Precipitation Forecasts Research Articles

Prediction of Summer Precipitation via Machine Learning with Key Climate Variables：A Case Study in Xinjiang, China

Study region: Xinjiang is located in the mid-latitude region of Eurasia in northwestern China. Precipitation is predominantly concentrated in northern Xinjiang, while southern Xinjiang remains comparatively arid. Summer precipitation accounts for 54.4% of the annual total. Study focus: This study aims to develop a machine learning model to predict summer precipitation (June–August) in XJ and explore the key variables contributing to summer precipitation in this region. The SHapley Additive exPlanations method was integrated with an extreme tree model to quantify the contributions of variables towards precipitation. Artificial neural networks, support vector machines, and extreme gradient boosting were considered to predict summer precipitation. To train the ML model, we used precipitation data from 1961 to 2012, whilst the forecast results from 2013 to 2017 were used for validation.New hydrological insights for the regions: The results demonstrated that the ANN model achieved robust performance during both the training and validation periods. For Northern and Southern XJ, the Mean Absolute Error and Root Mean Square Error of the ANN model were 15.34 (20.40) and 23.21 (30.01), respectively. The SHAP analysis showed that in the context of Northern Xinjiang, the Niño B Sea Surface Temperature Anomaly, Western Pacific Subtropical High Intensity, Pacific Subtropical High Intensity, and Multivariate ENSO Index play crucial roles in the prediction of summer precipitation. In Southern Xinjiang, the South China Sea Subtropical High Intensity, South China Sea Subtropical High Area, Western Pacific Warm Pool Strength, and Atlantic multidecadal oscillation have emerged as key variables affecting summer precipitation forecasting.

Deep Learning Integration of Multi-Model Forecast Precipitation Considering Long Lead Times

Reliable forecast precipitation can support disaster prevention and mitigation and sustainable socio-economic development. Improving forecast precipitation accuracy remains a challenge. Therefore, a novel method for multi-model forecast precipitation integration considering long lead times was proposed based on deep learning. First, the accuracy of numerical forecast precipitation was evaluated under different lead times. Secondly, an integrated model was built by coupling the attention mechanism and a long short-term memory neural network (LSTM). Finally, integrated forecast precipitation was obtained by taking high-precision numerical forecast precipitation as an input and examining its accuracy and applicability. Considering the example of the Yalong River, the results showed the following: (1) numerical forecast precipitation fails to forecast precipitation of a ≥10 mm/d intensity well, and is less applicable in streamflow forecast; (2) traditional machine learning methods for integrating multi-model forecast precipitation fail to forecast precipitation of a ≥25 mm/d intensity; (3) the LSTM-A integration model formed by attention weighting after the LSTM output can combine the advantages of numerical forecast precipitation under different intensities and improve the forecast precipitation accuracy for 7-day lead times; and (4) the LSTM-A integrated forecast precipitation has the best applicability in streamflow forecast, with an NSE above 0.82 and an MRE below 30% with 7-day lead times. These findings contribute to improving precipitation forecast accuracy at different intensities and enhancing defense against extreme weather events.

The Direct Assimilation of Radar Reflectivity Data with a Two-Moment Microphysics Scheme for a Landfalling Typhoon in an OSSE Framework

Despite the well-known importance of radar data assimilation, there are limited studies on landfalling typhoons in terms of directly assimilating radar reflectivity data, especially using a reflectivity operator based on double-moment microphysics. In this study, radar reflectivity data assimilation experiments are conducted with an ensemble Kalman filter (EnKF), using simulated observations in an Observing System Simulation Experiment (OSSE) framework for the landfalling typhoon In-Fa. With an OSSE, it is convenient to analyze the impact of assimilation of radar reflectivity on analysis and forecast for various state variables, especially for hydrometeors. Our results show that the direct assimilation of radar reflectivity with EnKF does not introduce non-physical hydrometeors and is able to adjust well, not only to hydrometers, but also to some large-scale variables which are not directly related to reflectivity, especially in terms of temperature and vertical velocity. Though the most notable reduction in the Root Mean Square Errors (RMSEs) is observed through mixing the ratio of rainwater and snow, the analysis of other variables is also significantly improved with the accumulation of assimilation cycles. The correlation analysis reveals the strongest correlation between radar reflectivity data and hydrometeor-related variables as well as the correlation with certain large-scale variables, indicating that these cross-variables are updated well through the reliable multivariate ensemble covariance in the EnKF. As a result, an obvious improvement in typhoon intensity and precipitation forecast is obtained in the data assimilation experiment. The impact of assimilation on radar reflectivity can last for up to 15–16 h.

