Search

Cloud-radiative interaction (CRI) is a fundamental process that modulates tropical circulation and intraseasonal variability, including the Madden-Julian Oscillation (MJO). In this study, we investigate how the mean state of the tropical atmosphere and MJO respond to CRI intensity changes and provide insights into the underlying mechanisms, using the aquaplanet configuration in CESM2. By enhancing CRI through tuning the DCS parameter (an auto-conversion threshold size in Morrison and Gettelman (2008) cloud microphysics scheme), we demonstrate that DCS-induced CRI intensification is linked to a warmer troposphere, increased tropical moisture, strengthened Hadley Cell (HC), stronger trade winds, and a stronger equatorward intertropical convergence zone (ITCZ) with more clouds and precipitation, reflecting stronger cloud-radiation-circulation feedback. The intensified CRI also leads to the intensification and slower propagation of the simulated MJO-like mode despite the MJO-like signals becoming less distinguishable from the background due to the influence of other waves. The MJO intensification is likely associated with the mean state changes that support the development of deep convection. Moreover, the CRI itself, especially the interaction with the longwave radiation, also directly influences the MJO’s maintenance and propagation, more contributing to the maintenance of column moist static energy (MSE) and deceleration of its eastward propagation on intraseasonal timescales.

Tropical cyclones (TCs) are expected to produce more intense precipitation under global warming. However, substantial uncertainties exist in the projection of coarse-resolution global climate models. Here, we use deep learning to aid targeted cloud-resolving simulations of extreme TCs. Contrary to the Clausius-Clapeyron (CC) scaling, which indicates a 7% moisture increase per K warming, our simulations reveal more complex responses of TC rainfall. TCs will not intensify via strengthened updrafts but through the expansion of deep convective cores with suppression of shallow cumulus and congestus. Consequently, while localized hourly rainfall may adhere to the CC scaling, precipitation accumulation over city-sized areas could surge by 18%K-1. This super-CC intensification due to changing TC structure has profound implications for floods and landslides. Estimations using Hong Kong’s slope data confirm this concern and suggest an up to 215% increase in landslide risks with 4-K warming, highlighting amplified threats from compound disasters under climate change. © The Author(s) 2024.

Aim Soil microbes can significantly influence restoration outcomes via interaction with plant community assembly processes, yet knowledge about variation in soil microbial communities – in particular functional variation – during oldfield succession is limited. Methods We divided a well-dated successional chronosequence on the Tibetan Plateau into five stages: stage 1 (continued arable land), stage 2 (arable abandoned for 2 years), stage 3 (arable abandoned for 10 years), stage 4 (arable abandoned for 20 years), and natural grassland. We investigated the changes in taxonomic and functional composition of bacterial and fungal communities in these successional stages. Results The richness of bacterial and fungal communities had a unimodal relationship with successional age, as the both were initially low and decreased again in natural grasslands. These changes were more correlated to soil properties. For both bacterial and fungal communities, taxonomic similarity to natural grasslands increased monotonously with successional age. The functional composition of bacterial communities shifted with successional age towards increased importance of strains involved in the C cycle rather than the N cycle, due to higher plant richness. For fungal communities, saprotrophs showed an increasing trend with successional age although low relative abundance in natural grasslands, which was regulated by belowground biomass. Symbiotrophs did not change during succession, but pathotrophic fungal relative abundance decreased rapidly after agricultural abandonment because of increased plant richness. Conclusions Overall, twenty years of oldfield succession did not appear to restore richness of soil bacterial and fungal communities to the levels of natural grasslands. Community taxonomic and functional composition in successional stages up to 20 y old were also different from natural grasslands. Our results suggest more important role of plant com, Aim: Soil microbes can significantly influence restoration outcomes via interaction with plant community assembly processes, yet knowledge about variation in soil microbial communities – in particular functional variation – during oldfield succession is limited. Methods: We divided a well-dated successional chronosequence on the Tibetan Plateau into five stages: stage 1 (continued arable land), stage 2 (arable abandoned for 2 years), stage 3 (arable abandoned for 10 years), stage 4 (arable abandoned for 20 years), and natural grassland. We investigated the changes in taxonomic and functional composition of bacterial and fungal communities in these successional stages. Results: The richness of bacterial and fungal communities had a unimodal relationship with successional age, as the both were initially low and decreased again in natural grasslands. These changes were more correlated to soil properties. For both bacterial and fungal communities, taxonomic similarity to natural grasslands increased monotonously with successional age. The functional composition of bacterial communities shifted with successional age towards increased importance of strains involved in the C cycle rather than the N cycle, due to higher plant richness. For fungal communities, saprotrophs showed an increasing trend with successional age although low relative abundance in natural grasslands, which was regulated by belowground biomass. Symbiotrophs did not change during succession, but pathotrophic fungal relative abundance decreased rapidly after agricultural abandonment because of increased plant richness. Conclusions: Overall, twenty years of oldfield succession did not appear to restore richness of soil bacterial and fungal communities to the levels of natural grasslands. Community taxonomic and functional composition in successional stages up to 20 y old were also different from natural grasslands. Our results suggest more important role of plant community than microbial community to s

