ResoNet:Robust and Explainable ENSO Forecasts with Hybrid Convolution and Transformer Networks

Recent studies have shown that deep learning(DL)models can skillfully forecast El Niño–Southern Oscillation(ENSO)events more than 1.5 years in advance.However,concerns regarding the reliability of predictions made by DL methods persist,including potential overfitting issues and lack of interpretability.Here,we propose ResoNet,a DL model that combines CNN(convolutional neural network)and transformer architectures.This hybrid architecture enables our model to adequately capture local sea surface temperature anomalies as well as long-range inter-basin interactions across oceans.We show that ResoNet can robustly predict ENSO at lead times of 19 months,thus outperforming existing approaches in terms of the forecast horizon.According to an explainability method applied to ResoNet predictions of El Niño and La Niña from 1-to 18-month leads,we find that it predicts the Niño-3.4 index based on multiple physically reasonable mechanisms,such as the recharge oscillator concept,seasonal footprint mechanism,and Indian Ocean capacitor effect.Moreover,we demonstrate for the first time that the asymmetry between El Niño and La Niña development can be captured by ResoNet.Our results could help to alleviate skepticism about applying DL models for ENSO prediction and encourage more attempts to discover and predict climate phenomena using AI methods.

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.

The Forecast Skills and Predictability Sources of Marine Heatwaves in the NUIST-CFS1.0 Hindcasts

Using monthly observations and ensemble hindcasts of the Nanjing University of Information Science and Technology Climate Forecast System(NUIST-CFS1.0) for the period 1983–2020, this study investigates the forecast skill of marine heatwaves(MHWs) over the globe and the predictability sources of the MHWs over the tropical oceans. The MHW forecasts are demonstrated to be skillful on seasonal-annual time scales, particularly in tropical oceans. The forecast skill of the MHWs over the tropical Pacific Ocean(TPO) remains high at lead times of 1–24 months, indicating a forecast better than random chance for up to two years. The forecast skill is subject to the spring predictability barrier of El Nino-Southern Oscillation(ENSO). The forecast skills for the MHWs over the tropical Indian Ocean(TIO), tropical Atlantic Ocean(TAO), and tropical Northwest Pacific(NWP) are lower than that in the TPO. A reliable forecast at lead times of up to two years is shown over the TIO, while a shorter reliable forecast window(less than 17 months) occurs for the TAO and NWP.Additionally, the forecast skills for the TIO, TAO, and NWP are seasonally dependent. Higher skills for the TIO and TAO appear in boreal spring, while a greater skill for the NWP emerges in late summer-early autumn. Further analyses suggest that ENSO serves as a critical source of predictability for MHWs over the TIO and TAO in spring and MHWs over the NWP in summer.

Prediction of ENSO using multivariable deep learning

本文基于残差神经网络和观测数据构建了一套深度学习多因子预报测模型,以改进厄尔尼诺-南方涛动(ENSO)的预报.该模型基于最大信息系数进行因子时空特征提取,并根据泰勒图的评估标准可自动确定关键预报因子进行预报.该模型在超前8个月以内的预报性能要优于当前传统的业务预报模式.2011–2018年间,该模型的预报性能优于多模式集成预报的结果.在超前6个月预报时效上,模型预报相关性可达0.82,标准化后的均方根误差仅为0.58°C,多模式集成预报的相关性和标准化后的均方根误差分别为0.70和0.73°C.该模型春季预报障碍问题有所缓解,并且自动选取的关键预报因子可用于解释热带和副热带热动力过程对于ENSO变化的影响.

Wave forecast in the Atlantic Ocean using a double-stage ConvLSTM network

海浪预报对海上运输安全至关重要.本研究提出了一种涵盖物理信息的深度学习模型Double-stage ConvLSTM(D-ConvLSTM)以改进大西洋的海浪预报.将D-ConvLSTM模型与海浪持续性预测和原始ConvLSTM模型的预测技巧进行对比.结果表明,预测误差随着预测时长的增加而增加.D-ConvLSTM模型在预测准确度方面优于前二者,且第三天预测的均方根误差低于0.4 m,距平相关系数约在0.8.此外,当使用IFS预测风替代再分析风时,能够产生相似的预测效果.这表明D-ConvLSTM模型的预测能力能够与ECMWF-WAM模式相当,且更节省计算资源和时间.