Tropical Interbasin Interaction as Effective Predictors of Late-Spring Precipitation Variability in the Southern Great Plains

Abstract The southern Great Plains experience fluctuating precipitation extremes that significantly impact agriculture and water management. Despite ongoing efforts to enhance forecast accuracy, the underlying causes of these climatic phenomena remain inadequately understood. This study elucidates the relative influence of the tropical Pacific and Atlantic basins on April–May–June precipitation variability in this region. Our partial ocean assimilation experiments using the Community Earth System Model unveil the prominent role of interbasin interaction, with the Pacific and Atlantic contributing approximately 70% and 30%, respectively, to these interbasin contrasts. Our statistical analyses suggest that these tropical interbasin contrasts could serve as a more reliable indicator for late-spring precipitation anomalies than El Niño–Southern Oscillation. The conclusions are reinforced by analyses of seven climate forecasting systems within the North American Multi-Model Ensemble, offering an optimistic outlook for enhancing real-time forecasting of late-spring precipitation in the southern plains. However, the current predictive skills of the interbasin contrasts across the prediction systems are hindered by the lower predictability of the tropical Atlantic Ocean, pointing to the need for future research to refine climate prediction models further. Significance Statement Agriculture and infrastructure in the southern plains face challenges from severe late-spring precipitation extremes. Traditional predictors like El Niño–Southern Oscillation (ENSO) lose effectiveness during the critical spring-to-summer transition, creating a forecasting gap. This study introduces the concept of tropical interbasin interactions, known to enhance seasonal predictability for late-spring precipitation in the southern plains. Novel climate model experiments highlight contributions from the tropical Pacific and Atlantic, offering a promising predictability that potentially surpasses the limitations of ENSO-based predictions. These outcomes hold the potential for developing operational forecasts of late-spring precipitation anomalies in the southern plains, enabling proactive risk management.

Research on National Short-Term Precipitation Forecast Method Based on Machine Learning

Abstract. Climate change is closely related to human lives. With the development of information technology, meteorological data gradually show the characteristics of big data. The purpose of this study is to compare the application effects of different machine learning models in short-term precipitation prediction, to improve the accuracy and efficiency of prediction. Based on the annual average precipitation data of prefecture-level cities in China from 1990 to 2022, this paper uses Random Forest, eXtreme Gradient Boosting and Neural Networks to construct prediction models, and comprehensively evaluate them. In this study, the data is preprocessed to ensure the data quality of the input model. Next, the data is fed into three machine learning algorithm models, Random Forest, eXtreme Gradient Boosting and Neural Networks. Finally, the prediction performance of each model is evaluated by various indexes. In this paper, it is found that the Random Forest model has the best performance, and its prediction accuracy is higher than the other two models, which has great application potential in the field of short-term precipitation prediction. This study shows that a reasonable selection of machine learning methods and optimization of model parameters can effectively improve the accuracy of short-term precipitation prediction. This paper provides some empirical evidence for precipitation prediction, which will help to make more effective decisions in dealing with extreme weather events and climate change challenges in the future.

Research on National Short-Term Precipitation Forecast Method Based on Machine Learning

Abstract. Climate change is closely related to human lives. With the development of information technology, meteorological data gradually show the characteristics of big data. The purpose of this study is to compare the application effects of different machine learning models in short-term precipitation prediction, to improve the accuracy and efficiency of prediction. Based on the annual average precipitation data of prefecture-level cities in China from 1990 to 2022, this paper uses Random Forest, eXtreme Gradient Boosting and Neural Networks to construct prediction models, and comprehensively evaluate them. In this study, the data is preprocessed to ensure the data quality of the input model. Next, the data is fed into three machine learning algorithm models, Random Forest, eXtreme Gradient Boosting and Neural Networks. Finally, the prediction performance of each model is evaluated by various indexes. In this paper, it is found that the Random Forest model has the best performance, and its prediction accuracy is higher than the other two models, which has great application potential in the field of short-term precipitation prediction. This study shows that a reasonable selection of machine learning methods and optimization of model parameters can effectively improve the accuracy of short-term precipitation prediction. This paper provides some empirical evidence for precipitation prediction, which will help to make more effective decisions in dealing with extreme weather events and climate change challenges in the future.

