Forecasting small-scale dynamics of fluid turbulence using deep neural networks.

Turbulence in fluid flows is characterized by a wide range of interacting scales. Since the scale range increases as some power of the flow Reynolds number, a faithful simulation of the entire scale range is prohibitively expensive at high Reynolds numbers. The most expensive aspect concerns the small-scale motions; thus, major emphasis is placed on understanding and modeling them, taking advantage of their putative universality. In this work, using physics-informed deep learning methods, we present a modeling framework to capture and predict the small-scale dynamics of turbulence, via the velocity gradient tensor. The model is based on obtaining functional closures for the pressure Hessian and viscous Laplacian contributions as functions of velocity gradient tensor. This task is accomplished using deep neural networks that are consistent with physical constraints and explicitly incorporate Reynolds number dependence to account for small-scale intermittency. We then utilize a massive direct numerical simulation database, spanning two orders of magnitude in the large-scale Reynolds number, for training and validation. The model learns from low to moderate Reynolds numbers and successfully predicts velocity gradient statistics at both seen and higher (unseen) Reynolds numbers. The success of our present approach demonstrates the viability of deep learning over traditional modeling approaches in capturing and predicting small-scale features of turbulence.

Climate presses and pulses mediate the decline of a migratory predator.

Long-term climate changes and extreme climate events differentially impact animal populations, yet whether and why these processes may act synergistically or antagonistically remains unknown. Disentangling these potentially interactive effects is critical for predicting population outcomes as the climate changes. Here, we leverage the "press-pulse" framework, which is used to describe ecological disturbances, to disentangle population responses in migratory Magellanic penguins to long-term changes in climate means and variability (presses) and extreme events (pulses) across multiple climate variables and life history stages. Using an unprecedented 38-y dataset monitoring 53,959 penguins, we show for the first time that the presses and pulses of climate change mediate the rate of population decline by differentially impacting different life stages. Moreover, we find that climate presses and pulses can work both synergistically and antagonistically to affect animal population persistence, necessitating the need to examine both processes in concert. Negative effects of terrestrial heat waves (pulses) on adult survival, for example, were countered by positive effects of long-term changes in oceanographic conditions in migratory grounds (presses) on juvenile and adult survival. Taken together, these effects led to predicted population extirpation under all future climate scenarios. This work underscores the importance of a holistic approach integrating multiple climate variables, life stages, and presses and pulses for predicting the persistence of animals under accelerating climate change.

Twenty-first century hydroclimate: A continually changing baseline, with more frequent extremes.

SignificanceTwenty-first century trends in hydroclimate are so large that future average conditions will, in most cases, fall into the range of what we would today consider extreme drought or pluvial states. Using large climate model ensembles, we remove the background trend and find that the risk of droughts and pluvials relative to that (changing) baseline is fairly similar to the 20th century risk. By continually adapting to long-term background changes, these risks could therefore perhaps be minimized. However, increases in the frequency of extremely wet and dry years are still present even after removing the trend, indicating that sustainably managing hydroclimate-driven risks in a warmer world will face increasingly difficult challenges.

A spatiotemporal decay model of human mobility when facing large-scale crises.

A common feature of large-scale extreme events, such as pandemics, wildfires, and major storms is that, despite their differences in etiology and duration, they significantly change routine human movement patterns. Such changes, which can be major or minor in size and duration and which differ across contexts, affect both the consequences of the events and the ability of governments to mount effective responses. Based on naturally tracked, anonymized mobility behavior from over 90 million people in the United States, we document these mobility differences in space and over time in six large-scale crises, including wildfires, major tropical storms, winter freeze and pandemics. We introduce a model that effectively captures the high-dimensional heterogeneity in human mobility changes following large-scale extreme events. Across five different metrics and regardless of spatial resolution, the changes in human mobility behavior exhibit a consistent hyperbolic decline, a pattern we characterize as "spatiotemporal decay." When applied to the case of COVID-19, our model also uncovers significant disparities in mobility changes-individuals from wealthy areas not only reduce their mobility at higher rates at the start of the pandemic but also maintain the change longer. Residents from lower-income regions show a faster and greater hyperbolic decay, which we suggest may help account for different COVID-19 rates. Our model represents a powerful tool to understand and forecast mobility patterns post emergency, and thus to help produce more effective responses.