Prediction of Seasonal Tropical Cyclone Activity in the NUIST-CFS1.0 Forecast System

Prediction skill for the seasonal tropical cyclone(TC)activity in the Northern Hemisphere is investigated using the coupled climate forecast system(version 1.0)of Nanjing University of Information Science and Technology(NUISTCFS1.0).This assessment is based on the seven-month(May to November)hindcasts consisting of nine ensemble members during 1982–2019.The predictions are compared with the Japanese 55-year Reanalysis and observed tropical storms in the Northern Hemisphere.The results show that the overall distributions of the TC genesis and track densities in model hindcasts agree well with the observations,although the seasonal mean TC frequency and accumulated cyclone energy(ACE)are underestimated in all basins due to the low resolution(T106)of the atmospheric component in the model.NUIST-CFS1.0 closely predicts the interannual variations of TC frequency and ACE in the North Atlantic(NA)and eastern North Pacific(ENP),which have a good relationship with indexes based on the sea surface temperature.In the western North Pacific(WNP),NUIST-CFS1.0 can closely capture ACE,which is significantly correlated with the El Ni?o–Southern Oscillation(ENSO),while it has difficulty forecasting the interannual variation of TC frequency in this area.When the WNP is further divided into eastern and western subregions,the model displays improved TC activity forecasting ability.Additionally,it is found that biases in predicted TC genesis locations lead to inaccurately represented TC–environment relationships,which may affect the capability of the model in reproducing the interannual variations of TC activity.

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.

印度洋海气相互作用对热带夏季大气环流气候态的影响

利用分辨率较高的SINTEX-F(Scale INTeraction EXperiment-FRCGC)海气耦合模式,进行多组长时间积分模拟和理想试验,分析研究热带印度洋海气耦合对夏季大气环流气候态的影响。主要结果有:(1)热带印度洋海气相互作用使热带东印度洋产生明显的东风变化,使热带中西太平洋赤道北部产生气旋性切变变化。(2)印度洋海气相互作用对大气环流气候态的影响绝大部分由于大气对海气相互作用的响应存在年际变化正负距平不对称性造成,这种年际变化不对称性包括正偶极子与负偶极子的不对称、海盆宽度正异常与海盆宽度负异常的不对称。(3)年际和季节内两种时间尺度海气相互作用对印度洋关键区大气环流平均态都有影响,约各占60%、40%;季节内尺度海气相互作用对太平洋近赤道区大气环流平均态有重要影响;年际尺度海气相互作用对太平洋赤道外地区大气环流平均态有重要影响。热带印度洋年际尺度、季节内尺度海气相互作用对大气环流气候态的影响,都存在年际变化以及年际变化正负距平不对称性。这两种尺度海气相互作用主要通过年际变化正负距平不对称性而对大气环流平均态产生影响。

Toward Understanding the Extreme Floods over Yangtze River Valley in June−July 2020:Role of Tropical Oceans

