Meteorological Applications最新文献

Data-driven weather models have shown the potential to match the accuracy of state-of-the-art numerical weather predictions (NWPs). However, existing data-driven forecasting models still have limitations in operational applications. For example, most of them are predominantly trained via fifth-generation climate reanalysis data (ERA5). However, in actual forecasting operations, the models are usually initiated by analysis fields instead of reanalysis data; this leads to a mismatch between the training data used by machine learning (ML) forecasting models and the actual operational data. To address this issue, we attempt to fine-tune the data-driven model with the initiation fields in operation. This study first develops a fine-tuned Pangu Weather Model (PGW) by integrating forecasting system (IFS) analysis data from 2021 to 2022 and conducts a comprehensive evaluation of its performance. By comparing the fine-tuned version (PGW_O) with the public version (PGW_P) against IFS models with different resolutions (IFS_L at 0.25° and IFS_H at 0.1°), this research highlights advancements in data-driven forecasting methodologies. The models are tested on data from South China, a region with dense meteorological observation networks, over a three-month period, encompassing a detailed case study of Tropical Cyclone Haikui (2023). The findings show that with the forecast activity (FA) level comparable to PGW_P, PGW_O significantly reduces the root mean square error (RMSE) and mean error (ME) across upper atmospheric variables and demonstrates superior accuracy in predicting surface elements. The operational relevance of these models is evaluated through both ERA5 reanalysis and surface observations, revealing that fine-tuning with IFS data enhances PGW compatibility and forecasting precision, particularly for severe weather events.

Rain-fed crop yields are heavily influenced by seasonal rainfall patterns and temperature, particularly during vegetative and reproductive growth stages. This study was conducted to investigate the effects of rainfall distribution indices (monthly, seasonal, and annual) on rain-fed wheat and barley yields using polynomial regression analysis across six different locations with varying elevations in Chaharmahal and Bakhtiari province, Iran. Additionally, the economic feasibility of rain-fed wheat and barley in all locations was evaluated. Results showed that the monthly rainfall distribution index could not accurately predict wheat/barley yield, where elevation exceeds 2000 m and the average annual minimum temperature is below 4°C (such as in Koohrang, Borujen, Shahrekord, and Farsan). Conversely, the monthly rainfall distribution index was able to predict the wheat/barley yield with high accuracy (R² > 0.75) in locations with lower elevation and higher average annual minimum temperature (such as Lordegan and Ardal). Compared to seasonal rainfall indices, annual rainfall indices showed weaker predictive accuracy in all locations. Furthermore, a significant relationship (p-value < 0.0001) with a high coefficient of determination (R² > 0.80) was found between spring rainfall index, spring minimum temperature, and wheat/barley yield in all locations. Therefore, incorporating minimum mean air temperature with the spring rainfall index is recommended for yield prediction for all locations. Economic analysis revealed that the internal return rates in Borujen, Farsan, Lordegan and Ardal exceeded the bank interest rate (14%), indicating that cultivating wheat and barley in these four locations was profitable and economic. Moreover, an exponential relationship between the average annual temperature and internal return rate was also established, offering a useful tool for farmers and planners to estimate the internal return rate based on only the average annual temperature.

Climate change poses significant challenges to agricultural production, particularly in Central Africa, where the livelihoods of millions depend on key crops such as maize, groundnut, soybean, and rice. The potential effects of climate projections on agricultural yields are significant, as variations in temperature, rainfall, humidity, and soil moisture can lead to substantial changes in crop performance. The research aims to model and predict crop yields based on these meteorological variables by utilizing machine learning models, including Gaussian process and Random forest. The findings demonstrate that regional agricultural production differences may arise from future climatic conditions. The random forest model aligned more closely with observed values, achieving better average accuracies depending on the season. The performance of the machine learning models is closely tied to the specific crops and countries within the study region. Furthermore, the insights gained can greatly benefit political decision-makers and stakeholders in developing targeted adaptation plans and policies.

Vertical mixing in the planetary boundary layer greatly influences thunderstorm activity. The sensitivity of two local (MYJ and MYNN) and one non-local (YSU) PBL schemes with a combination of Single Layer Urban Canopy Model (SLUCM) of the Weather Research and Forecasting (WRF) model is studied at 2 km horizontal resolution for the evolution of thunderstorms. Twelve thunderstorms over four cities in the eastern Indian region are identified during 2016–2021. Results highlighted that the YSU scheme performs better with a rainfall absolute percentage of error of 27%, while the MYJ and MYNN exhibited comparatively higher errors of 31% and 38%, respectively, within a 50 km area from the city center. The mean timing error of initiation and mature stage against GPM rainfall is 0–1 h in the YSU scheme and 0.5–2 h for both MYJ and MYNN. The lead–lag correlation (0.6 at 00 h) and quantitative rain rate verification also confirm the better performance of YSU. Surface (2 m) and atmospheric dynamical and thermodynamic profiles are replicated well with lower errors in YSU, except for 10 m wind speed. Diagnostic analysis indicates that higher frictional velocities and turbulent kinetic energy in YSU resemble the higher vertical mixing, leading to an unstable atmosphere with stronger updrafts. These PBL characteristics are relatively weaker in MYJ and MYNN as well as the stability indices. Overall, the better performance of the YSU scheme can be attributed to the better transport of surface characteristics, including turbulent fluxes and moisture, to the upper levels in an unstable atmosphere with strong vertical velocities. Further, results highlight that the simulation of urban thunderstorms improved with urban physics when compared with no-urban simulations. Thus, this study emphasizes the role of PBL along with urban physics in steering the dynamics of urban thunderstorms.