Time series forecasting is widely used in extensive applications, such as traffic planning and weather forecasting. However, real-world time series usually present intricate temporal variations, making forecasting extremely challenging. Going beyond the mainstream paradigms of plain decomposition and multiperiodicity analysis, we analyze temporal variations in a novel view of multiscale-mixing, which is based on an intuitive but important observation that time series present distinct patterns in different sampling scales. The microscopic and the macroscopic information are reflected in fine and coarse scales respectively, and thereby complex variations can be inherently disentangled. Based on this observation, we propose TimeMixer as a fully MLP-based architecture with Past-Decomposable-Mixing (PDM) and Future-Multipredictor-Mixing (FMM) blocks to take full advantage of disentangled multiscale series in both past extraction and future prediction phases. Concretely, PDM applies the decomposition to multiscale series and further mixes the decomposed seasonal and trend components in fine-to-coarse and coarse-to-fine directions separately, which successively aggregates the microscopic seasonal and macroscopic trend information. FMM further ensembles multiple predictors to utilize complementary forecasting capabilities in multiscale observations. Consequently, TimeMixer is able to achieve consistent state-of-the-art performances in both long-term and short-term forecasting tasks with favorable run-time efficiency.

This paper surveys and organizes research works on medical dialog systems, which is an important yet challenging task. Although these systems have been surveyed in the medical community from an application perspective, a systematic review from a rigorous technical perspective has to date remained noticeably absent. As a result, an overview of the categories, methods, and evaluation of medical dialogue systems remain limited and underspecified, hindering the further improvement of this area. To fill this gap, we investigate an initial pool of 325 papers from well-known computer science, and natural language processing conferences and journals, and make an overview. Recently, large language models have shown strong model capacity on downstream tasks, which also reshaped medical dialog systems' foundation. Despite the alluring practical application value, current medical dialogue systems still suffer from problems. To this end, this paper lists the grand challenges of medical dialog systems, especially of large language models.

Six-dimensional movable antenna (6DMA) is an effective solution for enhancing wireless network capacity through the adjustment of both 3D positions and 3D rotations of distributed antennas/antenna surfaces. Although freely positioning/rotating 6DMA surfaces offers the greatest flexibility and thus highest capacity improvement, its implementation may be challenging in practice due to the drastic architecture change required for existing base stations (BSs), which predominantly adopt fixed-position antenna (FPA) arrays (e.g., sector antenna arrays). Thus, we introduce in this letter a new BS architecture called hybrid fixed and movable antennas (HFMA), which consists of both conventional FPA arrays and position/rotation-adjustable 6DMA surfaces. For ease of implementation, we consider that all 6DMA surfaces can rotate along a circular track above the FPA arrays. We aim to maximize the network capacity via optimizing the rotation angles of all 6DMA surfaces based on the users' spatial distribution. Since this problem is combinatorial and its optimal solution requires prohibitively high computational complexity via exhaustive search, we propose an alternative adaptive Markov Chain Monte Carlo based method to solve it more efficiently. Finally, we present simulation results that show significant performance gains achieved by our proposed design over various benchmark schemes.

The intense and moist winds in a tropical cyclone (TC) environment can produce strong mountain waves and enhanced precipitation over complex terrain, yet few studies have investigated how the orographic precipitation in a TC environment might respond to global warming. Here, we use large-eddy simulation to estimate the global warming- induced change in the precipitation near an idealized mountain (1 km maximum height) with pseudo global warming. Two regions exhibit enhanced precipitation, one over the mountain and the other in the downstream region 25-45 km away from the mountain. The enhanced precipitation in both regions is related to the seeder-feeder mechanism, although the enhancement in the downstream regions differs from the conventional definition and is referred to as the pseudo-seeder-feeder mechanism (PSF). In the PSF, mountain waves generate an intense cloud formation center in the midtroposphere above the lee slope, and the resulting hydrometeors drift downstream, intensifying downstream convection when they fall into proper locations. Under warming, the overmountain precipitation maximum exhibits minimal changes, while the downstream precipitation maximum exhibits a large sensitivity of 18% K21. The small sensitivity of the first precipitation peak is due to the canceling effects of thermodynamic and dynamic changes. The large sensitivity in the downstream region is mainly due to the strengthening of the wave-induced midtroposphere cloud formation center, which supplies more hydrometeors to the downstream region and enhances precipitation efficiency through the enhanced PSF mechanism. However, the downstream precipitation sensitivity varies with mountain geometry. Higher mountain height enhances precipitation but lowers the sensitivity to warming.SIGNIFICANCE STATEMENT: The combination of typhoon environment and orography can produce intense precipitation and thereby severe flooding risks. Here, we investigate the global warming response of orographic

Previous studies on convective precipitation forecasting in South China have focused on the effects of multiscale dynamics and microphysics parameterizations. However, limited investigation has been conducted on how uncertainty in aerosol data might cause errors in quantitative precipitation forecast for South China's coastal convection. In this case study, we evaluated the impact of aerosol uncertainties on South China's severe coastal convection using convection-permitting simulations. We estimated the variability range of aerosol concentrations with observations for the pre-summer months. The simulation results suggest that the rainfall pattern and intensity change notably when aerosol concentrations are varied. Decreasing the concentration of water-friendly (WF) aerosols intensifies precipitation through reduced cloud water number concentration and increased droplet size. Increasing the concentration of ice-friendly (IF) aerosols results in up to 40% increase in vertical velocity and latent heat compared to minimal IF aerosol condition, by enhancing the heterogeneous process and dynamically intensifying convection. Consequently, the simulation with minimal WF and maximal IF aerosol concentrations shows prolonged intense precipitation over the entire life cycle of convection. However, when both WF and IF aerosols are set to minimal concentrations, the simulation produces the maximum peak rainfall rate, which is about 50% stronger than the simulation with the climatological mean concentration, due to an enhanced homogeneous process that results in a higher ice concentration and more efficient ice-phase precipitation growth. Meanwhile, variation in aerosol concentration affects convection initiation (CI), with a lower concentration of WF aerosol inducing earlier CI onset. Decreasing hygroscopicity leads to higher precipitation.