The extreme floods in the Middle/Lower Yangtze River Valley(MLYRV)during June−July 2020 caused more than 170 billion Chinese Yuan direct economic losses.Here,we examine the key features related to this extreme event and explore relative contributions of SST anomalies in different tropical oceans.Our results reveal that the extreme floods over the MLYRV were tightly related to a strong anomalous anticyclone persisting over the western North Pacific,which brought tropical warm moisture northward that converged over the MLYRV.In addition,despite the absence of a strong El Niño in 2019/2020 winter,the mean SST anomaly in the tropical Indian Ocean during June−July 2020 reached its highest value over the last 40 years,and 43%(57%)of it is attributed to the multi-decadal warming trend(interannual variability).Based on the NUIST CFS1.0 model that successfully predicted the wet conditions over the MLYRV in summer 2020 initiated from 1 March 2020(albeit the magnitude of the predicted precipitation was only about one-seventh of the observed),sensitivity experiment results suggest that the warm SST condition in the Indian Ocean played a dominant role in generating the extreme floods,compared to the contributions of SST anomalies in the Maritime Continent,central and eastern equatorial Pacific,and North Atlantic.Furthermore,both the multi-decadal warming trend and the interannual variability of the Indian Ocean SSTs had positive impacts on the extreme floods.Our results imply that the strong multi-decadal warming trend in the Indian Ocean needs to be taken into consideration for the prediction/projection of summer extreme floods over the MLYRV in the future.

印度洋海温年际异常与热带夏季季节内振荡的关系及其数值模拟研究

利用观测分析资料和SINTEX-F海气耦合长时间(70年)数值模拟结果,分析了印度洋海温年际异常与热带夏季季节内振荡(BSISO)各种传播模态之间关系及其物理过程。结果表明,印度洋海温年际异常与热带BSISO关系密切,当印度洋为正(负)偶极子情况,中东印度洋北传BSISO减弱(加强);当印度洋为正(负)海盆异常(BWA)情况,印度洋西太平洋赤道地区(40°E -180°)东传BSISO加强(减弱)。印度洋海温年际变化通过大气环流背景场和BSISO结构影响热带BSISO不同传播模态强度的年际变化。在负(正)偶极子年夏季,由于对流层大气垂直东风切变加强(减弱),对流扰动北侧的正压涡度、边界层水汽辐合加强更明显(不明显),导致形成BSISO较强(弱)的经向不对称结构,因此北传BSISO偏强(减弱)。印度洋BWA模态通过影响赤道西风背景以及海气界面热力交换,导致赤道东传BSISO强度产生变化。在正BWA年夏季,赤道地区西风较明显,当季节内振荡叠加在这种西风背景下,扰动中心的东侧(西侧)风速减弱(加强)更明显,海面蒸发及蒸发潜热减弱(加强)更明显,导致扰动中心的东侧(西侧)海温升高(降低)幅度更大,从而使边界层产生辐合(辐散)更强、水汽更多(少),因此赤道东传BSISO偏强;而在负BWA年,赤道地区西风背景减弱,以上物理过程受削弱使赤道东传BSISO偏弱。

Seasonal Prediction of Summer Precipitation over East Africa Using NUIST-CFS1.0

East Africa is particularly vulnerable to precipitation variability, as the livelihood of much of the population depends on rainfed agriculture. Seasonal forecasts of the precipitation anomalies, when skillful, can therefore improve implementation of coping mechanisms with respect to food security and water management. This study assesses the performance of Nanjing University of Information Science and Technology Climate Forecast System version 1.0(NUISTCFS1.0) on forecasting June–September(JJAS) seasonal precipitation anomalies over East Africa. The skill in predicting the JJAS mean precipitation initiated from 1 May for the period of 1982–2019 is evaluated using both deterministic and probabilistic verification metrics on grid cell and over six distinct clusters. The results show that NUIST-CFS1.0 captures the spatial pattern of observed seasonal precipitation climatology, albeit with dry and wet biases in a few parts of the region. The model has positive skill across a majority of Ethiopia, Kenya, Uganda, and Tanzania, whereas it doesn’t exceed the skill of climatological forecasts in parts of Sudan and southeastern Ethiopia. Positive forecast skill is found over regions where the model shows better performance in reproducing teleconnections related to oceanic SST. The prediction performance of NUIST-CFS1.0 is found to be on a level that is potentially useful over a majority of East Africa.

