Non-crossing Quantile Regression Neural Network as a Calibration Tool for Ensemble Weather Forecasts

Despite the maturity of ensemble numerical weather prediction(NWP),the resulting forecasts are still,more often than not,under-dispersed.As such,forecast calibration tools have become popular.Among those tools,quantile regression(QR)is highly competitive in terms of both flexibility and predictive performance.Nevertheless,a long-standing problem of QR is quantile crossing,which greatly limits the interpretability of QR-calibrated forecasts.On this point,this study proposes a non-crossing quantile regression neural network(NCQRNN),for calibrating ensemble NWP forecasts into a set of reliable quantile forecasts without crossing.The overarching design principle of NCQRNN is to add on top of the conventional QRNN structure another hidden layer,which imposes a non-decreasing mapping between the combined output from nodes of the last hidden layer to the nodes of the output layer,through a triangular weight matrix with positive entries.The empirical part of the work considers a solar irradiance case study,in which four years of ensemble irradiance forecasts at seven locations,issued by the European Centre for Medium-Range Weather Forecasts,are calibrated via NCQRNN,as well as via an eclectic mix of benchmarking models,ranging from the naïve climatology to the state-of-the-art deep-learning and other non-crossing models.Formal and stringent forecast verification suggests that the forecasts post-processed via NCQRNN attain the maximum sharpness subject to calibration,amongst all competitors.Furthermore,the proposed conception to resolve quantile crossing is remarkably simple yet general,and thus has broad applicability as it can be integrated with many shallow-and deep-learning-based neural networks.

Near-surface wind field characteristics of the desert-oasis transition zone in Dunhuang,China

The desert-oasis transition zone(DOTZ)serves as a buffer area between the desert and oasis.Understanding its wind field characteristics is of great significance for the prevention and control of aeolian disasters in the oasis.In this study,we used meteorological data during 2013–2019 from the portable meteorological stations at five sites(site A on the edge of the oasis,sites B,C,and D in the DOTZ,and site O in the desert)in Dunhuang,China to analyze the near-surface wind field characteristics and their causes,as well as to reveal the key role of the DOTZ in oasis protection.The results showed that the mean wind speed,frequency of sand-driving wind,and directional variability of wind decreased from west to east within the DOTZ,and wind speed was significantly affected by air temperature.The terrain influenced the prevailing winds in the region,mainly from northeast and southwest.Only some areas adjacent to the oasis were controlled by southeasterly wind.This indicated that the near-surface wind field characteristics of the DOTZ were caused by the combined effects of local terrain and surface hydrothermal difference.At site D,the annual drift potential(DP)was 24.95 vector units(VU),indicating a low wind energy environment,and the resultant drift direction(RDD)showed obvious seasonal differences.Additionally,the DOTZ played an important buffering role between the desert and oasis.Compared with the desert,the mean wind speed in the oasis decreased by 64.98%,and the prevailing wind direction was more concentrated.The results of this study will be useful in interpreting the aeolian activity of the DOTZ in Dunhuang.

3D near-surface P-wave velocity structure imaging with Distributed Acoustic Sensing and electric hammer source

Distributed Acoustic Sensing(DAS) is an emerging technique for ultra-dense seismic observation, which provides a new method for high-resolution sub-surface seismic imaging. Recently a large number of linear DAS arrays have been used for two-dimensional S-wave near-surface imaging in urban areas. In order to explore the feasibility of three-dimensional(3D) structure imaging using a DAS array, we carried out an active source experiment at the Beijing National Earth Observatory. We deployed a 1 km optical cable in a rectangular shape, and the optical cable was recast into 250 sensors with a channel spacing of 4 m. The DAS array clearly recorded the P, S and surface waves generated by a hammer source. The first-arrival P wave travel times were first picked with a ShortTerm Average/Long-Term Average(STA/LTA) method and further manually checked. The P-wave signals recorded by the DAS are consistent with those recorded by the horizontal components of short-period seismometers. At shorter source-receiver distances, the picked P-wave arrivals from the DAS recording are consistent with vertical component recordings of seismometers, but they clearly lag behind the latter at greater distances.This is likely due to a combination of the signal-to-noise ratio and the polarization of the incoming wave. Then,we used the Tomo DD software to invert the 3D P-wave velocity structure for the uppermost 50 m with a resolution of 10 m. The inverted P-wave velocity structures agree well with the S-wave velocity structure previously obtained through ambient noise tomography. Our study indicates the feasibility of 3D near-surface imaging with the active source and DAS array. However, the inverted absolute velocity values at large depths may be biased due to potential time shifts between the DAS recording and seismometer at large source-receiver distances.