Thirty-six years of butterfly monitoring, snow cover, and plant productivity reveal negative impacts of warmer winters and increased productivity on montane species.

Climate change is contributing to declines of insects through rising temperatures, altered precipitation patterns, and an increasing frequency of extreme events. The impacts of both gradual and sudden shifts in weather patterns are realized directly on insect physiology and indirectly through impacts on other trophic levels. Here, we investigated direct effects of seasonal weather on butterfly occurrences and indirect effects mediated by plant productivity using a temporally intensive butterfly monitoring dataset, in combination with high-resolution climate data and a remotely sensed indicator of plant primary productivity. Specifically, we used Bayesian hierarchical path analysis to quantify relationships between weather and weather-driven plant productivity on the occurrence of 94 butterfly species from three localities distributed across an elevational gradient. We found that snow pack exerted a strong direct positive effect on butterfly occurrence and that low snow pack was the primary driver of reductions during drought. Additionally, we found that plant primary productivity had a consistently negative effect on butterfly occurrence. These results highlight mechanisms of weather-driven declines in insect populations and the nuances of climate change effects involving snow melt, which have implications for ecological theories linking topographic complexity to ecological resilience in montane systems.

The impact of aragonite saturation variability on shelled pteropods: An attribution study in the California Current System.

Observations from the California Current System (CalCS) indicate that the long-term trend in ocean acidification (OA) and the naturally occurring corrosive conditions for the CaCO3 mineral aragonite (saturation state Ω < 1) have a damaging effect on shelled pteropods, a keystone group of calcifying organisms in the CalCS. Concern is heightened by recent findings suggesting that shell formation and developmental progress are already impacted when Ω falls below 1.5. Here, we quantify the impact of low Ω conditions on pteropods using an individual-based model (IBM) with life-stage-specific mortality, growth, and behavior in a high-resolution regional hindcast simulation of the CalCS between 1984 and 2019. Special attention is paid to attributing this impact to different processes that lead to such low Ω conditions, namely natural variability, long-term trend, and extreme events. We find that much of the observed damage in the CalCS, and specifically >70% of the shell CaCO3 loss, is due to the pteropods' exposure to naturally occurring low Ω conditions as a result of their diel vertical migration (DVM). Over the hindcast period, their exposure to damaging waters (Ω < 1.5) increases from 9% to 49%, doubling their shell CaCO3 loss, and increasing their mortality by ~40%. Most of this increased exposure is due to the shoaling of low Ω waters driven by the long-term trend in OA. Extreme OA events amplify this increase by ~40%. Our approach can quantify the health of pteropod populations under shifting environmental conditions, and attribute changes in fitness or population structure to changes in the stressor landscape across hierarchical time scales.

Impacts of marine heatwaves in coastal ecosystems depend on local environmental conditions.

Marine heatwaves (MHWs), increasing in duration and intensity because of climate change, are now a major threat to marine life and can have lasting effects on the structure and function of ecosystems. However, the responses of marine taxa and ecosystems to MHWs can be highly variable, making predicting and interpreting biological outcomes a challenge. Here, we review how biological responses to MHWs, from individuals to ecosystems, are mediated by fine-scale spatial variability in the coastal marine environment (hereafter, local gradients). Viewing observed responses through a lens of ecological theory, we present a simple framework of three 'resilience processes' (RPs) by which local gradients can influence the responses of marine taxa to MHWs. Local gradients (1) influence the amount of stress directly experienced by individuals, (2) facilitate local adaptation and acclimatization of individuals and populations, and (3) shape community composition which then influences responses to MHWs. We then synthesize known examples of fine-scale gradients that have affected responses of benthic foundation species to MHWs, including kelp forests, coral reefs, and seagrass meadows and link these varying responses to the RPs. We present a series of case studies from various marine ecosystems to illustrate the differential impacts of MHWs mediated by gradients in both temperature and other co-occurring drivers. In many cases, these gradients had large effect sizes with several examples of local gradients causing a 10-fold difference in impacts or more (e.g., survival, coverage). This review highlights the need for high-resolution environmental data to accurately predict and manage the consequences of MHWs in the context of ongoing climate change. While current tools may capture some of these gradients already, we advocate for enhanced monitoring and finer scale integration of local environmental heterogeneity into climate models. This will be essential for developing effective conservation strategies and mitigating future marine biodiversity loss.