The development of machine learning (ML) techniques enables data-driven parameterizations, which have been investigated in many recent studies. Some investigations suggest that a priori-trained ML models exhibit satisfying accuracy during training but poor performance when coupled to dynamical cores and tested. Here we use the evolution of the barotropic vorticity equation (BVE) with periodically reinforced shear instability as a prototype problem to develop and evaluate a model-consistent training strategy, which employs a numerical solver supporting automatic differentiation and includes the solver in the loss function for training ML-based subgrid-scale (SGS) turbulence models. This approach enables the interaction between the dynamical core and the ML-based parameterization during the model training phase. The BVE model was run at low, high, and ultrahigh (truth) resolutions. Our training dataset contains only a short period of coarsened high-resolution simulations. However, given initial conditions long after the training dataset time, the trained SGS model can still significantly increase the effective lead time of the BVE model running at the low resolution by up to 50% relative to the BVE simulation without an SGS model. We also tested using a covariance matrix to normalize the loss function and found it can notably boost the performance of the ML parameterization. The SGS model’s performance is further improved by conducting transfer learning using a limited number of discontinuous observations, increasing the forecast lead-time improvement to 73%. This study demonstrates a potential pathway to using machine learning to enhance the prediction skills of our climate and weather models.

Most medical dialogue systems assume that patients have clear goals (medicine querying, surgical operation querying, etc.) before medical consultation. However, in many real scenarios, due to the lack of medical knowledge, it is usually difficult for patients to determine clear goals with all necessary slots. In this paper, we identify this challenge as how to construct medical consultation dialogue systems to help patients clarify their goals. To mitigate this challenge, we propose a novel task and create a human-to-human mixed-type medical consultation dialogue corpus, termed MidMed, covering five dialogue types: task-oriented dialogue for diagnosis, recommendation, knowledge-grounded dialogue, QA, and chitchat. MidMed covers four departments (otorhinolaryngology, ophthalmology, skin, and digestive system), with 8,175 dialogues. Furthermore, we build baselines on MidMed and propose an instruction-guiding medical dialogue generation framework, termed InsMed, to address this task. Experimental results show the effectiveness of InsMed., Comment: Accepted by ACL 2023 main conference. The first two authors contributed equally to this work

Large language models have shown astonishing performance on a wide range of reasoning tasks. In this paper, we investigate whether they could reason about real-world events and help improve the prediction performance of event sequence models. We design LAMP, a framework that integrates a large language model in event prediction. Particularly, the language model performs abductive reasoning to assist an event sequence model: the event model proposes predictions on future events given the past; instructed by a few expert-annotated demonstrations, the language model learns to suggest possible causes for each proposal; a search module finds out the previous events that match the causes; a scoring function learns to examine whether the retrieved events could actually cause the proposal. Through extensive experiments on several challenging real-world datasets, we demonstrate that our framework -- thanks to the reasoning capabilities of large language models -- could significantly outperform the state-of-the-art event sequence models., Comment: NeurIPS 2023 camera-ready

METHODS: First, a set of evaluation criteria is designed based on a comprehensive literature review. Second, existing candidate criteria are optimized for using a Delphi method by five experts in medicine and engineering. Third, three clinical experts design a set of medical datasets to interact with LLMs. Finally, benchmarking experiments are conducted on the datasets. The responses generated by chatbots based on LLMs are recorded for blind evaluations by five licensed medical experts. RESULTS: The obtained evaluation criteria cover medical professional capabilities, social comprehensive capabilities, contextual capabilities, and computational robustness, with sixteen detailed indicators. The medical datasets include twenty-seven medical dialogues and seven case reports in Chinese. Three chatbots are evaluated, ChatGPT by OpenAI, ERNIE Bot by Baidu Inc., and Doctor PuJiang (Dr. PJ) by Shanghai Artificial Intelligence Laboratory. Experimental results show that Dr. PJ outperforms ChatGPT and ERNIE Bot in both multiple-turn medical dialogue and case report scenarios.

The rapid rise in cloud computing has resulted in an alarming increase in data centers' carbon emissions, which now accounts for >3% of global greenhouse gas emissions, necessitating immediate steps to combat their mounting strain on the global climate. An important focus of this effort is to improve resource utilization in order to save electricity usage. Our proposed Full Scaling Automation (FSA) mechanism is an effective method of dynamically adapting resources to accommodate changing workloads in large-scale cloud computing clusters, enabling the clusters in data centers to maintain their desired CPU utilization target and thus improve energy efficiency. FSA harnesses the power of deep representation learning to accurately predict the future workload of each service and automatically stabilize the corresponding target CPU usage level, unlike the previous autoscaling methods, such as Autopilot or FIRM, that need to adjust computing resources with statistical models and expert knowledge. Our approach achieves significant performance improvement compared to the existing work in real-world datasets. We also deployed FSA on large-scale cloud computing clusters in industrial data centers, and according to the certification of the China Environmental United Certification Center (CEC), a reduction of 947 tons of carbon dioxide, equivalent to a saving of 1538,000 kWh of electricity, was achieved during the Double 11 shopping festival of 2022, marking a critical step for our company's strategic goal towards carbon neutrality by 2030.

