Search Results (291)

Accurate forecasting of tropical cyclone (TC) tracks is critical for disaster preparedness and risk mitigation. While traditional numerical weather prediction (NWP) systems have long served as the backbone of operational forecasting, they face limitations in computational cost and sensitivity to initial conditions. In recent years, deep learning (DL) has emerged as a promising alternative, offering data-driven modeling capabilities for capturing nonlinear spatiotemporal patterns. This paper presents a comprehensive review of DL-based approaches for TC track forecasting. We categorize all DL-based TC tracking models according to the architecture, including recurrent neural networks (RNNs), convolutional neural networks (CNNs), Transformers, graph neural networks (GNNs), generative models, and Fourier-based operators. To enable rigorous performance comparison, we introduce a Unified Geodesic Distance Error (UGDE) metric that standardizes evaluation across diverse studies and lead times. Based on this metric, we conduct a critical comparison of state-of-the-art models and identify key insights into their relative strengths, limitations, and suitable application scenarios. Building on this framework, we conduct a critical cross-model analysis that reveals key trends, performance disparities, and architectural tradeoffs. Our analysis also highlights several persistent challenges, such as long-term forecast degradation, limited physical integration, and generalization to extreme events, pointing toward future directions for developing more robust and operationally viable DL models for TC track forecasting. To support reproducibility and facilitate standardized evaluation, we release an open-source UGDE conversion tool on GitHub. Full article

Some of the most intense thunderstorms and extreme weather events on Earth occur in the Hindu Kush Himalaya (HKH) region of Southern Asia. The need to provide end users, stakeholders, and decision makers with accurate forecasts and alerts of extreme weather is critical. To that end, a cutting edge weather modeling framework coined the High Impact Weather Assessment Toolkit (HIWAT) was created through the National Aeronautics and Space Administration (NASA) SERVIR Applied Sciences Team (AST) effort, which consists of a suite of varied numerical weather prediction (NWP) model runs to provide probabilities of straight-line damaging winds, hail, frequent lightning, and intense rainfall as part of a daily 54 h forecast tool. The HIWAT system was first deployed in 2018, and the recently released model archive hosted by the Global Hydrometeorology Resource Center (GHRC) Distributed Active Archive Center (DAAC) provides daily model outputs for the years of 2018–2022. With a nested modeling domain covering Nepal, Bangladesh, Bhutan, and Northeast India, the HIWAT archive spans the critical pre-monsoon and monsoon months of March–October when severe weather and flooding are most frequent. As part of NASA’s Transformation To Open Science (TOPS), this data archive is freely available to practitioners and researchers. Full article

Aerosols considerably reduce the upwelling radiance in the thermal infrared (IR) window; thus, it is worthwhile to understand the effects and challenges of assimilating aerosol-affected (i.e., hazy-sky) IR observations for all-sky data assimilation (DA). This study introduces an aerosol-aware DA framework for the Infrared Atmospheric Sounder Interferometer (IASI) to exploit hazy-sky IR observations and investigate the impact of assimilating hazy-sky IR observations on analyses and subsequent forecasts. The DA framework consists of the detection of hazy-sky pixels and an observation error model as the function of the aerosol effect. Compared to the baseline experiment, the experiment utilized an aerosol-aware framework that reduces biases in the sea surface temperature in the tropical region, particularly over the areas affected by heavy dust plumes. There are no significant differences in the evaluation of the analyses and the 7-day forecasts between the experiments. To further improve the aerosol-aware framework, the enhancements in quality control (e.g., aerosol detection) and bias correction need to be addressed in the future. Full article

Accurate initial conditions are crucial for improving numerical weather prediction (NWP). Variational data assimilation relies on a static background error covariance matrix (B), yet its variance estimation is often inaccurate, affecting assimilation and forecast performance. This study introduces DRL-AST, a deep reinforcement learning-based adaptive variance rescaling strategy that dynamically adjusts the variances of B to optimize forecast skill through improved assimilation performance. By formulating variance rescaling as a Markov Decision Process and employing an actor–critic framework with Proximal Policy Optimization, DRL-AST autonomously selects spatio-temporal rescaling factors, enhancing assimilation and forecast accuracy without additional computational cost. As a new paradigm for adaptive variance tuning, DRL-AST demonstrates competitive improvements in forecast skill in experiments with the Lorenz-96 model by generating initial states that better conform to model dynamical consistency. Given its adaptability and efficiency, DRL-AST holds great potential for application in high-dimensional NWP models, where deep learning-based dimensionality reduction and reinforcement learning techniques could further enhance its feasibility and effectiveness in complex assimilation frameworks. Full article