The Advanced Baseline Imager (ABI) onboard the geostationary satellite GOES-18, launched on March 14, 2022, and re-designated as GOES-West in January 2023, has been providing data to the National Centre for Medium Range Weather Forecasting (NCMRWF) since its operational commencement. This study endeavors to develop and evaluate GOES-18 satellite radiance data assimilation within the Global Data Assimilation System (GDAS) at NCMRWF, specifically on simulating Pacific hurricanes impacting the Western United States. The study includes two main components: (1) developing and assessing the reliability of the GOES-18 radiance observation assimilation capability in the NCMRWF Global Forecasting System (NGFS), and (2) simulating and analyzing the catastrophic Category-4 hurricane Hilary, which caused severe damage and heavy rain in the Western United States and Mexico. A month-long analysis of data reveals that GOES-18 provides a substantially larger number of observations compared to GOES-16, with a more significant proportion of observations being accepted during the assimilation cycle. Error metrics (e.g., spread, standard deviation, RMSE) were estimated for background fields without bias correction, with BC, and analysis with BC compared to observations. The results indicate a significant reduction in RMSE (~50%) in the analysis, thereby establishing a positive signature for the assimilation of GOES-18 observations. This study further investigates the efficacy of assimilating GOES-18 data in simulating hurricane Hilary using the NGFS with a focus on evaluating potential improvements in track, intensity, and inner core structure of the system.

Under the background of climate change, extreme weather events are apparently becoming more frequent. In Hong Kong, a record-breaking ‘Black’ Rainstorm on 7–8 September 2023 brought widespread flooding and caused landslides, paralysing the entire community. To enhance the community's response and resilience in coping with extreme weather, the Hong Kong Observatory has developed the Flood Risk Assessment System (FRAS), which embraces an impact-based forecasting method and a risk-based warning strategy. The main inputs are real-time rainfall data from rain gauges and the in-house developed probabilistic rainfall nowcast. The process from rain falling through the air to flooding observed on the ground is complicated, involving many non-meteorological and random factors. As a result, the corresponding impact assessment is highly non-trivial. The key technique adopted by FRAS is the use of a district-scale ‘rainfall-flooding impact’ statistical model, developed through in-depth study of historical flood reports and rainfall records. The risk-based warning strategy is designed largely based on the risk matrix recommended by the World Meteorological Organization. The performance of FRAS has been optimised in accordance with users' operational needs under the premises of a high safety margin and early alert. FRAS was launched in May 2024 for trial by government departments/bureaux, operating continuously in real-time and offering automatic flood risk assessment for all districts every minute during rainy seasons. This paper briefly presents the design, key techniques, and warning products of FRAS. Its performance as an early warning service is also examined through objective verification results and user feedback.

Typhoon-induced rainfall can trigger floods and landslides that pose significant hazards in Zhejiang. However, accurately forecasting its magnitude remains a significant challenge for global Numerical Weather Prediction (NWP) models, particularly in mountainous regions where complex orographic effects play a critical role. Using forecast data from three global NWP models (ECMWF, NCEP, and CMA-GFS) and observed station data from 2021 to 2024, this study revealed that these models consistently underestimated heavy rainfall in high-altitude areas, associated with their limited resolution in representing terrain-induced amplification. To address this, the terrain correction method proposed by Xu et al. (2019) was applied to the models to improve the estimation of typhoon-induced orographic rainfall in Zhejiang. Significant improvements in forecast skill were evidenced by increased Threat Scores (TS) and Probabilities of Detection (POD). In the ECMWF model, TS for rainstorms (≥ 50 mm·day⁻¹) increased from 0.33 to 0.35, while POD rose from 0.54 to 0.68. Larger gains were observed for heavy downpours (≥ 250 mm·day⁻¹), with TS rising from 0.02 to 0.08 and POD from near zero to 0.34. Similar improvements were found in the NCEP and CMA-GFS models. This study also discussed the limitations of terrain correction and identified two scenarios in which it underperformed. One occurred when the original forecast overestimated rainfall due to excessive moisture flux convergence, sometimes further amplifying errors (e.g., the forecast initialized at 00:00 UTC 24 July 2021). The other involved spatial displacement of the predicted rainfall field due to typhoon track errors, resulting in poor alignment with observations (e.g., 00:00 UTC 12 September 2021). Despite these limitations, the terrain correction notably improved forecasting skills, as indicated by TS and POD metrics, thereby enhancing local preparedness against typhoon-induced heavy rainfall and helping mitigate the risks of flooding and other related hazards in Zhejiang.