Predicting organismal response to marine heatwaves using dynamic thermal tolerance landscape models.

Marine heatwaves (MHWs) can cause thermal stress in marine organisms, experienced as extreme 'pulses' against the gradual trend of anthropogenic warming. When thermal stress exceeds organismal capacity to maintain homeostasis, organism survival becomes time-limited and can result in mass mortality events. Current methods of detecting and categorizing MHWs rely on statistical analysis of historic climatology and do not consider biological effects as a basis of MHW severity. The re-emergence of ectotherm thermal tolerance landscape models provides a physiological framework for assessing the lethal effects of MHWs by accounting for both the magnitude and duration of extreme heat events. Here, we used a simulation approach to understand the effects of a suite of MHW profiles on organism survival probability across (1) three thermal tolerance adaptive strategies, (2) interannual temperature variation and (3) seasonal timing of MHWs. We identified survival isoclines across MHW magnitude and duration where acute (short duration-high magnitude) and chronic (long duration-low magnitude) events had equivalent lethal effects on marine organisms. While most research attention has focused on chronic MHW events, we show similar lethal effects can be experienced by more common but neglected acute marine heat spikes. Critically, a statistical definition of MHWs does not accurately categorize biological mortality. By letting organism responses define the extremeness of a MHW event, we can build a mechanistic understanding of MHW effects from a physiological basis. Organism responses can then be transferred across scales of ecological organization and better predict marine ecosystem shifts to MHWs.

Avian responses to climate extremes: insights into abundance curves and species sensitivity using the UK Breeding Bird Survey.

Climate change remains one of the most urgent challenges for biodiversity conservation. Recent studies have highlighted that climate extremes (CLEXs) can lead to widespread and negative effects across all taxa and ecological levels, but most of these studies are based on short-term periods and small spatial scales and lack a multi-species approach. Here, using generalised additive models (GAMs) and the UK Breeding Bird Survey (BBS), we described response curves for the abundance of 100 resident bird species over large spatial and temporal scales and identified the species showing a greater sensitivity to CLEXs. We used five climatic indices computed at 1-km spatial resolution as proxies of CLEXs during the winter or breeding season and considered both 1- and 2-year lagged effects. The results demonstrated widespread and significant effects of CLEXs on bird abundances at both time lags and in both seasons. Winter frost days (FD0), summer days (SU25) during the breeding season and simple precipitation intensity index (SDII) during the breeding season mainly showed negative effects. Daily temperature range (DTR) in both winter and breeding season and dry days (DD) during the breeding season led to diversified responses across the species, with a prevalence of positive effects. A large proportion of species showed a high sensitivity to CLEXs, highlighting that these species may deserve attention in future studies aimed at biodiversity conservation. We demonstrated that CLEXs can represent a significant driver affecting population abundances over large spatial and temporal scales, emphasising the need for understanding mechanistic processes at the basis of the observed effects.

Rocky Mountain subalpine forests now burning more than any time in recent millennia.

The 2020 fire season punctuated a decades-long trend of increased fire activity across the western United States, nearly doubling the total area burned in the central Rocky Mountains since 1984. Understanding the causes and implications of such extreme fire seasons, particularly in subalpine forests that have historically burned infrequently, requires a long-term perspective not afforded by observational records. We place 21st century fire activity in subalpine forests in the context of climate and fire history spanning the past 2,000 y using a unique network of 20 paleofire records. Largely because of extensive burning in 2020, the 21st century fire rotation period is now 117 y, reflecting nearly double the average rate of burning over the past 2,000 y. More strikingly, contemporary rates of burning are now 22% higher than the maximum rate reconstructed over the past two millennia, during the early Medieval Climate Anomaly (MCA) (770 to 870 Common Era), when Northern Hemisphere temperatures were ∼0.3 °C above the 20th century average. The 2020 fire season thus exemplifies how extreme events are demarcating newly emerging fire regimes as climate warms. With 21st century temperatures now surpassing those during the MCA, fire activity in Rocky Mountain subalpine forests is exceeding the range of variability that shaped these ecosystems for millennia.

Humanitarian need drives multilateral disaster aid.