Accurate day-ahead photovoltaics (PV) power forecasting results are significant for power grid operation. According to different weather modes, the existing research has established a classification forecast framework to improve the accuracy of day-ahead forecasts. However, the existing framework still has the following two problems: (1) weather mode prediction and power forecasting are highly dependent on the accuracy of numerical weather prediction (NWP), but the existing classification forecasting framework ignores the impact from NWP errors; (2) the validity of the classification forecasting framework comes from the accurate prediction of weather modes, but the existing framework lacks the analysis and decision-making mechanism of the reliability of weather mode prediction results, which will lead to a significant decline in the overall accuracy when weather modes are wrongly predicted. Therefore, this paper proposes a day-ahead PV power forecasting method based on irradiance correction and weather mode reliability decision. Firstly, based on the measured irradiance, K-means clustering method is used to obtain the daily actual weather mode labels; secondly, considering the coupling relationship of meteorological elements, the graph convolutional network (GCN) model is used to correct the predicted irradiance by using multiple meteorological elements of NWP data; thirdly, the weather mode label is converted into one-heat code, and a weather mode reliability prediction model based on a convolutional neural network (CNN) is constructed, and then the prediction strategy of the day to be forecasted is decided; finally, based on the weather mode reliability prediction results, transformer model are established for unreliable weather and credible weather respectively. The simulation results of the ablation experiments show that classification prediction is an effective strategy to improve the forecasting accuracy of day-ahead PV output, which can be further improved by adding irradiance correction and weather mode reliability prediction modules. Full article

Accurate interval forecasting of wind power is crucial for ensuring the safe, stable, and cost-effective operation of power grids. In this paper, we propose a hybrid deep learning model for day-ahead wind power interval forecasting. The model begins by utilizing a Gaussian mixture model (GMM) to cluster daily data with similar distribution patterns. To optimize input features, a feature selection (FS) method is applied to remove irrelevant data. The empirical wavelet transform (EWT) is then employed to decompose both numerical weather prediction (NWP) and wind power data into frequency components, effectively isolating the high-frequency components that capture the inherent randomness and volatility of the data. A convolutional neural network (CNN) is used to extract spatial correlations and meteorological features, while the bidirectional gated recurrent unit (BiGRU) model captures temporal dependencies within the data sequence. To further enhance forecasting accuracy, a multi-head self-attention mechanism (MHSAM) is incorporated to assign greater weight to the most influential elements. This leads to the development of a day-ahead wind power interval forecasting model based on GMM-FS-EWT-CNN-BiGRU-MHSAM. The proposed model is validated through comparison with a benchmark forecasting model and demonstrates superior performance. Furthermore, a comparison with the interval forecasts generated using the NPKDE method shows that the new model achieves higher accuracy. Full article

Weather forecasting is essential for various applications such as agriculture and transportation, and relies heavily on meteorological sequential data such as multivariate time series collected from weather stations. Traditional numerical weather prediction (NWP) methods applied to multivariate time series forecasting are grounded in statistical principles such as Autoregressive Integrated Moving Average (ARIMA); however, they often struggle with capturing complex nonlinear patterns among meteorological variables and temporal variances. Currently, existing deep learning approaches such as Recurrent Neural Networks (RNNs) and transformers offer remarkable performance in handling complex patterns among meteorological multivariate time series, yet frequently fail to maintain weather-specific physical properties such as strict values constraints, while also incurring the significant computational costs of large parameter scales. In this paper, we present a novel deep learning plug-and-play framework named Post Constraint and Correction (PCC) to address these challenges by incorporating additional constraints and corrections based on weather-specific properties such as multivariant correlations and physical-based strict value constraints into the prediction process. Our method demonstrates notable computational efficiency, delivering significant improvements over existing deep learning time series models and helping to achieve better performance with far fewer parameters. Extensive experiments demonstrate the effectiveness, efficiency, and robustness of our method, highlighting its potential for real-world applications. Full article