It is becoming increasingly emphasis on the importance of LLM participating in clinical diagnosis decision-making. However, the low specialization refers to that current medical LLMs can not provide specific medical advice, which are more like a medical Q\&A. And there is no suitable clinical guidance tree data set that can be used directly with LLM. To address this issue, we first propose LLM-executavle clinical guidance tree(CGT), which can be directly used by large language models, and construct medical diagnostic decision-making dataset (MedDM), from flowcharts in clinical practice guidelines. We propose an approach to screen flowcharts from medical literature, followed by their identification and conversion into standardized diagnostic decision trees. Constructed a knowledge base with 1202 decision trees, which came from 5000 medical literature and covered 12 hospital departments, including internal medicine, surgery, psychiatry, and over 500 diseases.Moreover, we propose a method for reasoning on LLM-executable CGT and a Patient-LLM multi-turn dialogue framework.

Neural Temporal Point Processes (TPPs) are the prevalent paradigm for modeling continuous-time event sequences, such as user activities on the web and financial transactions. In real-world applications, event data is typically received in a \emph{streaming} manner, where the distribution of patterns may shift over time. Additionally, \emph{privacy and memory constraints} are commonly observed in practical scenarios, further compounding the challenges. Therefore, the continuous monitoring of a TPP to learn the streaming event sequence is an important yet under-explored problem. Our work paper addresses this challenge by adopting Continual Learning (CL), which makes the model capable of continuously learning a sequence of tasks without catastrophic forgetting under realistic constraints. Correspondingly, we propose a simple yet effective framework, PromptTPP\footnote{Our code is available at {\small \url{ https://github.com/yanyanSann/PromptTPP}}}, by integrating the base TPP with a continuous-time retrieval prompt pool. The prompts, small learnable parameters, are stored in a memory space and jointly optimized with the base TPP, ensuring that the model learns event streams sequentially without buffering past examples or task-specific attributes. We present a novel and realistic experimental setup for modeling event streams, where PromptTPP consistently achieves state-of-the-art performance across three real user behavior datasets., Comment: NeurIPS 2023 camera ready version

Time series forecasting holds significant importance in many real-world dynamic systems and has been extensively studied. Unlike natural language process (NLP) and computer vision (CV), where a single large model can tackle multiple tasks, models for time series forecasting are often specialized, necessitating distinct designs for different tasks and applications. While pre-trained foundation models have made impressive strides in NLP and CV, their development in time series domains has been constrained by data sparsity. Recent studies have revealed that large language models (LLMs) possess robust pattern recognition and reasoning abilities over complex sequences of tokens. However, the challenge remains in effectively aligning the modalities of time series data and natural language to leverage these capabilities. In this work, we present Time-LLM, a reprogramming framework to repurpose LLMs for general time series forecasting with the backbone language models kept intact. We begin by reprogramming the input time series with text prototypes before feeding it into the frozen LLM to align the two modalities. To augment the LLM's ability to reason with time series data, we propose Prompt-as-Prefix (PaP), which enriches the input context and directs the transformation of reprogrammed input patches. The transformed time series patches from the LLM are finally projected to obtain the forecasts. Our comprehensive evaluations demonstrate that Time-LLM is a powerful time series learner that outperforms state-of-the-art, specialized forecasting models. Moreover, Time-LLM excels in both few-shot and zero-shot learning scenarios., Comment: Accepted by the 12th International Conference on Learning Representations (ICLR 2024)

There is an increasing interest in developing LLMs for medical diagnosis to improve diagnosis efficiency. Despite their alluring technological potential, there is no unified and comprehensive evaluation criterion, leading to the inability to evaluate the quality and potential risks of medical LLMs, further hindering the application of LLMs in medical treatment scenarios. Besides, current evaluations heavily rely on labor-intensive interactions with LLMs to obtain diagnostic dialogues and human evaluation on the quality of diagnosis dialogue. To tackle the lack of unified and comprehensive evaluation criterion, we first initially establish an evaluation criterion, termed LLM-specific Mini-CEX to assess the diagnostic capabilities of LLMs effectively, based on original Mini-CEX. To address the labor-intensive interaction problem, we develop a patient simulator to engage in automatic conversations with LLMs, and utilize ChatGPT for evaluating diagnosis dialogues automatically. Experimental results show that the LLM-specific Mini-CEX is adequate and necessary to evaluate medical diagnosis dialogue. Besides, ChatGPT can replace manual evaluation on the metrics of humanistic qualities and provides reproducible and automated comparisons between different LLMs.

Continuous-time event sequences play a vital role in real-world domains such as healthcare, finance, online shopping, social networks, and so on. To model such data, temporal point processes (TPPs) have emerged as the most natural and competitive models, making a significant impact in both academic and application communities. Despite the emergence of many powerful models in recent years, there hasn't been a central benchmark for these models and future research endeavors. This lack of standardization impedes researchers and practitioners from comparing methods and reproducing results, potentially slowing down progress in this field. In this paper, we present EasyTPP, the first central repository of research assets (e.g., data, models, evaluation programs, documentations) in the area of event sequence modeling. Our EasyTPP makes several unique contributions to this area: a unified interface of using existing datasets and adding new datasets; a wide range of evaluation programs that are easy to use and extend as well as facilitate reproducible research; implementations of popular neural TPPs, together with a rich library of modules by composing which one could quickly build complex models. All the data and implementation can be found at https://github.com/ant-research/EasyTemporalPointProcess. We will actively maintain this benchmark and welcome contributions from other researchers and practitioners. Our benchmark will help promote reproducible research in this field, thus accelerating research progress as well as making more significant real-world impacts., Comment: ICLR 2024 camera ready