An evaluation of the Arctic clouds and surface radiative fluxes in CMIP6 models

To assess the performances of state-of-the-art global climate models on simulating the Arctic clouds and surface radiation balance,the 2001–2014 Arctic Basin surface radiation budget,clouds,and the cloud radiative effects(CREs)in 22 coupled model intercomparison project 6(CMIP6)models are evaluated against satellite observations.For the results from CMIP6 multi-model mean,cloud fraction(CF)peaks in autumn and is lowest in winter and spring,consistent with that from three satellite observation products(Cloud Sat-CALIPSO,CERESMODIS,and APP-x).Simulated CF also shows consistent spatial patterns with those in observations.However,almost all models overestimate the CF amount throughout the year when compared to CERES-MODIS and APP-x.On average,clouds warm the surface of the Arctic Basin mainly via the longwave(LW)radiation cloud warming effect in winter.Simulated surface energy loss of LW is less than that in CERES-EBAF observation,while the net surface shortwave(SW)flux is underestimated.The biases may result from the stronger cloud LW warming effect and SW cooling effect from the overestimated CF by the models.These two biases compensate each other,yielding similar net surface radiation flux between model output(3.0 W/m2)and CERES-EBAF observation(6.1 W/m2).During 2001–2014,significant increasing trend of spring CF is found in the multi-model mean,consistent with previous studies based on surface and satellite observations.Although most of the 22 CMIP6 models show common seasonal cycles of CF and liquid water path/ice water path(LWP/IWP),large inter-model spreads exist in the amounts of CF and LWP/IWP throughout the year,indicating the influences of different cloud parameterization schemes used in different models.Cloud Feedback Model Intercomparison Project(CFMIP)observation simulator package(COSP)is a great tool to accurately assess the performance of climate models on simulating clouds.More intuitive and credible evaluation results can be obtained based on the COSP model output.In the future,with the release of more COSP output of CMIP6 models,it is expected that those inter-model spreads and the model-observation biases can be substantially reduced.Longer term active satellite observations are also necessary to evaluate models’cloud simulations and to further explore the role of clouds in the rapid Arctic climate changes.

Distinct Evolution of the SST Anomalies in the Far Eastern Pacific between the 1997/98 and 2015/16 Extreme El Niños

The 2015/16 El Niño displayed a distinct feature in the SST anomalies over the far eastern Pacific(FEP)compared to the 1997/98 extreme case.In contrast to the strong warm SST anomalies in the FEP in the 1997/98 event,the FEP warm SST anomalies in the 2015/16 El Niño were modest and accompanied by strong southeasterly wind anomalies in the southeastern Pacific.Exploring possible underlying causes of this distinct difference in the FEP may improve understanding of the diversity of extreme El Niños.Here,we employ observational analyses and numerical model experiments to tackle this issue.Mixed-layer heat budget analysis suggests that compared to the 1997/98 event,the modest FEP SST warming in the 2015/16 event was closely related to strong vertical upwelling,strong westward current,and enhanced surface evaporation,which were caused by the strong southeasterly wind anomalies in the southeastern Pacific.The strong southeasterly wind anomalies were initially triggered by the combined effects of warm SST anomalies in the equatorial central and eastern Pacific(CEP)and cold SST anomalies in the southeastern subtropical Pacific in the antecedent winter,and then sustained by the warm SST anomalies over the northeastern subtropical Pacific and CEP.In contrast,southeasterly wind anomalies in the 1997/98 El Niño were partly restrained by strong anomalously negative sea level pressure and northwesterlies in the northeast flank of the related anomalous cyclone in the subtropical South Pacific.In addition,the strong southeasterly wind and modest SST anomalies in the 2015/16 El Niño may also have been partly related to decadal climate variability.