Statistical-dynamical analog ensemble system for real time quantitative precipitation forecasts (QPFs) at local scale in the north-west Himalaya (NWH), India

Statistical-dynamical analog ensemble system for real time quantitative precipitation forecasts (QPFs) at local scale in the north-west Himalaya (NWH), India

SAL Method Applied in Grid Forecasting Product Verification with Three-Source Fusion Product

Quantitative precipitation forecast (QPF) verification stands out as one of the most formidable endeavors in the realm of forecast verification. Traditional verification methods are not suitable for high-resolution forecasting products in some cases. Therefore, the SAL (structure, amplitude and location) method was proposed as a method of object-based spatial verification that studies precipitation verification in a certain range, which is combined with factors including structure, amplitude and location of the targets. However, the setting of the precipitation threshold would affect the result of the verification. This paper presented an improved method for determining the precipitation threshold using the QPF from ECMWF, which is an ensemble forecast model and three-source fusion product that was used in China from 1 July to 31 August 2020, and then the results obtained with this method were compared with the other two traditional methods. Furthermore, the SAL and the traditional verification methods were carried out for geometric, simulated and real cases, respectively. The results showed the following: (1) The proposed method in this paper for determining the threshold was more accurate at identifying the precipitation objects. (2) The verification area size was critical for SAL calculation. If the area selected was too large, the calculated SAL value had little significance. (3) ME (Mean Error) could not identify the displacement between prediction and observation, while HSS (Heidke Skill Score) was sensitive to the displacement of the prediction field. (4) Compared with the traditional verification methods, the SAL method was more straight forward and simple, and it could give a better representation of prediction ability. Therefore, forecasters can better understand the model prediction effect and what needs to be improved.

Multi-season evaluation of hurricane analysis and forecast system (HAFS) quantitative precipitation forecasts

Quantitative precipitation forecasts (QPF) from numerical weather prediction models need systematic verification to enable rigorous assessment and informed use, as well as model improvements. The United States (US) National Oceanic and Atmospheric Administration (NOAA) recently made a major update to its regional tropical cyclone modeling capabilities, introducing two new operational configurations of the Hurricane Analysis and Forecast System (HAFS). NOAA performed multi-season retrospective forecasts using the HAFS configurations during the period that the Hurricane Weather and Forecasting (HWRF) model was operational, which was used to assess HAFS performance for key tropical cyclone forecast metrics. However, systematic QPF verification was not an integral part of the initial evaluation. The first systematic QPF evaluation of the operational HAFS version 1 configurations is presented here for the 2021 and 2022 season re-forecasts as well as the first HAFS operational season, 2023. A suite of techniques, tools, and metrics within the enhanced Model Evaluation Tools (METplus) software suite are used. This includes shifting forecasts to mitigate track errors, regridding model and observed fields to a storm relative coordinate system, as well as object oriented verification. The HAFS configurations have better performance than HWRF for equitable threat score (ETS), but larger over forecast biases than HWRF. Storm relative and object oriented verification show the HAFS configurations have larger precipitation areas and less intense precipitation near the TC center as compared to observations and HWRF. HAFS QPF performance decreased for the 2023 season, but the general spatial patterns of the model QPF were very similar to 2021-2022.

Enhancing the Rainfall Forecasting Accuracy of Ensemble Numerical Prediction Systems via Convolutional Neural Networks