Remote sensing data play a critical role in improving numerical weather prediction (NWP). However, the physical principles of radiation dictate that data voids frequently exist in physical space (e.g., sub-cloud area for satellite infrared radiance or no-precipitation region for radar reflectivity). Such data gaps impair the accuracy of initial conditions derived from data assimilation (DA), which has a negative impact on NWP. We use the barotropic vorticity equation to demonstrate the potential of deep learning augmented data assimilation (DDA), which involves reconstructing spatially complete pseudo-observation fields from incomplete observations and using them for DA. By training a convolutional autoencoder (CAE) with a long simulation at a coarse “forecast” resolution (T63), we obtained a deep-learning approximation of the “reconstruction operator,” which maps spatially incomplete observations to a model state with full spatial coverage and resolution. The CAE was applied to an incomplete stream function observation (∼30% missing) from a high-resolution benchmark simulation and demonstrated satisfactory reconstruction performance, even when only very sparse (1/16 of T63 grid density) observations were used as input. When only spatially incomplete observations are used, the analysis fields obtained from Ensemble Square Root Filter (EnSRF) assimilation exhibit significant error. However, in DDA, when EnSRF takes in the combined data from the incomplete observations and CAE reconstruction, analysis error reduces significantly. Such gains are more pronounced with sparse observation and small ensemble size because the DDA analysis is much less sensitive to observation density and ensemble size than the conventional DA analysis, which is based solely on incomplete observations.

It is unequivocal that human influence has warmed the planet, which is seriously affecting the planetary health including human health. Adapting climate change should not only be a slogan, but requires a united, holistic action and a paradigm shift from crisis response to an ambitious and integrated approach immediately. Recognizing the urgent needs to tackle the risk connection between climate change and One Health, the four key messages and recommendations that with the intent to guide further research and to promote international cooperation to achieve a more climate-resilient world are provided.

Point process is the dominant paradigm for modeling event sequences occurring at irregular intervals. In this paper we aim at modeling latent dynamics of event propagation in graph, where the event sequence propagates in a directed weighted graph whose nodes represent event marks (e.g., event types). Most existing works have only considered encoding sequential event history into event representation and ignored the information from the latent graph structure. Besides they also suffer from poor model explainability, i.e., failing to uncover causal influence across a wide variety of nodes. To address these problems, we propose a Graph Regularized Point Process (GRPP) that can be decomposed into: 1) a graph propagation model that characterizes the event interactions across nodes with neighbors and inductively learns node representations; 2) a temporal attentive intensity model, whose excitation and time decay factors of past events on the current event are constructed via the contextualization of the node embedding. Moreover, by applying a graph regularization method, GRPP provides model interpretability by uncovering influence strengths between nodes. Numerical experiments on various datasets show that GRPP outperforms existing models on both the propagation time and node prediction by notable margins., Comment: IJCNN 2021

Continual pretraining is a popular way of building a domain-specific pretrained language model from a general-domain language model. In spite of its high efficiency, continual pretraining suffers from catastrophic forgetting, which may harm the model's performance in downstream tasks. To alleviate the issue, in this paper, we propose a continual pretraining method for the BERT-based model, named Attention-FFN Adapter. Its main idea is to introduce a small number of attention heads and hidden units inside each self-attention layer and feed-forward network. Furthermore, we train a domain-specific language model named AF Adapter based RoBERTa for the Chinese biomedical domain. In experiments, models are applied to downstream tasks for evaluation. The results demonstrate that with only about 17% of model parameters trained, AF Adapter achieves 0.6%, 2% gain in performance on average, compared to strong baselines. Further experimental results show that our method alleviates the catastrophic forgetting problem by 11% compared to the fine-tuning method.

In this paper, we tackle the important yet under-investigated problem of making long-horizon prediction of event sequences. Existing state-of-the-art models do not perform well at this task due to their autoregressive structure. We propose HYPRO, a hybridly normalized probabilistic model that naturally fits this task: its first part is an autoregressive base model that learns to propose predictions; its second part is an energy function that learns to reweight the proposals such that more realistic predictions end up with higher probabilities. We also propose efficient training and inference algorithms for this model. Experiments on multiple real-world datasets demonstrate that our proposed HYPRO model can significantly outperform previous models at making long-horizon predictions of future events. We also conduct a range of ablation studies to investigate the effectiveness of each component of our proposed methods., Comment: NeurIPS 2022 camera-ready

The intense winds in a tropical cyclone (TC) environment can produce significant mountain waves and substantially enhanced precipitation over mountainous region, yet few previous studies have been done in understanding how the orographic precipitation in a TC environment might respond to global warming. Here, we focus on this problem and use large-eddy simulation (LES) to estimate the global warming-induced response of precipitation over and near an idealized mountain with 10km half width and 1km height. The simulations were conducted with Cloud Model 1 (CM1) with 200m horizontal grid spacing in a small 51.2 km by 102.4 km domain, for which the large-scale forcing is extracted from the outer region of a convection-permitting simulation of Typhoon Vicente (2012). Pseudo-global warming experiments are further conducted by adding mean thermodynamic structure changes predicted by a CMIP6 climate simulation. In the domain, two regions exhibit locally enhanced precipitation. One is over the mountain with rainfall maximum on the lee slope; the other is in the downstream region 30 to 50 km away from the mountain. Regression analysis suggest the enhanced precipitation in both regions are related to the seeder-feeder mechanism. Though the enhancement in the downstream regions differs from the conventional definition. Mountain waves generate an intense cloud formation center in the upper troposphere above the mountain, resulting snow and graupel drift downstream which intensify downstream convection when they fall into a proper height. Under global warming, the first precipitation maximum barely changes, while the second precipitation maximum intensify and shift upwind. Correspondingly, the mountain lee slope region exhibits a mean sensitivity of ~-0.49%/K, whereas the downstream enhanced region exhibits mean sensitivity ~18%/K which is significantly higher than the Clausius–Clapeyron (CC) scaling of ~7%/K of surface warming. Analysis suggests that the little precipitation sen