As the climate changes, human livelihoods will increasingly be threatened by extreme weather events. To provide adequate disaster relief, states extensively rely on multilateral institutions, in particular the United Nations (UN). However, the determinants of this multilateral disaster aid channeled through the UN are poorly understood. To fill this gap, we examine the determinants of UN disaster aid using a dataset on UN aid covering almost 2,000 climate-related disasters occurring between 2006 and 2017. We make two principal contributions. First, we add to research on disaster impacts by linking existing disaster data from the Emergency Events Database (EM-DAT) to a meteorological reanalysis. We generate a uniquely global hazard severity measure that is comparable across different climate-related disaster types, and assess and bolster measurement validity of EM-DAT climate-related disasters. Second, by combining these data with social data on aid and its correlates, we contribute to the literature on aid disbursements. We show that UN disaster aid is primarily shaped by humanitarian considerations, rather than by strategic donor interests. These results are supported by a series of regression and out-of-sample prediction analyses and appear consistent with the view that multilateral institutions are able to shield aid allocation decisions from particular state interests to ensure that aid is motivated by need.

The vulnerability of sharks, skates, and rays to ocean deoxygenation: Physiological mechanisms, behavioral responses, and ecological impacts.

Levels of dissolved oxygen in open ocean and coastal waters are decreasing (ocean deoxygenation), with poorly understood effects on marine megafauna. All of the more than 1000 species of elasmobranchs (sharks, skates, and rays) are obligate water breathers, with a variety of life-history strategies and oxygen requirements. This review demonstrates that although many elasmobranchs typically avoid hypoxic water, they also appear capable of withstanding mild to moderate hypoxia with changes in activity, ventilatory responses, alterations to circulatory and hematological parameters, and morphological alterations to gill structures. However, such strategies may be insufficient to withstand severe, progressive, or prolonged hypoxia or anoxia where anaerobic metabolic pathways may be used for limited periods. As water temperatures increase with climate warming, ectothermic elasmobranchs will exhibit elevated metabolic rates and are likely to be less able to tolerate the effects of even mild hypoxia associated with deoxygenation. As a result, sustained hypoxic conditions in warmer coastal or surface-pelagic waters are likely to lead to shifts in elasmobranch distributions. Mass mortalities of elasmobranchs linked directly to deoxygenation have only rarely been observed but are likely underreported. One key concern is how reductions in habitat volume as a result of expanding hypoxia resulting from deoxygenation will influence interactions between elasmobranchs and industrial fisheries. Catch per unit of effort of threatened pelagic sharks by longline fisheries, for instance, has been shown to be higher above oxygen minimum zones compared to adjacent, normoxic regions, and attributed to vertical habitat compression of sharks overlapping with increased fishing effort. How a compound stressor such as marine heatwaves alters vulnerability to deoxygenation remains an open question. With over a third of elasmobranch species listed as endangered, a priority for conservation and management now lies in understanding and mitigating ocean deoxygenation effects in addition to population declines already occurring from overfishing.

Wildfires impact on PM2.5 concentration in galicia Spain.

Wildfire intensity and severity have been increasing in the Iberian Peninsula in recent years, particularly in the Galicia region, due to rising temperatures and accumulating drier combustible vegetation in unmanaged lands. This leads to substantial emissions of air pollutants, notably fine particles (PM2.5), posing a risk to public health. This study aims to assess the impact of local and regional wildfires on PM2.5 levels in Galicia's main cities and their implications for air quality and public health. Over a decade (2013-2022), PM2.5 data during wildfire seasons were analyzed using statistical methods and Lagrangian tracking to monitor smoke plume evolution. The results reveal a notable increase in PM2.5 concentration during the wildfire season (June-November) in Galicia, surpassing health guidelines during extreme events and posing a significant health risk to the population. Regional wildfire analyses indicate that smoke plumes from Northern Portugal contribute to pollution in Galician cities, influencing the seasonality of heightened PM2.5 levels. During extensive wildfires, elevated PM2.5 concentration values persisted for several days, potentially exacerbating health concerns in Galicia. These findings underscore the urgency of implementing air pollution prevention and management measures in the region, including developing effective alerts for large-scale events and improved wildfire management strategies to mitigate their impact on air quality in Galician cities.

Longitudinal assessment of extreme climate events in Kinnaur district, Himachal Pradesh, north-western Himalaya, India.