Evaluating the Robustness of MDSS Maintenance Forecasts Using Connected Vehicle Data

The Indiana Department of Transportation (INDOT) adopted the Maintenance Decision Support System (MDSS) for user-defined plowing segments in the winter of 2008-2009. Since then, many new data sources, including connected vehicle data, enhanced weather data, and fleet telematics, have been integrated into INDOT winter operations activities. The objective of this study was to use these new data sources to conduct a systematic evaluation of the robustness of the MDSS forecasts. During the 2023-2024 winter season, 26 unique MDSS forecast data attributes were collected at 0, 1, 3, 6, 12 and 23-hour intervals from the observed storm time for 6 roadway segments during 13 individual storms. In total, over 888,000 MDSS data points were archived for this evaluation. This study developed novel visualizations to compare MDSS forecasts to multiple other independent data sources, including connected vehicle data, National Oceanic and Atmospheric Administration (NOAA) weather data, road friction data and snowplow telematics. Three Indiana storms, with varying characteristics and severity, were analyzed in detailed case studies. Those storms occurred on January 6th, 2024, January 13th, 2024 and February 16th, 2024. Incorporating these visualizations into winter weather after-action reports increases the robustness of post-storm performance analysis and allows road weather stakeholders to better understand the capabilities of MDSS. The results of this analysis will provide a framework for future MDSS evaluations and implementations as well as training tools for winter operation stakeholders in Indiana and beyond.

Decadal Forecasts of Large Earthquakes along the Northern San Andreas Fault System, California: Increased Activity on Regional Creeping Faults Prior to Major and Great Events

The three largest earthquakes in northern California since 1849 were preceded by increased decadal activity for moderate-size shocks along surrounding nearby faults. Increased seismicity, double-difference precise locations of earthquakes since 1968, geodetic data and fault offsets for the 1906 great shock are used to re-examine the timing and locations of possible future large earthquakes. The physical mechanisms of regional faults like the Calaveras, Hayward and Sargent, which exhibit creep, differ from those of the northern San Andreas, which is currently locked and is not creeping. Much decadal forerunning activity occurred on creeping faults. Moderate-size earthquakes along those faults became more frequent as stresses in the region increased in the latter part of the cycle of stress restoration for major and great earthquakes along the San Andreas. They may be useful for decadal forecasts. Yearly to decadal forecasts, however, are based on only a few major to great events. Activity along closer faults like that in the two years prior to the 1989 Loma Prieta shock needs to be examined for possible yearly forerunning changes to large plate boundary earthquakes. Geodetic observations are needed to focus on identifying creeping faults close to the San Andreas. The distribution of moderate-size earthquakes increased significantly since 1990 along the Hayward fault but not adjacent to the San Andreas fault to the south of San Francisco compared to what took place in the decades prior to the three major historic earthquakes in the region. It is now clear from a re-examination of the 1989 mainshock that the increased level of moderate-size shocks in the one to two preceding decades occurred on nearby East Bay faults. Double-difference locations of small earthquakes provide structural information about faults in the region, especially their depths. The northern San Andreas fault is divided into several strongly coupled segments based on differences in seismicity.