Abstract Ensemble prediction systems are commonly used to demonstrate the potential uncertainty of weather forecasts. These systems also help provide weather predictions to prepare for disasters. Specifically, a type of prediction called a quantitative precipitation forecast (QPF) is calculated from the average of multiple forecasts. The probability-matched mean (PMM) method is used to improve these QPF predictions when they underestimate rainfall. However, the PMM method can exhibit limitations when dealing with certain types of rain patterns. To address this issue, this study focuses on improving short-term heavy rainfall predictions using an artificial intelligence (AI) algorithm that combines deep neural networks (DNNs), including convolutional neural networks (CNNs), with a space-based attention mechanism (convolutional block attention module [CBAM]). Four experiments were conducted to evaluate the new method, including those performed to optimize the timing of a 24-hour QPF model, adjust the training dataset, and refine the training algorithm. Two specific weather events were assessed: rainfall influenced by southwest winds after typhoons and afternoon thunderstorm rainfall. A comparison of the results obtained via the root mean square error (RMSE) and structural similarity index measure (SSIM) reveals that the accuracy and distribution of the QPF predictions significantly improved over those of the PMM method. Compared with the PMM method, our approach can reduce the RMSE from 77.51 to 19.52 in the Mei-yu case, an improvement of approximately 74.82%. The SSIM also increased from 0.33 to 0.56, indicating a 70.9% enhancement. Overall, the new approach successfully enhances rainfall prediction accuracy via AI techniques and has the potential to be applied in disaster preparedness operations.

Spatial Aligned Mean: A Method to Improve Consensus Forecasts of Precipitation from Convection-Allowing Model Ensembles

Abstract Ensembles of convection-allowing model (CAM) forecasts are increasingly being used in operational numerical weather forecasting. Several approaches have been devised to find consensus among ensemble forecast fields, including the arithmetic ensemble mean and, more recently, the patchwise localized probability-matched (LPM) mean. However, differences in spatial distribution and intensity of precipitation features among ensemble members make it difficult to construct an ensemble mean product that characterizes the consensus while preserving precipitation structures forecasted by the individual ensemble members. To overcome this problem, this study aims to develop and test a method for improving ensemble consensus precipitation forecasts by directly considering the spatial offsets among ensemble members. This study uses a multiscale spatial alignment technique to align the precipitation features of each ensemble member to a common location, and the spatial aligned mean (SAM) is obtained by averaging the realigned members. It is shown that implementing SAM and subsequently applying the LPM technique to the average of all aligned members (SAM–LPM) can significantly improve the warm season precipitation forecast scores using common metrics such as equitable threat score (ETS). Also, improvement in the structure of features of heavy rainfall is shown from summer 2023 flash-flooding cases. Thus, SAM and SAM–LPM can be excellent candidate methods for calculating an ensemble consensus and providing ensemble consensus guidance to forecasters. Significance Statement High-impact rainfall events, such as flash floods, result in many billion-dollar loss events in the United States each year. This study seeks to improve the prediction of such events when using guidance from convection-allowing model (CAM) ensemble forecasts, such as the U.S. operational High-Resolution Ensemble Forecast (HREF) and the nascent Rapid Refresh Forecast System (RRFS). The proposed method, the spatial aligned mean (SAM), directly addresses the common issue of disparity in the predicted location of convective systems among ensemble members that confounds traditional ensemble consensus methods. In this study, it is found that SAM improves ensemble consensus guidance for high-impact rainfall events in both the HREF and the Center for Analysis and Prediction of Storms (CAPS) Finite-Volume Cubed-Sphere (FV3)-limited area model (LAM) CAM ensemble forecast system, a proxy for the future RRFS.

Cold-Frontal Rainbands over the United Kingdom: Sensitivity of Precipitation Processes to Microphysics Schemes in WRF Simulations

Abstract A narrow cold-frontal rainband (NCFR) and wide cold-frontal rainband (WCFR) over the United Kingdom on 24 January 2018 are selected to evaluate the sensitivity of cold-season precipitation to three bulk cloud microphysics schemes in the Weather Research and Forecasting (WRF) Model. The three simulations tend to overpredict the intensity and the areal coverage of NCFRs but underpredict those of WCFRs against the Met Office radar imagery. Nonetheless, the WRF double-moment 6-class (WDM6) simulation produces the most realistic NCFR and WCFR. More specifically, the intensity of the NCFR in the WDM6 simulation is overpredicted in the developing stage and underpredicted in the dissipating stage when compared to the Thompson and Morrison simulations. This phenomenon is mainly attributed to the rain mixing ratio at 1 km. The more intense and larger area of the WCFR in the WDM6 simulation is associated with more low-level snow, graupel, and rain than the other two simulations. In the WDM6 simulation, the larger low-level snow mixing ratios are due to the higher rates of aggregation between cloud ice and snow, whereas the larger rain mixing ratio is due to higher melting rates. Thus, differences in low-level ice-phase processes between microphysics schemes are large, strongly influencing the intensity of WCFRs, but the differences in warm-rain processes are smaller. Additionally, the small size of raindrops in the WDM6 simulation leads to wider and more intense WCFR as these raindrops are preferentially sent rearward relative to the cold front. Significance Statement Models are useful tools to predict weather but have errors in predicting different weather types. In the United Kingdom, precipitation is often predicted to be too strong at cold fronts and too weak behind the fronts compared to radar observations. Choosing different options for how precipitation is represented in one model produced different precipitation forecasts, with WDM6 being the best. One reason why WDM6 was the best was because it produced more and smaller raindrops, which would be transported behind the front, and it had higher melting rates at low levels, resulting in more precipitation behind the fronts than other options. Our results help identify why different options produce different forecasts, leading to possible future improvements in these options.

