 hours. A series of (observing system simulation experiment) OSSE-type tests and a real case study are conducted to investigate the performance of this algorithm under different conditions. It is shown that by using the retrieved three-dimensional wind, thermodynamic, and microphysical parameters to reinitialize a fine-resolution numerical model, its QPN skill can be significantly improved. Since the Gal-Chen method requires the horizontal average properties of the weather system at each altitude, utilization of in situ radiosonde(s) to obtain this additional information for the retrieval is tested. When sounding data are not available, it is demonstrated that using the model results to replace the role played by observing devices is also a feasible choice. The moisture field is obtained through a simple, but effective, adjusting scheme and is found to be beneficial to the rainfall forecast within the first hour after the reinitialization of the model. Since this algorithm retrieves the unobserved state variables instantaneously from the wind measurements and directly uses them to reinitialize the model, fewer radar data and a shorter model spinup time are needed to correct the rainfall forecasts, in comparison with other data assimilation techniques such as four-dimensional variational data assimilation (4DVAR) or ensemble Kalman filter (EnKF) methods.

Probabilistic radar rainfall nowcasts using empirical and theoretical uncertainty models

Weather radar has a potential to provide accurate short-term (0–3 h) forecasts of rainfall (i.e. radar nowcasts), which are of great importance in warnings and risk management for hydro-meteorological events. However, radar nowcasts are affected by large uncertainties, which are not only linked to limitations in the forecast method but also because of errors in the radar rainfall measurement. The probabilistic quantitative precipitation nowcasting approach attempts to quantify these uncertainties by delivering the forecasts in a probabilistic form. This study implements two forms of probabilistic quantitative precipitation nowcasting for a hilly area in the south of Manchester, namely, the theoretically based scheme [ensemble rainfall forecasts (ERF)-TN] and the empirically based scheme (ERF-EM), and explores which one exhibits higher predictive skill. The ERF-TN scheme generates ensemble forecasts of rainfall in which each ensemble member is determined by the stochastic realisation of a theoretical noise component. The so-called ERF-EM scheme proposed and applied for the first time in this study, aims to use an empirically based error model to measure and quantify the combined effect of all the error sources in the radar rainfall forecasts. The essence of the error model is formulated into an empirical relation between the radar rainfall forecasts and the corresponding ‘ground truth’ represented by the rainfall field from rain gauges measurements. The ensemble members generated by the two schemes have been compared with the rain gauge rainfall. The hit rate and the false alarm rate statistics have been computed and combined into relative operating characteristic curves. The comparison of the performance scores for the two schemes shows that the ERF-EM achieves better performance than the ERF-TN at 1-h lead time. The predictive skills of both schemes are almost identical when the lead time increases to 2 h. In addition, the relation between uncertainty in the radar rainfall forecasts and lead time is also investigated by computing the dispersion of the generated ensemble members. Copyright © 2013 John Wiley & Sons, Ltd.

Application of multi-scale tracking radar echoes scheme in quantitative precipitation nowcasting

A new radar echo tracking algorithm known as multi-scale tracking radar echoes by cross-correlation (MTREC) was developed in this study to analyze movements of radar echoes at different spatial scales. Movement of radar echoes, particularly associated with convective storms, exhibits different characteristics at various spatial scales as a result of complex interactions among meteorological systems leading to the formation of convective storms. For the null echo region, the usual correlation technique produces zero or a very small magnitude of motion vectors. To mitigate these constraints, MTREC uses the tracking radar echoes by correlation (TREC) technique with a large “box” to determine the systematic movement driven by steering wind, and MTREC applies the TREC technique with a small “box” to estimate small-scale internal motion vectors. Eventually, the MTREC vectors are obtained by synthesizing the systematic motion and the small-scale internal motion. Performance of the MTREC technique was compared with TREC technique using case studies: the Khanun typhoon on 11 September 2005 observed by Wenzhou radar and a squall-line system on 23 June 2011 detected by Beijing radar. The results demonstrate that more spatially smoothed and continuous vector fields can be generated by the MTREC technique, which leads to improvements in tracking the entire radar reflectivity pattern. The new multi-scale tracking scheme was applied to study its impact on the performance of quantitative precipitation nowcasting. The location and intensity of heavy precipitation at a 1-h lead time was more consistent with quantitative precipitation estimates using radar and rain gauges.

A Bayesian state space modelling approach to probabilistic quantitative precipitation forecasting

The generation of very short range forecasts of precipitation in the 0–6 h time window is traditionally referred to as nowcasting. Most existing nowcasting systems essentially extrapolate radar observations in some manner, however, very few systems account for the uncertainties involved. Thus deterministic forecast are produced, which have a limited use when decisions must be made, since they have no measure of confidence or spread of the forecast. This paper develops a Bayesian state space modelling framework for quantitative precipitation nowcasting which is probabilistic from conception. The model treats the observations (radar) as noisy realisations of the underlying true precipitation process, recognising that this process can never be completely known, and thus must be represented probabilistically. In the model presented here the dynamics of the precipitation are dominated by advection, so this is a probabilistic extrapolation forecast. The model is designed in such a way as to minimise the computational burden, while maintaining a full, joint representation of the probability density function of the precipitation process. The update and evolution equations avoid the need to sample, thus only one model needs be run as opposed to the more traditional ensemble route. It is shown that the model works well on both simulated and real data, but that further work is required before the model can be used operationally.

The Nowcasting of Precipitation during Sydney 2000: An Appraisal of the QPF Algorithms

Statistical and case study–oriented comparisons of the quantitative precipitation nowcasting (QPN) schemes demonstrated during the first World Weather Research Programme (WWRP) Forecast Demonstration Project (FDP), held in Sydney, Australia, during 2000, served to confirm many of the earlier reported findings regarding QPN algorithm design and performance. With a few notable exceptions, nowcasting algorithms based upon the linear extrapolation of observed precipitation motion (Lagrangian persistence) were generally superior to more sophisticated, nonlinear nowcasting methods. Centroid trackers [Thunderstorm Identification, Tracking, Analysis and Nowcasting System (TITAN)] and pattern matching extrapolators using multiple vectors (Auto-nowcaster and Nimrod) were most reliable in convective scenarios. During widespread, stratiform rain events, the pattern matching extrapolators were superior to centroid trackers and wind advection techniques (Gandolf, Nimrod). There is some limited case study and statistical evidence from the FDP to support the use of more sophisticated, nonlinear QPN algorithms. In a companion paper in this issue, Wilson et al. demonstrate the advantages of combining linear extrapolation with algorithms designed to predict convective initiation, growth, and decay in the Auto-nowcaster. Ebert et al. show that the application of a nonlinear scheme [Spectral Prognosis (S-PROG)] designed to smooth precipitation features at a rate consistent with their observed temporal persistence tends to produce a nowcast that is superior to Lagrangian persistence in terms of rms error. However, the value of this approach in severe weather forecasting is called into question due to the rapid smoothing of high-intensity precipitation features.