Downscaling Seasonal Precipitation Forecasts over East Africa with Deep Convolutional Neural Networks

This study assesses the suitability of convolutional neural networks(CNNs) for downscaling precipitation over East Africa in the context of seasonal forecasting. To achieve this, we design a set of experiments that compare different CNN configurations and deployed the best-performing architecture to downscale one-month lead seasonal forecasts of June–July–August–September(JJAS) precipitation from the Nanjing University of Information Science and Technology Climate Forecast System version 1.0(NUIST-CFS1.0) for 1982–2020. We also perform hyper-parameter optimization and introduce predictors over a larger area to include information about the main large-scale circulations that drive precipitation over the East Africa region, which improves the downscaling results. Finally, we validate the raw model and downscaled forecasts in terms of both deterministic and probabilistic verification metrics, as well as their ability to reproduce the observed precipitation extreme and spell indicator indices. The results show that the CNN-based downscaling consistently improves the raw model forecasts, with lower bias and more accurate representations of the observed mean and extreme precipitation spatial patterns. Besides, CNN-based downscaling yields a much more accurate forecast of extreme and spell indicators and reduces the significant relative biases exhibited by the raw model predictions. Moreover, our results show that CNN-based downscaling yields better skill scores than the raw model forecasts over most portions of East Africa. The results demonstrate the potential usefulness of CNN in downscaling seasonal precipitation predictions over East Africa,particularly in providing improved forecast products which are essential for end users.

Growth and Interactions of Multi-Source Perturbations in Convection-Allowing Ensemble Forecasts

This study investigated the growth of forecast errors stemming from initial conditions(ICs),lateral boundary conditions(LBCs),and model(MO)perturbations,as well as their interactions,by conducting seven 36 h convectionallowing ensemble forecast(CAEF)experiments.Two cases,one with strong-forcing(SF)and the other with weak-forcing(WF),occurred over the Yangtze-Huai River basin(YHRB)in East China,were selected to examine the sources of uncertainties associated with perturbation growth under varying forcing backgrounds and the influence of these backgrounds on growth.The perturbations exhibited distinct characteristics in terms of temporal evolution,spatial propagation,and vertical distribution under different forcing backgrounds,indicating a dependence between perturbation growth and forcing background.A comparison of the perturbation growth in different precipitation areas revealed that IC and LBC perturbations were significantly influenced by the location of precipitation in the SF case,while MO perturbations were more responsive to convection triggering and dominated in the WF case.The vertical distribution of perturbations showed that the sources of uncertainties and the performance of perturbations varied between SF and WF cases,with LBC perturbations displaying notable case dependence.Furthermore,the interactions between perturbations were considered by exploring the added values of different source perturbations.For the SF case,the added values of IC,LBC,and MO perturbations were reflected in different forecast periods and different source uncertainties,suggesting that the combination of multi-source perturbations can yield positive interactions.In the WF case,MO perturbations provided a more accurate estimation of uncertainties downstream of the Dabie Mountain and need to be prioritized in the research on perturbation development.

Bias-Corrected Short-Range Ensemble Forecasts for Near-Surface Variables during the Summer Season of 2010 in Northern China

A running mean bias （RMB） correction ap- proach was applied to the forecasts of near-surface variables in a seasonal short-range ensemble forecasting experiment with 57 consecutive cases during summer 2010 in the northern China region. To determine a proper training window length for calculating RMB, window lengths from 2 to 20 days were evaluated, and 16 days was taken as an optimal window length, since it receives most of the benefit from extending the window length. The raw and 16-day RMB corrected ensembles were then evaluated for their ensemble mean forecast skills. The results show that the raw ensemble has obvious bias in all near-surface variables. The RMB correction can remove the bias reasonably well, and generate an unbiased ensemble. The bias correction not only reduces the ensemble mean forecast error, but also results in a better spreaderror relationship. Moreover, two methods for computing calibrated probabilistic forecast （PF） were also evaluated through the 57 case dates： 1） using the relative frequency from the RMB-eorrected ensemble; 2） computing the forecasting probabilities based on a historical rank histogram. The first method outperforms the second one, as it can improve both the reliability and the resolution of the PFs, while the second method only has a small effect on the reliability, indicating the necessity and importance of removing the systematic errors from the ensemble.