Heavy rainfall events are the main cause of flooding, especially in regions like Kerala, India. Kerala is vulnerable to extreme weather due to its geographical location in the Western Ghats. Accurate forecasting of rainfall events is essential for minimizing the impact of floods on life, infrastructure, and agriculture. For accurate forecasting of heavy rainfall events in this region, region-specific evaluations of NWP model performance are very important. This study evaluated the performance of six Numerical Weather Prediction (NWP) models—NCEP, NCMRWF, ECMWF, CMA, UKMO, and JMA—in forecasting heavy rainfall events in Kerala. A comprehensive assessment of these models was performed using traditional performance metrics, categorical precipitation metrics, and Fractional Skill Scores (FSSs) across different forecast lead times. FSSs were calculated for different rainfall thresholds (100 mm, 50 mm, 5 mm). The results reveal that all models captured rainfall patterns well for the lower threshold of 5 mm, but most of the models struggled to accurately forecast heavy rainfall, especially for longer lead times. JMA performed well overall in most of the metrics except False Alarm Ratio (FAR). It showed high FAR, which revealed that it may predict false rainfall events. ECMWF demonstrated consistent performance. NCEP and UKMO performed moderately well. CMA, and NCMRWF had the lowest accuracy either due to more errors or biases. The findings underscore the trade-offs in model performance, suggesting that model selection should depend on the accuracy required or rainfall event prediction capability. This study recommends the use of Multi-Model Ensembles (MME) to improve forecasting accuracy, integrate the strengths of the best-performing models, and reduce biases. Future research can also focus on expanding observational networks and employing advanced data assimilation techniques for more reliable predictions, particularly in regions with complex terrain such as Kerala. Full article

Winter mixed-phase precipitation (P) impacts transportation, electric power grids, and homes. Forecasting winter precipitation such as freezing precipitation (ZP), freezing rain (ZR), freezing drizzle (ZL), ice pellets (IPs), and the snow (S) and rain (R) boundary remains challenging due to the complex cloud microphysical and dynamical processes involved, which are difficult to predict with the current numerical weather prediction (NWP) models. Understanding these processes based on observations is crucial for improving NWP models. To aid this effort, Environment and Climate Change Canada deployed specialized instruments such as the Vaisala FD71P and OTT PARSIVEL disdrometers, which measure P type (PT), particle size distributions, and fall velocity (V). The liquid water content (LWC) and mean mass-weighted diameter (${\mathrm{D}}_{\mathrm{m}}$) were derived based on the PARSIVEL data during ZP events. Additionally, a Micro Rain Radar (MRR) and an OTT Pluvio2 P gauge were used as part of the Winter Precipitation Type Research Multi-Scale Experiment (WINTRE-MIX) field campaign at Sorel, Quebec. The dataset included manual measurements of the snow water equivalent (SWE), PT, and radiosonde profiles. The analysis revealed that the FD71P and PARSIVEL instruments generally agreed in detecting P and snow events. However, FD71P tended to overestimate ZR and underestimate IPs, while PARSIVEL showed superior detection of R, ZR, and S. Conversely, the FD71P performed better in identifying ZL. These discrepancies may stem from uncertainties in the velocity–diameter (V-D) relationship used to diagnose ZR and IPs. Observations from the MRR, radiosondes, and surface data linked ZR and IP events to melting layers (MLs). IP events were associated with colder surface temperatures (Ts) compared to ZP events. Most ZR and ZL occurrences were characterized by light P with low LWC and specific intensity and ${\mathrm{D}}_{\mathrm{m}}$ thresholds. Additionally, snow events were more common at warmer T compared to liquid P under low surface relative humidity conditions. The Pluvio2 gauge significantly underestimated snowfall compared to the optical probes and manual measurements. However, snowfall estimates derived from PARSIVEL data, adjusted for snow density to account for riming effects, closely matched measurements from the FD71P and manual observations. Full article

Many Caribbean low-latitude small island states lack wind maps tailored to capture their wind features at high resolutions. However, high-resolution mesoscale modeling is computationally expensive. This study proposes a statistical–dynamical downscaling (SDD) method that integrates an atmospheric-circulation-type (CT) approach with a high-resolution numerical weather prediction (NWP) model to map the wind resources of a case study, Trinidad and Tobago. The SDD method uses a novel wind class generation technique derived directly from reanalysis wind field patterns. For the Caribbean, 82 wind classes were defined from an atmospheric circulation catalog of seven types derived from 850 hPa daily wind fields from the NCEP-DOE reanalysis over 32 years. Each wind class was downscaled using the Weather Research and Forecasting (WRF) model and weighted by frequency to produce 1 km × 1 km climatological wind maps. The 10 m wind maps, validated using measured wind data at Piarco and Crown Point, exhibit a small positive average bias (+0.5 m/s in wind speed and +11 W m⁻² in wind power density (WPD)) and capture the shape of the wind speed distributions and a significant proportion of the interannual variability. The 80 m wind map indicates from good to moderate wind resources, suitable for determining priority areas for a detailed wind measurement program in Trinidad and Tobago. The proposed SDD methodology is applicable to other regions worldwide beyond low-latitude tropical islands. Full article