It is vital to keep an eye on changes in climatic extremes because they set the stage for current and potential future climate, which usually have a reasonable adverse impact on ecosystems and society. The present study examines the variability and trends in precipitation and temperature across seasons in the Kinnaur district, offering valuable insights into the complex dynamics of the Himalayan climate. Using Climatic Research Unit gridded Time Series (CRU TS) datasets from 1951 to 2021, the study analyzes the data to produce 28 climate indices based on India Meteorological Department (IMD) convention indices and Expert Team on Climate Change Detection and Indices (ETCCDI). Although there may be considerable variation in climate indices in terms of absolute values within different products, there is consensus in both long-term trends and inter-annual variability. Analysis shows that even within a small area, there is variability in the magnitude and direction of historic temperature trends. Initially, the data were subjected to rigorous quality control procedures, which involved identifying anomalies. Statistical analysis like trend analysis, employing Mann-Kendall test and Sen's slope estimator, reveal significant (p < 0.05) increase in consecutive dry days (CDD) at 0.03 days/year and decrease in consecutive wet days (CWD) at 0.02 days/year. Notably, the frequency of heavy precipitation occurrences showed an increasing trend. Changes in precipitation in the Western Himalaya are driven by a complex interplay of orographic effects, monsoonal dynamics, atmospheric circulation patterns, climate change, and localized factors such as topography, atmospheric circulation patterns, moisture sources, land-sea temperature contrasts, and anthropogenic influences. Moreover, in case of temperature indices, there is significant increasing trend observed. Temperature indices indicate a significant annual increase in warm nights (TN90p) at 0.06%/year and warm days (TX90p) at 0.11%/year. Extreme temperature events have been trending upward, with monthly daily maximum temperature (TXx) increasing by 1.5 °C yearly. This study enhances our comprehension of the global warming phenomenon and underscores the importance of acknowledging alterations in the water cycle and their repercussions on hydrologic resources, agriculture, and livelihoods in the cold desert of the northwestern Indian Himalaya.

Investigating the utility of satellite-based precipitation products for simulating extreme discharge events: an exhaustive model-driven approach for a tropical river basin in India.

Satellite-based precipitation estimates are a critical source of information for understanding and predicting hydrological processes at regional or global scales. Given the potential variability in the accuracy and reliability of these estimates, comprehensive performance assessments are essential before their application in specific hydrological contexts. In this study, six satellite-based precipitation products (SPPs), namely, CHIRPS, CMORPH, GSMaP, IMERG, MSWEP, and PERSIANN, were evaluated for their utility in hydrological modeling, specifically in simulating streamflow using the Variable Infiltration Capacity (VIC) model. The performance of the VIC model under varying flow conditions and timescales was assessed using statistical indicators, viz., R2, KGE, PBias, RMSE, and RSR. The findings of the study demonstrate the effectiveness of VIC model in simulating hydrological components and its applicability in evaluating the accuracy and reliability of SPPs. The SPPs were shown to be valuable for streamflow simulation at monthly and daily timescales, as confirmed by various performance measures. Moreover, the performance of SPPs for simulating extreme flow events (streamflow above 75%, 90%, and 95%) using the VIC model was assessed and a significant decrease in the performance was observed for high-flow events. Comparative analysis revealed the superiority of IMERG and CMORPH for streamflow simulation at daily timescale and high-flow conditions. In contrast, the performances of CHIRPS and PERSIANN were found to be poor. This study highlights the importance of thoroughly assessing the SPPs in modeling diverse flow conditions.

Metabolic constraints on the body size scaling of extreme population densities.

Pest outbreaks, harmful algal blooms and population collapses are extreme events with critical consequences for ecosystems. Therefore, understanding the ecological mechanisms underlying these extreme events is crucial. We evaluated theoretical predictions on the size scaling and variance of extreme population abundance by combining (i) the generalized extreme value (GEV) theory and (ii) the resource-limited metabolic restriction hypothesis for population abundance. Using the phytoplankton data from the L4 station in the English Channel, we showed a negative size scaling of the expected value of maximal density, whose confidence interval included the predicted metabolic scaling (α = -1) supporting theoretical predictions. The role of resources and temperature in the distribution of the size-abundance pattern and residuals was well characterized by the GEV distribution. This comprehensive modelling framework will allow to elucidate community structure and fluctuations and provide unbiased return times estimates, thereby improving the prediction accuracy of the timing of the population outbreaks.