Evaluation of Tianji and ECMWF high-resolution precipitation forecasts for extreme rainfall event in Henan in July 2021

The extreme rainfall event of July 17 to 22, 2021 in Henan Province, China, led to severe urban waterlogging and flood disasters. This study investigated the performance of high-resolution weather forecasts in predicting this extreme event and the feasibility of weather forecast-based hydrological forecasts. To achieve this goal, high-resolution precipitation forecasts from the Tianji weather system and the forecast system of the European Centre for Medium-Range Weather Forecasts (ECMWF) were evaluated with the spatial verification metrics of structure, amplitude, and location. The results showed that Tianji weather forecasts accurately predicted the amplitude of 12-h accumulated precipitation with a lead time of 12 h. The location and structure of the rainfall areas in Tianji forecasts were closer to the observations than ECMWF forecasts. Tianji hourly precipitation forecasts were also more accurate than ECMWF hourly forecasts, especially at lead times shorter than 8 h. The precipitation forecasts were used as the inputs to a hydrological model to evaluate their hydrological applications. The results showed that the runoff forecasts driven by Tianji weather forecasts could effectively predict the extreme flood event. The runoff forecasts driven by Tianji forecasts were more accurate than those driven by ECMWF forecasts in terms of amplitude and location. This study demonstrates that high-resolution weather forecasts and corresponding hydrological forecasts can provide valuable information in advance for disaster warnings and leave time for people to act on the event. The results encourage further hydrological applications of high-resolution weather forecasts, such as Tianji weather forecasts, in the future.

Skilful Forecasts of Summer Rainfall in the Yangtze River Basin from November

Variability in the East Asian summer monsoon(EASM)brings the risk of heavy flooding or drought to the Yangtze River basin,with potentially devastating impacts.Early forecasts of the likelihood of enhanced or reduced monsoon rainfall can enable better management of water and hydropower resources by decision-makers,supporting livelihoods and major economic and population centres across eastern China.This paper demonstrates that the EASM is predictable in a dynamical forecast model from the preceding November,and that this allows skilful forecasts of summer mean rainfall in the Yangtze River basin at a lead time of six months.The skill for May–June–July rainfall is of a similar magnitude to seasonal forecasts initialised in spring,although the skill in June–July–August is much weaker and not consistently significant.However,there is some evidence for enhanced skill following El Niño events.The potential for decadal-scale variability in forecast skill is also examined,although we find no evidence for significant variation.

Effects of Dropsonde Data in Field Campaigns on Forecasts of Tropical Cyclones over the Western North Pacific in 2020 and the Role of CNOP Sensitivity

Valuable dropsonde data were obtained from multiple field campaigns targeting tropical cyclones,namely Higos,Nangka,Saudel,and Atsani,over the western North Pacific by the Hong Kong Observatory and Taiwan Central Weather Bureau in 2020.The conditional nonlinear optimal perturbation(CNOP)method has been utilized in real-time to identify the sensitive regions for targeting observations adhering to the procedure of real-time field campaigns for the first time.The observing system experiments were conducted to evaluate the effect of dropsonde data and CNOP sensitivity on TC forecasts in terms of track and intensity,using the Weather Research and Forecasting model.It is shown that the impact of assimilating all dropsonde data on both track and intensity forecasts is case-dependent.However,assimilation using only the dropsonde data inside the sensitive regions displays unanimously positive effects on both the track and intensity forecast,either of which obtains comparable benefits to or greatly reduces deterioration of the skill when assimilating all dropsonde data.Therefore,these results encourage us to further carry out targeting observations for the forecast of tropical cyclones according to CNOP sensitivity.

