Precipitation Nowcasting Research Articles

FDNet: A Deep Learning Approach with Two Parallel Cross Encoding Pathways for Precipitation Nowcasting

FDNet: A Deep Learning Approach with Two Parallel Cross Encoding Pathways for Precipitation Nowcasting

Value of Spatially Distributed Rainfall Design Events—Creating Basin-Scale Stochastic Design Storm Ensembles

Current design storms used in hydrological modeling, urban planning, and dimensioning of structures are typically point-scale rainfall events with a steady rainfall intensity or a simple temporal intensity pattern. This can lead to oversimplified results because real rainfall events have more complex patterns than simple design series. In addition, the interest of hydrologists is usually in areal estimates rather than point values, most commonly in river-basin-wide areal mean rainfall estimates. By utilizing weather radar data and the short-term ensemble prediction system pySTEPS, which has so far been used for precipitation nowcasting, ensembles of high-resolution stochastic design storms with desired statistical properties and spatial structure evolving in time are generated. pySTEPS is complemented by adding time-series models for areal average rainfall over the simulation domain and field advection vectors. The selected study area is the Kokemäenjoki river basin located in Western Finland, and the model parametrization is carried out utilizing the Finnish Meteorological Institute’s weather radar data from the years 2013 to 2016. The results demonstrate how simulated events with similar large-scale mean areal rainfall can produce drastically different total event rainfalls in smaller scales. The sampling method, areal vs. gauge estimate, is also shown to have a prominent effect on total event rainfall across different spatial scales. The outlined method paves the way towards a more thorough and wide-spread assessment of the hydrological impacts of spatiotemporal rainfall characteristics.

TAFFNet: Time‐Aware Adaptive Feature Fusion Network for Very Short‐Term Precipitation Forecasts

AbstractDeep learning models based on radar echo extrapolation have been widely used in precipitation nowcasting. However, they face the challenge of insufficient input information when extending the forecast lead time, requiring the incorporation of physics‐based numerical weather prediction (NWP). Given that the strengths of radar and NWP data vary depending on the forecast time, effectively fusing these two data sources in a unified deep learning model remains an open research problem. In this study, we propose a Time‐aware Adaptive Feature Fusion Network (TAFFNet) for very short‐term precipitation forecasts up to 12 hr. TAFFNet fuses features adaptively according to their relative contributions to forecast skill at different times. Experimental results demonstrate that TAFFNet performs the best for very short‐term precipitation forecasts. The case studies show that adaptively fusing NWP with radar can improve the accuracy of precipitation forecasts, especially for predicting the initiation and dissipation of storms at longer lead times.

Unsupervised Clustering of Geostationary Satellite Cloud Properties for Estimating Precipitation Probabilities of Tropical Convective Clouds

Abstract Understanding the growth of tropical convective clouds (TCCs) is of vital importance for the early detection of heavy rainfall. This study explores the properties of TCCs that can cause them to develop into clouds with a high probability of precipitation. Remotely sensed cloud properties, such as cloud-top temperature (CTT), cloud optical thickness (COT), and cloud effective radius (CER) as measured by a geostationary satellite are trained by a neural network. First, the image segmentation algorithm identifies TCC objects with different cloud properties. Second, a self-organizing map (SOM) algorithm clusters TCC objects with similar cloud microphysical properties. Third, the precipitation probability (PP) for each cluster of TCCs is calculated based on the proportion of precipitating TCCs among the total number of TCCs. Precipitating TCCs can be distinguished from nonprecipitating TCCs using Integrated Multi-Satellite Retrievals for Global Precipitation Measurement precipitation data. Results show that SOM clusters with a high PP (>70%) satisfy a certain range of cloud properties: CER ≥ 20 μm and CTT < 230 K. PP generally increases with increasing COT, but COT cannot be a clear cloud property to confirm a high PP. For relatively thin clouds (COT < 30), however, CER should be much larger than 20 μm to have a high PP. More importantly, these TCC conditions associated with a PP ≥ 70% are consistent across regions and periods. We expect our results will be useful for satellite nowcasting of tropical precipitation using geostationary satellite cloud properties. Significance Statement We aim to identify the properties of tropical convective clouds (TCCs) that have a high precipitation probability. We designed a two-step framework to identify TCC objects and the conditions of cloud properties for TCCs to have a high precipitation probability. The TCCs with a precipitation probability > 70% tend to have a low cloud-top temperature and a cloud particle effective radius ≥ 20 μm. Cloud optical thicknesses are distributed over a wide range, but thinning requires a particle radius larger than 20 μm. These conditions of cloud properties appear to be unchanged under various spatial–temporal conditions over the tropics. This important observational finding advances our understanding of the cloud–precipitation relationship in TCCs and can be applied to satellite nowcasting of precipitation in the tropics, where numerical weather forecasts are limited.