The Australian Bureau of Meteorology provides flood forecasting and warning services across Australia for most major rivers in Australia, in cooperation with other government, local agencies and emergency services. As part of this service, the Bureau issues a flood watch product to provide early advice on a developing situation that may lead to flooding up to 4 days prior to an event. This service is based on (a) an ensemble of available Numerical Weather Prediction (NWP) rainfall forecasts, (b) antecedent soil moisture, stream and dam conditions, (c) hydrological forecasts using event-based models and (d) expert meteorological and hydrological input by Bureau of Meteorology staff, to estimate the risk of reaching pre-specified river height thresholds at locations across the continent. A flood watch provides information about a developing weather situation including forecasting rainfall totals, catchments at risk of flooding, and indicative severity where required. Although there is uncertainty attached to a flood watch, its early dissemination can help individuals and communities to be better prepared should flooding eventuate. This paper investigates the utility of forecasts of daily gridded national runoff to inform the risk of riverine flooding up to 7 days in advance. The gridded national water balance model (AWRA-L) runoff outputs generated using post-processed 9-day Numerical Weather Prediction hindcasts were evaluated as to whether they could accurately predict exceedance probabilities of runoff at gauged locations. The approach was trialed over 75 forecast locations across North East Australia (Queensland). Forecast 3-, 5- and 7-day accumulations of runoff over the catchment corresponding to each location were produced, identifying whether accumulated runoff reached either 95% or 99% historical levels (analogous to minor, moderate and major threshold levels). The performance of AWRA-L runoff-based flood likelihood was benchmarked against that based on precipitation only (i.e., not rainfall–runoff transformation). Both products were evaluated against the observed runoff data measured at the site. Our analysis confirmed that this runoff-based flood likelihood guidance could be used to support the generation of flood watch products. Full article

The accurate forecasting of time, intensity, and spatial distribution is fundamental to weather prediction. However, the limitations of numerical weather prediction (NWP) models, as well as uncertainties in inital conditions, often lead to temporal biases in forecasts. This study addresses these biases by employing visible reflectance data from the Himawari-9/AHI satellite and RTTOV (TOVS radiation transfer) simulations derived from CMA-MESO model outputs. The time-shift method was applied to analyze two precipitation events—20 October 2023 and 30 April 2024—in order to assess its impact on precipitation forecasts. The results indicate the following: (1) the time-shift method improved cloud simulations, necessitating a 30 min advance for Case 1 and a 3.5 h delay for Case 2; (2) time-shifting reduced the standard deviation of observation-minus-background (OMB) bias in certain regions and enhanced spatial uniformity; (3) the threat score (TS) demonstrated an improvement in forecast accuracy, particularly in cases exhibiting significant movement patterns. The comparative analysis demonstrates that the time-shift method effectively corrects temporal biases in NWP models, providing forecasters with a valuable tool to optimize predictions through the integration of high-temporal- and spatial-resolution visible light data, thereby leading to more accurate and reliable weather forecasts. Full article

Water vapor is an active trace component in the troposphere and has a significant impact on meteorology and the atmospheric environment. In order to meet demands for high-precision water vapor and aerosol observations for numerical weather prediction (NWP), the China Meteorological Administration (CMA) deployed 49 Raman aerosol lidar systems and established the first Raman–Mie scattering lidar network in China (CARLNET) for routine measurements. In this paper, we focus on the water vapor measurement capabilities of the CARLNET. The uncertainty of the water vapor Raman channel calibration coefficient (Cw) is determined using an error propagation formula. The theoretical relationship between the uncertainty of the calibration coefficient and the water vapor mixing ratio (WVMR) is constructed based on least squares fitting. Based on the distribution of lidar in regions with different humidity conditions, the method of real-time calibration and quality control based on radiosonde data is established for the first time. Based on the uncertainty requirements of the World Meteorological Organization for water vapor in data assimilation, the calibration and quality control thresholds of the WVMR in regions with different humidity conditions are determined by fitting real-time lidar and radiosonde data. Lastly, based on the radiosonde results, the calibration algorithm established in this study is used to calibrate CARLNET data from October to December 2023. Compared with traditional calibration results, the results show that the stability and detection accuracy of the CARLNET significantly improved after calibration in regions with different humidity conditions. The deviation of the Cw decreased from 12.84~18.47% to 5.41~11.54%. The inversion error of the WVMR compared to radiosonde decreased from 1.05~0.46 g/kg to 0.82~0.34 g/kg. The reliability of the improved calibration algorithm and the CARLNET’s performance have been verified, enabling them to provide high-precision water vapor products for NWP. Full article