A Deep Learning Approach for Forecasting Thunderstorm Gusts in the Beijing–Tianjin–Hebei Region

Thunderstorm gusts are a common form of severe convective weather in the warm season in North China,and it is of great importance to correctly forecast them.At present,the forecasting of thunderstorm gusts is mainly based on traditional subjective methods,which fails to achieve high-resolution and high-frequency gridded forecasts based on multiple observation sources.In this paper,we propose a deep learning method called Thunderstorm Gusts TransU-net(TGTransUnet)to forecast thunderstorm gusts in North China based on multi-source gridded product data from the Institute of Urban Meteorology(IUM)with a lead time of 1 to 6 h.To determine the specific range of thunderstorm gusts,we combine three meteorological variables:radar reflectivity factor,lightning location,and 1-h maximum instantaneous wind speed from automatic weather stations(AWSs),and obtain a reasonable ground truth of thunderstorm gusts.Then,we transform the forecasting problem into an image-to-image problem in deep learning under the TG-TransUnet architecture,which is based on convolutional neural networks and a transformer.The analysis and forecast data of the enriched multi-source gridded comprehensive forecasting system for the period 2021–23 are then used as training,validation,and testing datasets.Finally,the performance of TG-TransUnet is compared with other methods.The results show that TG-TransUnet has the best prediction results at 1–6 h.The IUM is currently using this model to support the forecasting of thunderstorm gusts in North China.

Enhancing Deep Learning Soil Moisture Forecasting Models by Integrating Physics-based Models

Accurate soil moisture(SM)prediction is critical for understanding hydrological processes.Physics-based(PB)models exhibit large uncertainties in SM predictions arising from uncertain parameterizations and insufficient representation of land-surface processes.In addition to PB models,deep learning(DL)models have been widely used in SM predictions recently.However,few pure DL models have notably high success rates due to lacking physical information.Thus,we developed hybrid models to effectively integrate the outputs of PB models into DL models to improve SM predictions.To this end,we first developed a hybrid model based on the attention mechanism to take advantage of PB models at each forecast time scale(attention model).We further built an ensemble model that combined the advantages of different hybrid schemes(ensemble model).We utilized SM forecasts from the Global Forecast System to enhance the convolutional long short-term memory(ConvLSTM)model for 1–16 days of SM predictions.The performances of the proposed hybrid models were investigated and compared with two existing hybrid models.The results showed that the attention model could leverage benefits of PB models and achieved the best predictability of drought events among the different hybrid models.Moreover,the ensemble model performed best among all hybrid models at all forecast time scales and different soil conditions.It is highlighted that the ensemble model outperformed the pure DL model over 79.5%of in situ stations for 16-day predictions.These findings suggest that our proposed hybrid models can adequately exploit the benefits of PB model outputs to aid DL models in making SM predictions.

Assimilation of Ocean Surface Wind Data by the HY-2B Satellite in GRAPES: Impacts on Analyses and Forecasts

The ocean surface wind(OSW)data retrieved from microwave scatterometers have high spatial accuracy and represent the only wind data assimilated by global numerical models on the ocean surface,thus playing an important role in improving the forecast skills of global medium-range weather prediction models.To improve the forecast skills of the Global/Regional Assimilation and Prediction System Global Forecast System(GRAPES_GFS),the HY-2B OSW data is assimilated into the GRAPES_GFS four-dimensional variational assimilation(4DVAR)system.Then,the impacts of the HY-2B OSW data assimilation on the analyses and forecasts of GRAPES_GFS are analyzed based on one-month assimilation cycle experiments.The results show that after assimilating the HY-2B OSW data,the analysis errors of the wind fields in the lower-middle troposphere(1000-600 hPa)of the tropics and the southern hemisphere(SH)are significantly reduced by an average rate of about 5%.The impacts of the HY-2B OSW data assimilation on the analysis fields of wind,geopotential height,and temperature are not solely limited to the boundary layer but also extend throughout the entire troposphere after about two days of cycling assimilation.Furthermore,assimilating the HY-2B OSW data can significantly improve the forecast skill of wind,geopotential height,and temperature in the troposphere of the tropics and SH.