The Northern Annular Mode (NAM) is a zonally symmetric pattern of the sea level pressure (SLP) and the leading mode of tropospheric winter climate variability over the Northern Hemisphere (NH). Characterized with two main centers of action over the northern Pacific and Atlantic basins, the NAM exhibits a negative or positive phase defined by anomalous low or high SLP over those centers. As the leading mode of midlatitude atmospheric circulation variability over the NH, the NAM influences the winter surface temperature over several NH regions, including North America. However, the Pacific center of NAM is projected to shift and enhance under the anthropogenic forcing. As a result, a change in the winter surface pattern is also expected in the future (2051-2100). This study investigates the NAM-related surface changes based on the mean pattern of 40 ensemble members of the Community Earth System Model version 1 (CESM1) Large Ensemble (LENS). In the present climate (1951-2000), the anomalous high over the Pacific center is associated with anomalously warm winters over the Northern Great Plains during the positive phase of the NAM. However, under anthropogenic forcing, the North American wintertime minimum surface temperature decreases during positive NAM conditions compared to the historical run. In addition, such weakening of the NAM-surface temperature teleconnection is also shown in nine out of fifteen models listed on Coupled Model Intercomparison Project Phase 6 (CMIP6). Furthermore, the projected winter Eddy Kinetic Energy intensifies and shifts along with the Pacific storm track during future positive and negative NAM conditions. These shifts align with the future shifts in the NAM’s Pacific center. This change in the storm track also leads to increased winter SST and precipitation over the Northeastern Pacific during the negative NAM phase. Moreover, we hypothesize that the projected NAM changes are related to weakened winter stratospheric temperature anomalies

Limited studies attempted to incorporate the dynamic urban growth into the regional modeling system over intensely urbanized city agglomerations in China. The Weather Research and Forecasting (WRF version 4.2) coupled with the single-layer urban canopy model (SLUCM) simulates the long-term climate impacts of LULC change in China's two fast-growing delta regions. The physical parameterizations in the PRD and YRD are first investigated in a high-resolution (4km) downscaled modeling. Then, the sensitivity experiments and new classified urban data are tested with fine-tuned building parameters and anthropogenic heat input in the SLUCM. The second part of the paper focuses on the impact of transient urbanization on the current and future climate. The most significant response to urbanization impacts comes from the SSP5 in the 2030s. The climatic interactions of different LULC in the nested domains are distinct that the PRD has warming and lower relative humidity on a local urban scale. In contrast, the YRD has broader feedback on a regional scale. Also, the urban effect on diurnal temperature varied significantly based on the forms and location of urban areas. The urbanization-induced warming mainly happened at night in the newly developed urban areas in the PRD, while such warming occurs in the late afternoon in the YRD. The local warming trend induced by urban expansion has already caused heat stress in these highly populated regions. Driven by the NorESM RCP8.5 scenario and the corresponding SSP5 urban growth data, the synergic effects of urban development and global warming are further quantified. Global warming is contributing more climatic impacts than LULC change, while the urban landcover can have an amplified effect on the local climate. The PRD and the YRD will experience dangerous heat stress under the synergic effects, with the risks of extreme in YRD developed more.

Convection-permitting resolutions, which refer to kilometer-scale horizontal grid spacings, have become increasingly popular in regional numerical weather prediction and climate studies. However, this resolution range is in the gray zone for the simulation of convection, where conventional cumulus convection and subgrid-scale (SGS) turbulence parameterizations are inadequate for such grid spacings due to invalid assumptions and simplifications. Recent studies demonstrated that the magnitudes of SGS fluxes of momentum and scalars are comparable to those of resolved fluxes at convection-permitting resolutions and that horizontal SGS components are as important as the vertical SGS component. Thus, it appears necessary to adapt available schemes to model the SGS effects of convective motions for the gray zone. Here, we investigated the efficacy of separately parameterizing the vertical and horizontal SGS effects in improving the convection-permitting simulation of Typhoon Vicente (2012). To represent the vertical SGS turbulence effect, we evaluated the Grell-3, Tiedtke, and multiscale Kain-Fritsch (MSKF) schemes in the Weather Research and Forecasting (WRF) Model; the MSKF scheme is scale adaptive, whereas the other two are conventional cumulus schemes. For horizontal SGS turbulence, we evaluated the effects of the traditional Smagorinsky scheme and our newly developed reconstruction and nonlinear anisotropy (RNA) model, which models not only downgradient diffusion but also backscatter. We found that the simulation combining the MSKF and RNA schemes exhibits the best skill in predicting precipitation, especially rainfall extremes. The advantages are rooted in the MSKF scheme's scale-awareness and parameterized cloud-radiation feedback and in the backscatter-enabling capability of the RNA model. Significance StatementOperational numerical weather prediction and some climate simulations have approached kilometer-scale horizontal resolutions, called convection-permitting r

We consider a sequential decision making problem where the agent faces the environment characterized by the stochastic discrete events and seeks an optimal intervention policy such that its long-term reward is maximized. This problem exists ubiquitously in social media, finance and health informatics but is rarely investigated by the conventional research in reinforcement learning. To this end, we present a novel framework of the model-based reinforcement learning where the agent's actions and observations are asynchronous stochastic discrete events occurring in continuous-time. We model the dynamics of the environment by Hawkes process with external intervention control term and develop an algorithm to embed such process in the Bellman equation which guides the direction of the value gradient. We demonstrate the superiority of our method in both synthetic simulator and real-world problem.