In this study, the indicative effect of the steering flow and the ventilation flow on the motion of tropical cyclones (TCs) is studied when the TCs are nearshore or landfalling. Results show that the steering flow derived with a radius of 3°–5° has smaller directional deviation and better directional indication effect for nearshore or landfalling TCs compared with traditional calculations with a radius of 5°–7° over the ocean, with an average deviation of about 25°–26° for a radius of 3°–5° and about 37°–40° for a radius of 5°–7°. Moreover, ventilation flow with a radius of 3° has a further better directional indication effect and smaller directional deviation, with the average deviation of about 18°–23° for nearshore or landfalling TCs, which is further improved compared with that of the steering flow. Although the directional deviation of steering flow and ventilation flow both increase as TCs approach the mainland, the closer to the land, the weaker the TC or the slower the TC motion, the more the guiding effect of ventilation flow improves compared to steering flow. Therefore, the ventilation flow is more suitable as an indicator for the TC path than steering flow when TCs approach the mainland or make landfall. Statistical analysis shows that the translation speed and the intensity of TCs contribute most to the asymmetry of TCs, which has a high relationship with the ventilation flow. In addition, the high asymmetry of TCs usually occurs when TCs move into regions with large vorticity at low level, low temperature at high- or low-level, large zonal wind at high level, and large deep environmental vertical wind shear.

Lightning is one of the most hazardous natural phenomena, causing significant damage to life and property. In India, lightning activity peaks during pre-monsoon and monsoon seasons. In 2022, 36% of the deaths from natural disasters were attributed to lightning. Accurate forecasting is critical for preparedness and mitigation, but complex convection processes often lead to spatial mismatches in forecasts. Spatial verification methods offer valuable insights into the accuracy of modeling systems. This study evaluates the performance of a high-resolution (4 km) regional model, operational at the National Centre for Medium Range Weather Forecasting (NCMRWF), that is, NCUMR (NCMRWF Regional Model), in predicting lightning strikes over India during pre-monsoon and monsoon seasons from 2021 to 2024. The Method for Object-Based Diagnostic Evaluation (MODE) was applied to assess the model's ability to predict lightning-prone regions. The primary objectives of this study are (i) to analyze the performance of the NCUMR model in predicting regions affected by lightning and (ii) to determine whether MODE can be used as an effective tool for forecasting lightning-prone areas. Results demonstrate that the NCUMR model is capable of forecasting the spatial structure and distribution of lightning events with reasonable accuracy up to 3 days in advance. On Day 1, more than 88% of lightning objects for thresholds above 5 strikes/day show boundary overlap with observations, with centroid distances for 50% of matched objects remaining below 55 km. For Day 2 lead time, 83%–85% of objects show boundary overlap. On Day 3, although displacement errors increase slightly, over 85% of objects still exhibit zero boundary distance at lower thresholds, and centroid distances remain within 1°–1.5°. For all lead times, 75% of the forecasted objects have area ratios exceeding 0.7, and complexity ratios consistently above 0.7, indicating good structural agreement. While intensity is generally under-forecasted, 90^th percentile intensity ratios exceed 0.5 in most cases. The model performs better for lower thresholds and shows improved object correspondence during the monsoon season compared to pre-monsoon. These results confirm the utility of object-based verification using MODE in capturing spatial aspects of lightning forecasts and highlight its potential application for real-time impact-based forecasting and early warning systems.

The article presents results from a computationally low-cost regional numerical weather prediction chain based on the Weather Research and Forecasting (WRF) model and its data assimilation (DA) suite WRFDA. Experiments with 24-h forecasts were performed twice daily (at 00 and 12 UTC) over a domain encompassing the European Alps and their surroundings with a 3.5 km grid spacing. The assimilation of surface observations with the 3D-Var algorithm improves near-surface temperature and humidity forecasts compared to control runs without assimilation. The forecast skill for near-surface variables is evaluated using independent surface observations. In the first six forecast hours, it is generally better in the assimilation experiments than in the control ones, with a mean error reduction of 0.26 K for temperature and 0.13 g kg⁻¹ for specific humidity in the 00 UTC runs, and of 0.12 K for temperature and 0.18 g kg⁻¹ for specific humidity in the 12 UTC runs. The assimilation reduces the standard deviation of the errors by a factor between 7% and 10% both for temperature and specific humidity. Verification with radiosonde measurements shows that assimilating surface observations increases the mean error in temperature and humidity forecasts within the planetary boundary layer (PBL), relative to the control. We show that the vertical structure of the adjustments to the model state resulting from DA (the analysis increments) is such that model biases are reduced near the surface but amplified higher up in the PBL. Finally, the assimilation of surface observations has a different impact on surface temperature forecasts in mountainous regions compared to adjacent plains. The error reduction is substantially higher in the plains than in the mountains, which likely depends on the inappropriate spreading of information along terrain-following model levels by the static covariances in 3D-Var. The relative accuracy of surface temperature forecasts in these two regions has a diurnal variability, with larger mean errors in the mountains during the day and in the plains at night.