Significant wave height forecasts integrating ensemble empirical mode decomposition with sequence-to-sequence model

As wave height is an important parameter in marine climate measurement,its accurate prediction is crucial in ocean engineering.It also plays an important role in marine disaster early warning and ship design,etc.However,challenges in the large demand for computing resources and the improvement of accuracy are currently encountered.To resolve the above mentioned problems,sequence-to-sequence deep learning model(Seq-to-Seq)is applied to intelligently explore the internal law between the continuous wave height data output by the model,so as to realize fast and accurate predictions on wave height data.Simultaneously,ensemble empirical mode decomposition(EEMD)is adopted to reduce the non-stationarity of wave height data and solve the problem of modal aliasing caused by empirical mode decomposition(EMD),and then improves the prediction accuracy.A significant wave height forecast method integrating EEMD with the Seq-to-Seq model(EEMD-Seq-to-Seq)is proposed in this paper,and the prediction models under different time spans are established.Compared with the long short-term memory model,the novel method demonstrates increased continuity for long-term prediction and reduces prediction errors.The experiments of wave height prediction on four buoys show that the EEMD-Seq-to-Seq algorithm effectively improves the prediction accuracy in short-term(3-h,6-h,12-h and 24-h forecast horizon)and long-term(48-h and 72-h forecast horizon)predictions.

Seasonal Forecasts of Precipitation during the First Rainy Season in South China Based on NUIST-CFS1.0

Current dynamical models experience great difficulties providing reliable seasonal forecasts of regional/local rainfall in South China.This study evaluates seasonal forecast skill for precipitation in the first rainy season(FRS,i.e.,April–June)over South China from 1982 to 2020 based on the global real-time Climate Forecast System of Nanjing University of Information Science and Technology(NUIST-CFS1.0,previously known as SINTEX-F).The potential predictability and the practical forecast skill of NUIST-CFS1.0 for FRS precipitation remain low in general.But NUIST-CFS1.0 still performs better than the average of nine international models in terms of correlation coefficient skill in predicting the interannual precipitation anomaly and its related circulation index.NUIST-CFS1.0 captures the anomalous Philippines anticyclone,which transports moisture and heat northward to South China,favoring more precipitation in South China during the FRS.By examining the correlations between sea surface temperature(SST)and FRS precipitation and the Philippines anticyclone,we find that the model reasonably captures SST-associated precipitation and circulation anomalies,which partly explains the predictability of FRS precipitation.A dynamical downscaling model with 30-km resolution forced by the large-scale circulations of the NUIST-CFS1.0 predictions could improve forecasts of the climatological states and extreme precipitation events.Our results also reveal interesting interdecadal changes in the predictive skill for FRS precipitation in South China based on the NUIST-CFS1.0 hindcasts.These results help improve the understanding and forecasts for FRS precipitation in South China.

Artificial Intelligence Technique in Hydrological Forecasts Supporting for Water Resources Management of a Large River Basin in Vietnam

Hydrological forecasting plays an important role in water resource management, supporting socio-economic development and managing water-related risks in river basins. There are many flow forecasting techniques that have been developed several centuries ago, ranging from physical models, physics-based models, conceptual models, and data-driven models. Recently, Artificial Intelligence (AI) has become an advanced technique applied as an effective data-driven model in hydrological forecasting. The main advantage of these models is that they give results with compatible accuracy, and require short computation time, thus increasing forecasting time and reducing human and financial effort. This study evaluates the applicability of machine learning and deep learning in Hanoi water level forecasting where it is controlled for flood management and water supply in the Red River Delta, Vietnam. Accordingly, SANN (machine learning algorithm) and LSTM (deep learning algorithm) were tested and compared with a Physics-Based Model (PBM) for the Red River Delta. The results show that SANN and LSTM give high accuracy. The R-squared coefficient is greater than 0.8, the mean squared error (MSE) is less than 20 cm, the correlation coefficient of the forecast hydrology is greater than 0.9 and the level of assurance of the forecast plan ranges from 80% to 90% in both cases. In addition, the calculation time is much reduced compared to the requirement of PBM, which is its limitation in hydrological forecasting for large river basins such as the Red River in Vietnam. Therefore, SANN and LSTM are expected to help increase lead time, thereby supporting water resource management for sustainable development and management of water-related risks in the Red River Delta.