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.

Skilful nowcasting of extreme precipitation with NowcastNet

Extreme precipitation is a considerable contributor to meteorological disasters and there is a great need to mitigate its socioeconomic effects through skilful nowcasting that has high resolution, long lead times and local details1–3. Current methods are subject to blur, dissipation, intensity or location errors, with physics-based numerical methods struggling to capture pivotal chaotic dynamics such as convective initiation4 and data-driven learning methods failing to obey intrinsic physical laws such as advective conservation5. We present NowcastNet, a nonlinear nowcasting model for extreme precipitation that unifies physical-evolution schemes and conditional-learning methods into a neural-network framework with end-to-end forecast error optimization. On the basis of radar observations from the USA and China, our model produces physically plausible precipitation nowcasts with sharp multiscale patterns over regions of 2,048 km × 2,048 km and with lead times of up to 3 h. In a systematic evaluation by 62 professional meteorologists from across China, our model ranks first in 71% of cases against the leading methods. NowcastNet provides skilful forecasts at light-to-heavy rain rates, particularly for extreme-precipitation events accompanied by advective or convective processes that were previously considered intractable.

Comparison of Three Radar-Based Precipitation Nowcasts for the Extreme July 2021 Flooding Event in Germany

Abstract Quantitative precipitation nowcasts (QPN) can improve the accuracy of flood forecasts, especially for lead times up to 12 h, but their evaluation depends on a variety of factors, namely, the choice of the hydrological model and the benchmark. We tested three precipitation nowcasting techniques based on radar observations for the disastrous mid-July 2021 event in seven German catchments (140–1670 km2). Two deterministic [advection-based and spectral prognosis (S-PROG)] and one probabilistic [Short-Term Ensemble Prediction System (STEPS)] QPN with a maximum lead time of 3 h were used as input to two hydrological models: a physically based, 3D-distributed model (ParFlowCLM) and a conceptual, lumped model (GR4H). We quantified the hydrological added value of QPN compared with hydrological persistence and zero-precipitation nowcasts as benchmarks. For the 14 July 2021 event, we obtained the following key results. 1) According to the quality of the forecasted hydrographs, exploiting QPN improved the lead times by up to 4 h (8 h) compared with adopting zero-precipitation nowcasts (hydrological persistence) as a benchmark. Using a skill-based approach, obtained improvements were up to 7–12 h depending on the benchmark. 2) The three QPN techniques obtained similar performances regardless of the applied hydrological model. 3) Using zero-precipitation nowcasts instead of hydrological persistence as benchmark reduced the added value of QPN. These results highlight the need for combining a skill-based approach with an analysis of the quality of forecasted hydrographs to rigorously estimate the added value of QPN.

LSTMAtU-Net: A Precipitation Nowcasting Model Based on ECSA Module.

Precipitation nowcasting refers to the use of specific meteorological elements to predict precipitation in the next 0-2 h. Existing methods use radar echo maps and the Z-R relationship to directly predict future rainfall rates through deep learning methods, which are not physically constrained, but suffer from severe loss of predicted image details. This paper proposes a new model framework to effectively solve this problem, namely LSTMAtU-Net. It is based on the U-Net architecture, equipped with a Convolutional LSTM (ConvLSTM) unit with the vertical flow direction and depthwise-separable convolution, and we propose a new component, the Efficient Channel and Space Attention (ECSA) module. The ConvLSTM unit with the vertical flow direction memorizes temporal changes by extracting features from different levels of the convolutional layers, while the ECSA module innovatively integrates different structural information of each layer of U-Net into the channelwise attention mechanism to learn channel and spatial information, thereby enhancing attention to the details of precipitation images. The experimental results showed that the performance of the model on the test dataset was better than other examined models and improved the accuracy of medium- and high-intensity precipitation nowcasting.