Predictive autoscaling (autoscaling with workload forecasting) is an important mechanism that supports autonomous adjustment of computing resources in accordance with fluctuating workload demands in the Cloud. In recent works, Reinforcement Learning (RL) has been introduced as a promising approach to learn the resource management policies to guide the scaling actions under the dynamic and uncertain cloud environment. However, RL methods face the following challenges in steering predictive autoscaling, such as lack of accuracy in decision-making, inefficient sampling and significant variability in workload patterns that may cause policies to fail at test time. To this end, we propose an end-to-end predictive meta model-based RL algorithm, aiming to optimally allocate resource to maintain a stable CPU utilization level, which incorporates a specially-designed deep periodic workload prediction model as the input and embeds the Neural Process to guide the learning of the optimal scaling actions over numerous application services in the Cloud. Our algorithm not only ensures the predictability and accuracy of the scaling strategy, but also enables the scaling decisions to adapt to the changing workloads with high sample efficiency. Our method has achieved significant performance improvement compared to the existing algorithms and has been deployed online at Alipay, supporting the autoscaling of applications for the world-leading payment platform., Comment: Accepted by KDD'22 Applied Research Track

Remote sensing data play a critical role in improving numerical weather prediction (NWP). However, the physical principles of radiation dictate that data voids frequently exist in physical space (e.g., sub-cloud area for satellite infrared radiance or no-precipitation region for radar reflectivity). Such data gaps impair the accuracy of initial conditions derived from data assimilation (DA), which has a negative impact on NWP. We use the barotropic vorticity equation to demonstrate the potential of deep learning augmented data assimilation (DDA), which involves reconstructing spatially complete pseudo-observation fields from incomplete observations and using them for DA. By training a convolutional autoencoder (CAE) with a long simulation at a coarse “forecast” resolution (T63), we obtained a deep-learning approximation of the “reconstruction operator,” which maps spatially incomplete observations to a model state with full spatial coverage and resolution. The CAE was applied to an incomplete stream function observation (∼30% missing) from a high-resolution benchmark simulation and demonstrated satisfactory reconstruction performance, even when only very sparse (1/16 of T63 grid density) observations were used as input. When only spatially incomplete observations are used, the analysis fields obtained from Ensemble Square Root Filter (EnSRF) assimilation exhibit significant error. However, in DDA, when EnSRF takes in the combined data from the incomplete observations and CAE reconstruction, analysis error reduces significantly. Such gains are more pronounced with sparse observation and small ensemble size because the DDA analysis is much less sensitive to observation density and ensemble size than the conventional DA analysis, which is based solely on incomplete observations.

It is unequivocal that human influence has warmed the planet, which is seriously affecting the planetary health including human health. Adapting climate change should not only be a slogan, but requires a united, holistic action and a paradigm shift from crisis response to an ambitious and integrated approach immediately. Recognizing the urgent needs to tackle the risk connection between climate change and One Health, the four key messages and recommendations that with the intent to guide further research and to promote international cooperation to achieve a more climate-resilient world are provided.

It is unequivocal that human influence has warmed the planet, which is seriously affecting the planetary health including human health. Adapting climate change should not only be a slogan, but requires a united, holistic action and a paradigm shift from crisis response to an ambitious and integrated approach immediately. Recognizing the urgent needs to tackle the risk connection between climate change and One Health, the four key messages and recommendations that with the intent to guide further research and to promote international cooperation to achieve a more climate-resilient world are provided.

In vitro and in vivo experimental studies have shown garlic has protective effects on the aging process; however, there is no evidence that garlic consumption is associated with all-cause mortality among oldest-old individuals (≥80 years). From 1998 to 2011, 27,437 oldest-old participants (mean age: 92.9 years) were recruited from 23 provinces in China. The frequencies of garlic consumption at baseline and at age 60 were collected. Cox proportional hazards models adjusted for potential covariates were constructed to estimate hazard ratios (HRs) relating garlic consumption to all-cause mortality. Among 92,505 person-years of follow-up from baseline to September 1, 2014, 22,321 participants died. Participants who often (≥5 times/week) or occasionally (1-4 times/week) consumed garlic survived longer than those who rarely (less than once/week) consumed it (p < 0.001). Participants who consumed garlic occasionally or often had a lower risk for mortality than those who rarely consumed garlic at baseline; the adjusted HRs for mortality were 0.92(0.89-0.94) and 0.89(0.85-0.92), respectively. The inverse associations between garlic consumption and all-cause mortality were robust in sensitivity analyses and subgroup analyses. In this study, habitual consumption of garlic was associated with a lower all-cause mortality risk; this advocates further investigation into garlic consumption for promoting longevity.