Seasonal Predictions of Summer Precipitation in the Middle-lower Reaches of the Yangtze River with Global and Regional Models Based on NUIST-CFS1.0

Accurate prediction of the summer precipitation over the middle and lower reaches of the Yangtze River(MLYR)is of urgent demand for the local economic and societal development.This study assesses the seasonal forecast skill in predicting summer precipitation over the MLYR region based on the global Climate Forecast System of Nanjing University of Information Science and Technology(NUIST-CFS1.0,previously SINTEX-F).The results show that the model can provide moderate skill in predicting the interannual variations of the MLYR rainbands,initialized from 1 March.In addition,the nine-member ensemble mean can realistically reproduce the links between the MLYR precipitation and tropical sea surface temperature(SST)anomalies,but the individual members show great discrepancies,indicating large uncertainty in the forecasts.Furthermore,the NUIST-CFS1.0 can predict five of the seven extreme summer precipitation anomalies over the MLYR during 1982-2020,albeit with underestimated magnitudes.The Weather Forecast and Research(WRF)downscaling hindcast experiments with a finer resolution of 30 km,which are forced by the large-scale information of the NUIST-CFS1.0 predictions with a spectral nudging method,display improved predictions of the extreme summer precipitation anomalies to some extent.However,the performance of the downscaling predictions is highly dependent on the global model forecast skill,suggesting that further improvements on both the global and regional climate models are needed.

Verification of Seasonal Prediction by the Upgraded China Multi-Model Ensemble Prediction System (CMMEv2.0)

Based on a combination of six Chinese climate models and three international operational models,the China multimodel ensemble(CMME)prediction system has been upgraded into its version 2(CMMEv2.0)at the National Climate Centre(NCC)of the China Meteorological Administration(CMA)by including new model members and expanding prediction products.A comprehensive assessment of the performance of the upgraded CMME during its hindcast(1993–2016)and real-time prediction(2021–present)periods is conducted in this study.The results demonstrate that CMMEv2.0 outperforms all the individual models by capturing more realistic equatorial sea surface temperature(SST)variability.It exhibits better prediction skills for precipitation and 2-m temperature anomalies,and the improvements in prediction skill of CMMEv2.0 are significant over East Asia.The superiority of CMMEv2.0 can be attributed to its better projection of El Niño–Southern Oscillation(ENSO;with the temporal correlation coefficient score for Niño3.4 index reaching 0.87 at 6-month lead)and ENSO-related teleconnections.As for the real-time prediction in recent three years,CMMEv2.0 has also yielded relatively stable skills;it successfully predicted the primary rainbelt over northern China in summers of 2021–2023 and the warm conditions in winters of 2022/2023.Beyond that,ensemble sampling experiments indicate that the CMMEv2.0 skills become saturated after the ensemble model number increased to 5–6,indicating that selection of only an optimal subgroup of ensemble models could benefit the prediction performance,especially over the extratropics,yet the underlying reasons await future investigation.

A Review of Research on Tropical Air–Sea Interaction, ENSO Dynamics, and ENSO Prediction in China

Remarkable progress has been made in observations,theories,and simulations of the ocean–atmosphere system,laying a solid foundation for the improvement of short-term climate prediction,among which Chinese scientists have made important contributions.This paper reviews Chinese research on tropical air–sea interaction,ENSO dynamics,and ENSO prediction in the past 70 years.Review of the tropical air–sea interaction mainly focuses on four aspects:characteristics of the tropical Pacific climate system and ENSO;main modes of tropical Indian Ocean SSTs and their interactions with the tropical Pacific;main modes of tropical Atlantic SSTs and inter-basin interactions;and influences of the mid–high-latitude air–sea system on ENSO.Review of the ENSO dynamics involves seven aspects:fundamental theories of ENSO;diagnosis and simulation of ENSO;the two types of ENSO;mechanisms of ENSO initiation;the interactions between ENSO and other phenomena;external forcings and teleconnections;and climate change and the ENSO response.The ENSO prediction part briefly summarizes the dynamical–statistical methods used in ENSO prediction,as well as the operational ENSO prediction systems and their applications.Lastly,we discuss some of the issues in these areas that are in need of further study.