Seasonal Characteristics of Forecasting Uncertainties in Surface PM_(2.5)Concentration Associated with Forecast Lead Time over the Beijing-Tianjin-Hebei Region

Forecasting uncertainties among meteorological fields have long been recognized as the main limitation on the accuracy and predictability of air quality forecasts.However,the particular impact of meteorological forecasting uncertainties on air quality forecasts specific to different seasons is still not well known.In this study,a series of forecasts with different forecast lead times for January,April,July,and October of 2018 are conducted over the Beijing-Tianjin-Hebei(BTH)region and the impacts of meteorological forecasting uncertainties on surface PM_(2.5)concentration forecasts with each lead time are investigated.With increased lead time,the forecasted PM_(2.5)concentrations significantly change and demonstrate obvious seasonal variations.In general,the forecasting uncertainties in monthly mean surface PM_(2.5)concentrations in the BTH region due to lead time are the largest(80%)in spring,followed by autumn(~50%),summer(~40%),and winter(20%).In winter,the forecasting uncertainties in total surface PM_(2.5)mass due to lead time are mainly due to the uncertainties in PBL heights and hence the PBL mixing of anthropogenic primary particles.In spring,the forecasting uncertainties are mainly from the impacts of lead time on lower-tropospheric northwesterly winds,thereby further enhancing the condensation production of anthropogenic secondary particles by the long-range transport of natural dust.In summer,the forecasting uncertainties result mainly from the decrease in dry and wet deposition rates,which are associated with the reduction of near-surface wind speed and precipitation rate.In autumn,the forecasting uncertainties arise mainly from the change in the transport of remote natural dust and anthropogenic particles,which is associated with changes in the large-scale circulation.

Promising Results Predict Role for Artificial Intelligence in Weather Forecasting

Artificial intelligence(AI)has already demonstrated its proficiency at difficult scientific tasks like predicting how proteins will fold and identifying new astronomical objects in masses of observational data[1].Now,recent results suggest that AI also excels at weather forecasting.For global predictions,GraphCast,an AI system developed by Google subsidiary DeepMind(London,UK),outperforms the state-of-the-art model from the European Centre for Medium-Range Weather Forecasts(ECMWF),providing more accurate projections of variables such as temperature and humidity 90%of the time[2,3].Other AI systems,including Pangu-Weather from the Chinese tech company Huawei(Shenzhen,China)[4],can also match or beat traditional global forecasting models.

Scientific Advances and Weather Services of the China Meteorological Administration’s National Forecasting Systems during the Beijing 2022 Winter Olympics

Since the Beijing 2022 Winter Olympics was the first Winter Olympics in history held in continental winter monsoon climate conditions across complex terrain areas,there is a deficiency of relevant research,operational techniques,and experience.This made providing meteorological services for this event particularly challenging.The China Meteorological Administration(CMA)Earth System Modeling and Prediction Centre,achieved breakthroughs in research on short-and medium-term deterministic and ensemble numerical predictions.Several key technologies crucial for precise winter weather services during the Winter Olympics were developed.A comprehensive framework,known as the Operational System for High-Precision Weather Forecasting for the Winter Olympics,was established.Some of these advancements represent the highest level of capabilities currently available in China.The meteorological service provided to the Beijing 2022 Games also exceeded previous Winter Olympic Games in both variety and quality.This included achievements such as the“100-meter level,minute level”downscaled spatiotemporal resolution and forecasts spanning 1 to 15 days.Around 30 new technologies and over 60 kinds of products that align with the requirements of the Winter Olympics Organizing Committee were developed,and many of these techniques have since been integrated into the CMA’s operational national forecasting systems.These accomplishments were facilitated by a dedicated weather forecasting and research initiative,in conjunction with the preexisting real-time operational forecasting systems of the CMA.This program represents one of the five subprograms of the WMO’s high-impact weather forecasting demonstration project(SMART2022),and continues to play an important role in their Regional Association(RA)II Research Development Project(Hangzhou RDP).Therefore,the research accomplishments and meteorological service experiences from this program will be carried forward into forthcoming highimpact weather forecasting activities.This article provides an overview and assessment of this program and the operational national forecasting systems.