Conventional weather forecasting relies on numerical weather prediction (NWP), which solves atmospheric equations using numerical methods. The Korea Meteorological Administration (KMA) adopted the Met Office Global Seasonal Forecasting System version 6 (GloSea6) NWP model from the UK and runs it on a supercomputer. However, due to high task demands, the limited resources of the supercomputer have caused job queue delays. To address this, the KMA developed a low-resolution version, Low GloSea6, for smaller-scale servers at universities and research institutions. Despite its ability to run on less powerful servers, Low GloSea6 still requires significant computational resources like those of high-performance computing (HPC) clusters. We integrated deep learning with Low GloSea6 to reduce execution time and improve meteorological research efficiency. Through profiling, we confirmed that deep learning models can be integrated without altering the original configuration of Low GloSea6 or complicating physical interpretation. The profiling identified “tri_sor.F90” as the main CPU time hotspot. By combining the biconjugate gradient stabilized (BiCGStab) method, used for solving the Helmholtz problem, with a deep learning model, we reduced unnecessary hotspot calls, shortening execution time. We also propose a convolutional block attention module-based Half-UNet (CH-UNet), a lightweight 3D-based U-Net architecture, for faster deep-learning computations. In experiments, CH-UNet showed 10.24% lower RMSE than Half-UNet, which has fewer FLOPs. Integrating CH-UNet into Low GloSea6 reduced execution time by up to 71 s per timestep, averaging a 2.6% reduction compared to the original Low GloSea6, and 6.8% compared to using Half-UNet. This demonstrates that CH-UNet, with balanced FLOPs and high predictive accuracy, offers more significant execution time reductions than models with fewer FLOPs. Full article

Pangu is an AI-based model designed for rapid and accurate numerical weather forecasting. To evaluate Pangu’s short- to medium-term weather forecasting skill over various meteorological parameters, this paper validated its performance in predicting temperature, wind speed, wind direction, and barometric pressure using data from over 2000 weather stations in China. Pangu’s performance was compared with ECMWF-HRES and GFS to assess its effectiveness relative to traditional high-precision NWP models under real meteorological conditions. Furthermore, the more recent FuXi and FengWu models were included in the analysis to further validate Pangu’s forecasting skill. The study examined Pangu’s forecast performance from spatial perspectives, evaluated the dispersion of forecast deviations, and analyzed its performance at different lead times and with various initial fields. The iteration precision of Pangu’s four forecast models with lead times of 1 h, 3 h, 6 h, and 24 h was also assessed. Finally, a case study on typhoon track forecasting was conducted to evaluate Pangu’s performance in predicting typhoon paths. The results indicate that Pangu surpasses traditional NWP systems in temperature forecasting, while its performance in predicting wind direction, wind speed and pressure is comparable to them. Additionally, the forecast skill of Pangu diminishes as the lead time extends, but it tends to surpass traditional NWP systems with longer lead times. Moreover, FuXi and FengWu demonstrate even higher accuracy compared to Pangu. Pangu’s performance is also dependent on initial fields, and the temperature forecasting of Pangu is more sensitive to the initial field compared with other meteorological parameters. Furthermore, the iteration precision of Pangu’s 1 h forecast model is significantly lower than that of the other models, but this discrepancy in precision may not be prominently reflected in Pangu’s actual forecasting process due to the greedy algorithm employed. In the case study on typhoon forecasting, Pangu, along with FuXi and FengWu, demonstrates comparable performance in predicting Bebinca’s track compared to ECMWF and outperforms GFS in its track predictions. This study demonstrated Pangu’s applicability in short- to medium-term forecasting of meteorological parameters, showcasing the significant potential of AI-based numerical weather models in enhancing forecast performance. Full article