Detection and Attribution of Changes in Summer Compound Hot and Dry Events over Northeastern China with CMIP6 Models

Northeastern China has experienced a significant increase in summer compound hot and dry events(CHDEs),posing a threat to local agricultural production and sustainable development.This study investigates the detectable anthropogenic signal in the long-term trend of CHDE and quantifies the contribution of different external forcings.A probability-based index(PI)is constructed through the joint probability distribution to measure the severity of CHDE,with lower values representing more severe cases.Response of CHDE to external forcing was assessed with simulations from the Coupled Model Intercomparison Project phase 6(CMIP6).The results show a significant increase in the severity of CHDE over northeastern China during the past decades.The trend of regional averaged PI is-0.28(90%confidence interval:-0.43 to-0.13)per 54 yr and it is well reproduced in the historical forcing simulations.The attribution method of optimal fingerprinting was firstly applied to a two-signal configuration with anthropogenic forcing and natural forcing;the anthropogenic impact was robustly detected and it explains most of the observed trend of PI.Similarly,three-signal analysis further demonstrated that the anthropogenic greenhouse gases dominantly contribute to the observed change,while the anthropogenic aerosol and natural forcing have almost no contribution to the observed changes.For a compound event concurrently exceeding the 95 th percentile of surface air temperature and precipitation reversal in the current period,its likelihood exhibits little change at 1.5℃global warming,but almost doubled at 2.0℃global warming.

Extreme fire weather is the major driver of severe bushfires in southeast Australia

In Australia,the proportion of forest area that burns in a typical fire season is less than for other vegetation types.However,the 2019-2020 austral spring-summer was an exception,with over four times the previous maximum area burnt in southeast Australian temperate forests.Temperate forest fires have extensive socio-economic,human health,greenhouse gas emissions,and biodiversity impacts due to high fire intensities.A robust model that identifies driving factors of forest fires and relates impact thresholds to fire activity at regional scales would help land managers and fire-fighting agencies prepare for potentially hazardous fire in Australia.Here,we developed a machine-learning diagnostic model to quantify nonlinear relationships between monthly burnt area and biophysical factors in southeast Australian forests for 2001-2020 on a 0.25°grid based on several biophysical parameters,notably fire weather and vegetation productivity.Our model explained over 80%of the variation in the burnt area.We identified that burnt area dynamics in southeast Australian forest were primarily controlled by extreme fire weather,which mainly linked to fluctuations in the Southern Annular Mode(SAM)and Indian Ocean Dipole(IOD),with a relatively smaller contribution from the central Pacific El Niño Southern Oscillation(ENSO).Our fire diagnostic model and the non-linear relationships between burnt area and environmental covariates can provide useful guidance to decision-makers who manage preparations for an upcoming fire season,and model developers working on improved early warning systems for forest fires.

Improving global weather and ocean wave forecast with large artificial intelligence models

The rapid advancement of artificial intelligence technologies,particularly in recent years,has led to the emergence of several large parameter artificial intelligence weather forecast models.These models represent a significant breakthrough,overcoming the limitations of traditional numerical weather prediction models and indicating the emergence of profound potential tools for atmosphere-ocean forecasts.This study explores the evolution of these advanced artificial intelligence forecast models,and based on the identified commonalities,proposes the“Three Large Rules”for large weather forecast models:a large number of parameters,a large number of predictands,and large potential applications.We discuss the capacity of artificial intelligence to revolutionize numerical weather prediction,briefly outlining the underlying reasons for the significant improvement in weather forecasting.While acknowledging the high accuracy,computational efficiency,and ease of deployment of large artificial intelligence forecast models,we also emphasize the irreplaceable values of traditional numerical forecasts and explore the challenges in the future development of large-scale artificial intelligence atmosphere-ocean forecast models.We believe that the optimal future of atmosphere-ocean weather forecast lies in achieving a seamless integration of artificial intelligence and traditional numerical models.Such a synthesis is anticipated to offer a more advanced and reliable approach for improved atmosphere-ocean forecasts.Finally,we illustrate how forecasters can leverage the large weather forecast models through an example by building an artificial intelligence model for global ocean wave forecast.