Weather Radar Echo Extrapolation with Dynamic Weight Loss

Precipitation nowcasting is an important tool for economic and social services, especially for forecasting severe weather. The crucial and challenging part of radar echo image prediction is the focus of radar-based precipitation nowcasting. Recently, a number of deep learning models have been designed to solve the problem of extrapolating radar images. Although these methods can generate better results than traditional extrapolation methods, the issue of error accumulation in precipitation forecasting is exacerbated by using only the mean square error (MSE) and mean absolute error (MAE) as loss functions. In this paper, we approach the problem from the perspective of the loss function and propose dynamic weight loss (DWL), a simple but effective loss function for radar echo extrapolation. The method adds model self-adjusted dynamic weights to the weighted loss function and structural similarity index measures. Radar echo extrapolation experiments are performed on four models, ConvLSTM, ConvGRU, PredRNN, and PredRNN++. Radar reflectivity is predicted using Nanjing University C-band Polarimetric (NJU-CPOL) weather radar data. The quantitative statistics show that using the DWL method reduces the MAE of the four models by up to 10.61%, 5.31%, 14.8%, and 13.63%, respectively, over a 1 h prediction period. The results show that the DWL approach is effective in reducing the accumulation of errors over time, improving the predictive performance of currently popular deep learning models.

CLGAN: a generative adversarial network (GAN)-based video prediction model for precipitation nowcasting

Abstract. The prediction of precipitation patterns up to 2 h ahead, also known as precipitation nowcasting, at high spatiotemporal resolutions is of great relevance in weather-dependent decision-making and early warning systems. In this study, we are aiming to provide an efficient and easy-to-understand deep neural network – CLGAN (convolutional long short-term memory generative adversarial network) – to improve the nowcasting skills of heavy precipitation events. The model constitutes a generative adversarial network (GAN) architecture, whose generator is built upon a u-shaped encoder–decoder network (U-Net) and is equipped with recurrent long short-term memory (LSTM) cells to capture spatiotemporal features. The optical flow model DenseRotation and the competitive video prediction models ConvLSTM (convolutional LSTM) and PredRNN-v2 (predictive recurrent neural network version 2) are used as the competitors. A series of evaluation metrics, including the root mean square error, the critical success index, the fractions skill score, and object-based diagnostic evaluation, are utilized for a comprehensive comparison against competing baseline models. We show that CLGAN outperforms the competitors in terms of scores for dichotomous events and object-based diagnostics. A sensitivity analysis on the weight of the GAN component indicates that the GAN-based architecture helps to capture heavy precipitation events. The results encourage future work based on the proposed CLGAN architecture to improve the precipitation nowcasting and early warning systems.

Precipitation Nowcasting Based on Deep Learning over Guizhou, China

Accurate precipitation nowcasting (lead time: 0–2 h), which requires high spatiotemporal resolution data, is of great relevance in many weather-dependent social and operational activities. In this study, we are aiming to construct highly accurate deep learning (DL) models to directly obtain precipitation nowcasting at 6-min intervals for the lead time of 0–2 h. The Convolutional Long Short-Term Memory (ConvLSTM) and Predictive Recurrent Neural Network (PredRNN) models were used as comparative DL models, and the Lucas–Kanade (LK) Optical Flow method was selected as a traditional extrapolation baseline. The models were trained with high-quality datasets (resolution: 1 min) created from precipitation observations recorded by automatic weather stations in Guizhou Province (China). A comprehensive evaluation of the precipitation nowcasting was performed, which included consideration of the root mean square error, equitable threat score (ETS), and probability of detection (POD). The evaluation indicated that the reduction of the number of missing values and data normalization boosted training efficiency and improved the forecasting skill of the DL models. Increasing the time series length of the training set and the number of training samples both improved the POD and ETS of the DL models and enhanced nowcasting stability with time. Training with the Hea-P dataset further improved the forecasting skill of the DL models and sharply increased the ETS for thresholds of 2.5, 8, and 15 mm, especially for the 1-h lead time. The PredRNN model trained with the Hea-P dataset (time series length: 8 years) outperformed the traditional LK Optical Flow method for all thresholds (0.1, 1, 2.5, 8, and 15 mm) and obtained the best performance of all the models considered in this study in terms of ETS. Moreover, the Method for Object-Based Diagnostic Evaluation on a rainstorm case revealed that the PredRNN model, trained well with high-quality observation data, could both capture complex nonlinear characteristics of precipitation more accurately than achievable using the LK Optical Flow method and establish a reasonable mapping network during drastic changes in precipitation. Thus, its results more closely matched the observations, and its forecasting skill for thresholds exceeding 8 mm was improved substantially.