Impact‐Based Skill Evaluation of Seasonal Precipitation Forecasts

AbstractWe introduce an impact‐based framework to evaluate seasonal forecast model skill in capturing extreme weather and climate events over regions prone to natural disasters such as floods and wildfires. Forecasting hydroclimatic extremes holds significant importance in an era of increasing hazards such as wildfires, floods, and droughts. We evaluate the performance of five Copernicus Climate Change Service (C3S) seasonal forecast models (CMCC, DWD, ECCC, UK‐Met, and Météo‐France) in predicting extreme precipitation events from 1993 to 2016 using 14 indices reflecting timing and intensity (using absolute and locally defined thresholds) of precipitation at a seasonal timescale. Performance metrics, including Percent Bias, Kendall Tau Rank Correlation Score, and models' discrimination capacity, are used for skill evaluation. Our findings indicate that the performance of models varies markedly across regions and seasons. While models generally show good skill in the tropical regions, their skill in extra‐tropical regions is markedly lower. Elevated precipitation thresholds (i.e., higher intensity indices) correlate with heightened model biases, indicating deficiencies in modeling severe precipitation events. Our analysis using an impact‐based framework highlights the superior predictive capabilities of the UK‐Met and Météo‐France models in capturing the underlying processes that drive precipitation events, or lack thereof, across many regions and seasons. Other models exhibit strong performance in specific regions and/or seasons, but not globally. These results advance our understanding of an impact‐based framework in capturing a broad spectrum of extreme weather and climatic events, and inform strategic amalgamation of diverse models across different regions and seasons, thereby offering valuable insights for disaster management and risk analysis.

RainForests: A Machine Learning Approach to Calibrating NWP Precipitation Forecasts

Abstract Probabilistic forecasts derived from ensemble prediction systems (EPSs) have become the standard basis for many products and services produced by modern operational forecasting centers. However, statistical postprocessing is generally required to ensure forecasts have the desired properties expected for probability-based outputs. Precipitation, a core component of any operational forecast, is particularly challenging to calibrate due to its discontinuous nature and the extreme skew in rainfall amounts. A skillful forecasting system must maintain accuracy for low-to-moderate precipitation amounts, but preserve resolvability for high-to-extreme rainfall amounts, which, though rare, are important to forecast accurately in the interest of public safety. Existing statistical and machine learning approaches to rainfall calibration address this problem, but each has drawbacks in design, training approaches, and/or performance. We describe RainForests, a machine learning approach for calibrating ensemble rainfall forecasts using gradient-boosted decision trees. The model is based on the ecPoint system recently developed at ECMWF by Hewson and Pillosu, but uses machine learning models in place of the semisubjective decision trees of ecPoint, along with some other improvements to the model structure. We evaluate RainForests on the Australian domain against some simple benchmarks and show that it outperforms standard calibration approaches both in overall skill and in accurately forecasting high rainfall conditions, while being computationally efficient enough to be used in an operational forecasting system. Significance Statement Weather forecasting typically uses physical models to simulate the atmosphere and produce a set of scenarios, called an ensemble, for each forecast time. Forecast calibration is a set of statistical techniques for using these ensemble forecasts to obtain probability forecasts of weather variables. For example, on a given day, we may predict that there are a 90% chance of at least 1 mm of rain and a 50% chance of at least 3 mm. We introduce RainForests, a new machine learning method for calibrating ensemble forecasts, which uses ensemble forecasts of rainfall and other related variables to produce more accurate probability forecasts of rainfall amounts in operational forecasting settings.