In vitro and in vivo experimental studies have shown garlic has protective effects on the aging process; however, there is no evidence that garlic consumption is associated with all-cause mortality among oldest-old individuals (≥80 years). From 1998 to 2011, 27,437 oldest-old participants (mean age: 92.9 years) were recruited from 23 provinces in China. The frequencies of garlic consumption at baseline and at age 60 were collected. Cox proportional hazards models adjusted for potential covariates were constructed to estimate hazard ratios (HRs) relating garlic consumption to all-cause mortality. Among 92,505 person-years of follow-up from baseline to September 1, 2014, 22,321 participants died. Participants who often (≥5 times/week) or occasionally (1-4 times/week) consumed garlic survived longer than those who rarely (less than once/week) consumed it (p < 0.001). Participants who consumed garlic occasionally or often had a lower risk for mortality than those who rarely consumed garlic at baseline; the adjusted HRs for mortality were 0.92(0.89-0.94) and 0.89(0.85-0.92), respectively. The inverse associations between garlic consumption and all-cause mortality were robust in sensitivity analyses and subgroup analyses. In this study, habitual consumption of garlic was associated with a lower all-cause mortality risk; this advocates further investigation into garlic consumption for promoting longevity.

Ocean variability plays an essential role in the climate system at different time scales through air–sea interactions. Recent studies have addressed the importance of the ocean mixed layer in cooling feedback to tropical cyclones (TCs). However, using constant sea surface temperature (SST) in short-range weather forecasts remains common, especially in high-resolution regional models. This study investigates the role of subsurface ocean mixing in the short-range forecast of non-TC extreme rainfall with the Weather Research and Forecast (WRF) model. In the simulations of 26 heavy rainfall cases, we found that using a one-dimensional mixed layer model leads to a 15% enhancement (reduction) of rainfall maximum in six (two) cases compared to using constant SST. When the initial depth of the mixed layer model is perturbed by the amount of daily variability, 13 cases exhibit larger than 15% increases or decreases. A detailed analysis of one case suggests that the radiative process dominates the overall response of SST. The warming and moistening of boundary layer air cause significant strengthening of updrafts in convection. Although the SST change in most cases due to varying mixed layer model setups is less than 0.5 K, convective motions in some cases are surprisingly sensitive to small changes. © 2021 by the authors. Licensee MDPI, Basel, Switzerland.

This study investigates the mechanisms by which small-scale turbulence and cloud physics determine the organization of large-scale convection in radiative-convective equilibrium (RCE), an idealization of the tropical atmosphere. Under uniform forcings similar to typical tropical conditions, the atmosphere in RCE might spontaneously separate into dry and moist regions on scales of 100–1,000 km, with convective clouds aggregating into a cluster in the latter. This phenomenon is known as convective self-aggregation. Herein, we demonstrate that subtle changes in assumptions related to cloud physics and turbulence on scales of E 1 km can dictate the emergence or suppression of convective self-aggregation, resulting from a bifurcation of the dynamical system. The bifurcation occurs when a small dry patch forms in the domain and is sustained because it contributes to negative effective diffusivity of the circulation. Cloud-radiation feedbacks and turbulence circulation interactions govern the formation of such dry patches, thereby modulating the bifurcation. This sensitive dependence on subgrid process models might be a fundamental barrier to climate predictability in light of inherent uncertainties in microscale processes. Because without the capability to include exact representations of those processes in climate models, slight differences in the different approximations used by modelers can lead to qualitative changes in climate predictions, at least for some processes.

Convective self-aggregation in radiative-convective equilibrium (RCE) is a phenomenon of significant interest because of its potential relevance to the organization of tropical storms. However, some previous theories on RCE instability neglected or simplified cloud-radiative interaction in theoretical models and suggested that water vapor-radiation feedback is responsible for destabilizing a homogeneous RCE state. Here, through numerical experiments of nonrotating RCE, we first confirm that the predictions from these theories are valid insomuch that dry patches can form in an initially homogeneous domain with uniform sea-surface temperature when cloud-radiation interaction is suppressed or eliminated. However, sustained growth of those dry patches is not guaranteed. Developing deep convective circulation can engulf dry patches, and water vapor feedback is not sufficient for overcoming this barrier. Only when the early-stage cloud radiative forcing is sufficiently strong can self-aggregation fully develop. Therefore, cloud-radiative feedback is essential for the eventual realization of self-aggregation in nonrotating RCE.

The projection of extreme convective precipitation by global climate models (GCM) exhibits significant uncertainty due to coarse resolutions. Direct dynamical downscaling (DDD) of regional climate at kilometer-scale resolutions provides valuable insight into extreme precipitation changes, but its computational expense is formidable. Here we document the effectiveness of machine learning in enabling smart dynamical downscaling (SDD), which selects a small subset of GCM data to conduct downscaling. Three machine learning models are evaluated, including support vector machine, an 8-layer neural network, and a sophisticated 58-layer neural network with residual connections. Trained with data for three subtropical/tropical regions, the convolutional neural networks (CNNs) can retain 92% to 98% of extreme precipitation events (rain intensity higher than the 99th percentile) while filtering out 88% to 95% of circulation data. When applied to two different reanalysis data sets, the CNNs' skill in retaining extremes decreases modestly in subtropical regions but sharply in the deep tropics. Nonetheless, one of the CNNs can still retain 62% of all extreme events in the deep tropical region in the worst case. These deep learning models make SDD become possible, and therefore the study of extreme events under global warming more affordable for the community.