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.

Mutual Information Boosted Precipitation Nowcasting from Radar Images

Precipitation nowcasting has long been a challenging problem in meteorology. While recent studies have introduced deep neural networks into this area and achieved promising results, these models still struggle with the rapid evolution of rainfall and extremely imbalanced data distribution, resulting in poor forecasting performance for convective scenarios. In this article, we evaluate the amount of information in different precipitation nowcasting tasks of varying lengths using mutual information. We propose two strategies: the mutual information-based reweighting strategy (MIR) and a mutual information-based training strategy (time superimposing strategy (TSS)). MIR reinforces neural network models to improve the forecasting accuracy for convective scenarios while maintaining prediction performance for rainless scenarios and overall nowcasting image quality. The TSS strategy enhances the model’s forecasting performance by adopting a curriculum learning-like method. Although the proposed strategies are simple, the experimental results show that they are effective and can be applied to various state-of-the-art models.

IMPROVER: The New Probabilistic Postprocessing System at the Met Office

Abstract The Met Office in the United Kingdom has developed a completely new probabilistic postprocessing system called IMPROVER to operate on outputs from its operational numerical weather prediction (NWP) forecasts and precipitation nowcasts. The aim is to improve weather forecast information to the public and other stakeholders while better exploiting the current and future generations of underpinning kilometer-scale NWP ensembles. We wish to provide seamless forecasts from nowcasting to medium range, provide consistency between gridded and site-specific forecasts, and be able to verify every stage of the processing. The software is written in a modern modular framework that is easy to maintain, develop, and share. IMPROVER allows forecast information to be provided with greater spatial and temporal detail and a faster update frequency than previous postprocessing. Independent probabilistic processing chains are constructed for each meteorological variable consisting of a series of processing stages that operate on predefined grids and blend outputs from several NWP inputs to give a frequently updated, probabilistic forecast solution. Probabilistic information is produced as standard, with the option of extracting a most likely or yes–no outcome if required. Verification can be performed at all stages, although it is only currently switched on for the most significant stages when run in real time. IMPROVER has been producing real-time output since March 2021 and became operational in spring 2022.

Data for paper "Quality Aware Conditional Generative Adversarial Networks for Precipitation Nowcasting" submitted to AAI.

Data for paper "Quality Aware Conditional Generative Adversarial Networks for Precipitation Nowcasting" submitted to AAI.

MM-RNN: A Multimodal RNN for Precipitation Nowcasting

Precipitation nowcasting, the high-resolution forecasting of precipitation in a short term, is essential in various applications in the real world. Previous deep learning methods use huge samples to learn potential laws, and the learning process lacks regularity, making it difficult to model the complex nonlinear precipitation phenomenon. Inspired by traditional numerical weather prediction models, we propose the MultiModal RNN (MM-RNN), which introduces knowledge of elements to guide precipitation prediction. This constraint forces the movement of precipitation to follow the underlying atmospheric motion laws. MM-RNN not only can provide accurate precipitation nowcasting but other meteorological elements predictions. Besides, it has high flexibility and is compatible with multiple RNN models, such as ConvLSTM, PredRNN, MIM, MotionRNN, etc. We conduct experiments on two multimodal datasets (MeteoNet and RAIN-F) and the results indicate that MM-RNN is superior to common RNN (MultiScale RNN, MS-RNN) using a single radar modality. For the MeteoNet, compared to MS-MotionRNN, the CSI (R ⩾ 10) of MM-MotionRNN increases by 23.4%, and the MSE of MM-MotionRNN decreases by 6.7%. For the RAIN-F, compared to MS-MIM, the HSS (R ⩾ 5) of MM-MIM increases by 209.4%, and the B-MSE of MM-MIM decreases by 4.6%.