Scale‐ and Variable‐Dependent Localization for 3DEnVar Data Assimilation in the Rapid Refresh Forecast System

AbstractThis study demonstrates the advantages of scale‐ and variable‐dependent localization (SDL and VDL) on three‐dimensional ensemble variational data assimilation of the hourly‐updated high‐resolution regional forecast system, the Rapid Refresh Forecast System (RRFS). SDL and VDL apply different localization radii for each spatial scale and variable, respectively, by extended control vectors. Single‐observation assimilation tests and cycling experiments with RRFS indicated that SDL can enlarge the localization radius without increasing the sampling error caused by the small ensemble size and decreased associated imbalance of the analysis field, which was effective at decreasing the bias of temperature and humidity forecasts. Moreover, simultaneous assimilation of conventional and radar reflectivity data with VDL, where a smaller localization radius was applied only for hydrometeors and vertical wind, improved precipitation forecasts without introducing noisy analysis increments. Statistical verification showed that these impacts contributed to forecast error reduction, especially for low‐level temperature and heavy precipitation.

Optimizing Precipitation Forecasting and Agricultural Water Resource Allocation Using the Gaussian-Stacked-LSTM Model

Our study investigates the use of machine learning models for daily precipitation prediction using data from 56 meteorological stations in Jilin Province, China. We evaluate Stacked Long Short-Term Memory (LSTM), Transformer, and Support Vector Regression (SVR) models, with Stacked-LSTM showing the best performance in terms of accuracy and stability, as measured by the Root Mean Square Error (RMSE). To improve robustness, Gaussian noise was introduced, particularly enhancing predictions for zero-precipitation days. Key predictors identified through variable attribution analysis include temperature, dew point, prior precipitation, and air pressure. Additionally, we demonstrate the practical benefits of precipitation forecasts in optimizing water resource allocation. A prediction-based strategy outperforms equal distribution in managing resources efficiently, as shown in a case study using 2022 Beidahu data. Overall, our research advances precipitation forecasting through deep learning and offers valuable insights for water resource management.

Temperature and precipitation projection in the lower Mahanadi Basin through machine learning methods

This study examined climate change dynamics in the lower Mahanadi River basin by integrating observed and climate model data. Historical precipitation and temperature data (1979–2020) from the India Meteorological Department (IMD) and monthly climate model data from the CORDEX-SMHI-MIROC model via the Earth System Grid Federation (ESGF) are utilized. Four machine learning models (Fbprophet, Holt-Winters, LSTM RNN, and SARIMAX) are applied to forecast precipitation, Tmax, and Tmin, and are compared across different representative concentration pathway (RCP 2.6, 4.5, and 8.5) scenarios. Diverse trajectories emerge, highlighting potential shifts in precipitation and temperature dynamics over near, mid, and far-term intervals. Fbprophet and SARIMAX are identified as superior models through performance evaluation metrics (R2, RMSE, r, P-bias, and NSE). Spatial analysis using ArcGIS and IDW interpolation reveals spatial variations in climate projections, aiding in visualizing future climate trends within the Mahanadi Basin. This study acknowledges limitations such as historical data uncertainties, socio-economic indicators, and unpredictable RCP trajectories, introducing a novel method to integrate machine learning with climate model data for assessing reliability. It also explores anticipated shifts in monthly precipitation and temperature patterns, providing insights into future climate variations.

Impact of Directly Assimilating Radar Reflectivity Using a Reflectivity Operator Based on a Double-Moment Microphysics Scheme on the Analysis and Forecast of Typhoon Lekima (1909)

In previous studies, radar reflectivity is often directly assimilated using reflectivity operators based on a single-moment (SM) microphysics scheme, though the forecast model uses a double-moment (DM) microphysics scheme. With the fixed number concentrations, only the mixing ratios of hydrometeors are directly updated during the assimilation, which leads to a mismatch between the analyzed microphysical state variables and the microphysics scheme of the forecast model. In this study, the radar reflectivity is directly assimilated through an observation operator consistent with the DM Thompson microphysics scheme used in numerical integrations, and the impact of reflectivity operators based on SM and DM schemes on the analysis performance of the ensemble Kalman filter for typhoon Lekima on 9 August 2019 is evaluated. Reflectivity observations from a single operational weather radar in Wenzhou City, Zhejiang Province of China are assimilated. In addition, the dual-polarization observations from the same radar are used to evaluate the quality of the analysis. The analyzed reflectivity and dual-polarization characteristics obtained by different reflectivity operators are examined in detail. Compared with the experiments applying the reflectivity operator based on the SM Lin scheme, the use of a reflectivity operator consistent with the DM Thompson scheme adopted in the forecast model results in analyzed reflectivity and polarization characteristics that are more consistent with the observed characteristics in terms of general structure, location, and intensity. Forecasted reflectivity, 3 h accumulated precipitation, and typhoon intensity and track are also evaluated. The application of the reflectivity operator based on the DM scheme makes better forecasts of typhoon intensity, precipitation, and reflectivity, which also improves the forecast skills on typhoon tracks to a certain extent.

A Method Based on Deep Learning for Severe Convective Weather Forecast: CNN-BiLSTM-AM (Version 1.0)

In this study, we propose a model called CNN-BiLSTM-AM that utilizes deep learning techniques to forecast severe convective weather based on ERA5 hourly data and observations. The model integrates a CNN with a Bidirectional Long Short-Term Memory (BiLSTM) system and an Attention Mechanism (AM). The CNN is tasked with extracting features from the input data, while the BiLSTM effectively captures temporal dependencies. The AM enhances the results by considering the impact of past feature states on severe weather phenomena. Additionally, we assess the performance of our model in comparison to traditional network architectures, including ConvLSTM, Predrnn++, CNN, FC-LSTM, and LSTM. Our results indicate that the CNN-BiLSTM-AM model exhibits superior accuracy in precipitation forecasting. Especially with the extension of the forecast time, the model performs well across multiple evaluation metrics. Furthermore, an interpretability analysis of the convective weather mechanisms utilizing machine learning highlights the critical role of total precipitable water (PWAT) in short-term heavy precipitation forecasts. It also emphasizes the significant impact of regional variables on convective weather patterns and the role of convective available potential energy (CAPE) in fostering conditions conducive to convection. These findings not only confirm the effectiveness of deep learning in the automatic identification of severe weather features but also validate the suitability of the sample dataset employed. Given its remarkable performance and robustness, we advocate for the adoption of this model to enhance the forecast of severe convective weather across various business applications.

Kilometer-Scale Precipitation Forecasting Utilizing Convolutional Neural Networks: A Case Study of Jiangsu’s Coastal Regions

High-resolution precipitation forecasts play a pivotal role in formulating comprehensive disaster prevention and mitigation plans. As spatial resolution enhances, striking a balance between computation, storage, and simulation accuracy becomes imperative to ensure optimal cost-effectiveness. Convolutional neural networks (CNNs), a cornerstone of deep learning, are examined in this study for their downscaling capabilities in precipitation simulation. During a precipitation event on 23 June 2022, in Jiangsu Province, China, distinct rain belts emerged in both southern and northern Jiangsu, precisely captured by a numerical model (the Weather Research and Forecasting, WRF) with a 3 km spatial resolution. Specifically, precipitation was prevalent in northern Jiangsu from 00:00 to 11:00 Beijing Time (BJT), transitioning to southern Jiangsu from 12:00 to 23:00 BJT on the same day. Upon dynamic downscaling, the model reproduced precipitation in these periods with an average error of 12.35 mm at 3 km and 12.48 mm at 1 km spatial resolutions. Employing CNN technology for statistical downscaling to a 1 km spatial resolution, samples from the initial period were utilized for training, while those from the subsequent period served for validation. Following dynamic downscaling, CNNs with four, five, six, and seven layers exhibited average errors of 8.86 mm, 8.93 mm, 9.71 mm, and 9.70 mm, respectively, accompanied by correlation coefficients of 0.550, 0.570, 0.574, and 0.578, respectively. This analysis indicates that for this precipitation event, a shallower CNN depth yields a lower average error and correlation coefficient, whereas a deeper architecture enhances the correlation coefficient. By employing deep network architectures, CNNs are capable of capturing nonlinear patterns and subtle local features from complex meteorological data, thereby providing more accurate predictions during the downscaling process. Leveraging faster computation and reduced storage requirements, machine learning has demonstrated immense potential in high-resolution forecasting research. There is significant scope for advancing technologies that integrate numerical models with machine learning to achieve higher-resolution numerical forecasts.