CEMA-LSTM: Enhancing Contextual Feature Correlation for Radar Extrapolation Using Fine-Grained Echo Datasets

Accurate precipitation nowcasting can provide great convenience to the public so they can conduct corresponding arrangements in advance to deal with the possible impact of upcoming heavy rain. Recent relevant research activities have shown their concerns on various deep learning models for radar echo extrapolation, where radar echo maps were used to predict their consequent moment, so as to recognize potential severe convective weather events. However, these approaches suffer from an inaccurate prediction of echo dynamics and unreliable depiction of echo aggregation or dissipation, due to the size limitation of convolution filter, lack of global feature, and less attention to features from previous states. To address the problems, this paper proposes a CEMA-LSTM recurrent unit, which is embedded with a Contextual Feature Correlation Enhancement Block (CEB) and a Multi-Attention Mechanism Block (MAB). The CEB enhances contextual feature correlation and supports its model to memorize significant features for near-future prediction; the MAB uses a position and channel attention mechanism to capture global features of radar echoes. Two practical radar echo datasets were used involving the FREM and CIKM 2017 datasets. Both quantification and visualization of comparative experimental results have demonstrated outperformance of the proposed CEMA-LSTM over recent models, e.g., PhyDNet, MIM and PredRNN++, etc. In particular, compared with the second-ranked model, its average POD, FAR and CSI have been improved by 3.87%, 1.65% and 1.79%, respectively on the FREM, and by 1.42%, 5.60% and 3.16%, respectively on the CIKM 2017.

STUNNER: Radar Echo Extrapolation Model Based on Spatiotemporal Fusion Neural Network

Radar echo extrapolation based on deep learning is an important method for conducting precipitation nowcasting. Radar echo sequence data have spatio-temporal correlations and non-rigid movements of the radar echo. According to the characteristics of radar data, this study proposes a new spatio-temporal fusion neural network called STUNNER. STUNNER implements a two-stream spatio-temporal fusion strategy to extract and fuse spatial and temporal signals. Specifically, it uses a novel cross-network embedding method to achieve efficient spatio-temporal fusion; the fusion integrates a temporal differencing network and a spatio-temporal trajectory network. The temporal differencing network models the high-order non-stationarity of the radar sequence to learn the motion trend. The spatio-temporal trajectory network optimizes the convolution operator in a spatio-temporal long short-term memory model to extract the transient variations from the radar images. We compare STUNNER with other five models on two public datasets. On the radar echo extrapolation task, STUNNER achieves the best performance.

WF-UNet: Weather Data Fusion using 3D-UNet for Precipitation Nowcasting

Designing early warning systems for harsh weather and its effects, such as urban flooding or landslides, requires accurate short-term forecasts (nowcasts) of precipitation. Nowcasting is a significant task with several environmental applications, such as agricultural management or increasing flight safety. In this study, we investigate the use of a UNet core model and its extension for precipitation nowcasting in western Europe for up to 3 hours ahead. In particular, we propose the Weather Fusion UNet (WF-UNet) model, which utilizes the Core 3D-UNet model and integrates precipitation and wind speed variables as input in the learning process and analyzes its influences on the precipitation target task. We have collected six years of precipitation and wind radar images from Jan 2016 to Dec 2021 of 14 European countries, with 1-hour temporal resolution and 31 square km spatial resolution based on the ERA5 dataset, provided by Copernicus, the European Union's Earth observation programme. We compare the proposed WF-UNet model to the persistence model as well as other UNet-based architectures that are trained only using precipitation radar input data. The obtained results show that WF-UNet outperforms the other examined best-performing architectures by 22%, 8% and 3% lower MSE at a horizon of 1, 2 and 3 hours respectively.

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.