Marine heatwaves and upwelling shape stress responses in a keystone predator.

Climate change increases the frequency and intensifies the magnitude and duration of extreme events in the sea, particularly so in coastal habitats. However, the interplay of multiple extremes and the consequences for species and ecosystems remain unknown. We experimentally tested the impacts of summer heatwaves of differing intensities and durations, and a subsequent upwelling event on a temperate keystone predator, the starfish Asterias rubens. We recorded mussel consumption throughout the experiment and assessed activity and growth at strategically chosen time points. The upwelling event overall impaired starfish feeding and activity, likely driven by the acidification and low oxygen concentrations in the upwelled seawater. Prior exposure to a present-day heatwave (+5°C above climatology) alleviated upwelling-induced stress, indicating cross-stress tolerance. Heatwaves of present-day intensity decreased starfish feeding and growth. While the imposed heatwaves of limited duration (9 days) caused slight impacts but allowed for recovery, the prolonged (13 days) heatwave impaired overall growth. Projected future heatwaves (+8°C above climatology) caused 100% mortality of starfish. Our findings indicate a positive ecological memory imposed by successive stress events. Yet, starfish populations may still suffer extensive mortality during intensified end-of-century heatwave conditions.

Integrated analysis shows how the effects of extreme flooding events propagate through fish communities to impact amphibians.

Research Highlight: Davis, C. L., Walls, S. C., Barichivich, W. J., Brown, M. E., & Miller, D. A. (2022). Disentangling direct and indirect effects of extreme events on coastal wetland communities. Journal of Animal Ecology, https://doi.org/10.1111/1365-2656.13874. Catastrophic events such as floods, hurricanes, winter storms, droughts and wildfires increasingly touch our lives either directly or indirectly. These events draw our attention to the seriousness of changes in climate not only to human well-being but also to the integrity of ecological systems upon which we depend. Understanding the impacts of extreme events on ecological systems requires the ability to characterize the cascading effects of environmental changes on the environments in which organisms live and the altered biological interactions produced. This scientific ambition represents no small challenge for the study of animal communities, which are typically difficult to census as well as dynamic in time and space. Davis et al. (2022) in a recent study in the Journal of Animal Ecology examined the amphibian and fish communities found in depressional coastal wetlands to better understand how they respond to major rainfall and flooding events. Data from the U.S. Geological Survey's Amphibian Research and Monitoring Initiative provided an 8-year record of observations as well as environmental measurements. For this study, the authors integrated techniques for assessing the dynamics of animal populations with a Bayesian implementation of structural equation modelling. Using their integrated methodological approach permitted the authors to reveal the direct and indirect effects of extreme weather events on co-occurring amphibian and fish communities while accounting for observational uncertainty and temporal variation in population-level processes. Their findings indicate that the most prominent effects of flooding on the amphibian community were caused by changes in the fish community that led to increased predation and resource competition. In their conclusions, the authors emphasize the importance of understanding networks of abiotic and biotic effects if we are to predict and mitigate the influence of extreme weather events.

Heat waves accelerate the spread of infectious diseases.

COVID-19 pandemic appeared summer surge in 2022 worldwide and this contradicts its seasonal fluctuations. Even as high temperature and intense ultraviolet radiation can inhibit viral activity, the number of new cases worldwide has increased to >78% in only 1 month since the summer of 2022 under unchanged virus mutation influence and control policies. Using the attribution analysis based on the theoretical infectious diseases model simulation, we found the mechanism of the severe COVID-19 outbreak in the summer of 2022 and identified the amplification effect of heat wave events on its magnitude. The results suggest that approximately 69.3% of COVID-19 cases this summer could have been avoided if there is no heat waves. The collision between the pandemic and the heatwave is not an accident. Climate change is leading to more frequent extreme climate events and an increasing number of infectious diseases, posing an urgent threat to human health and life. Therefore, public health authorities must quickly develop coordinated management plans to deal with the simultaneous occurrence of extreme climate events and infectious diseases.

Using machine learning to predict extreme events in complex systems.

Extreme events and the related anomalous statistics are ubiquitously observed in many natural systems, and the development of efficient methods to understand and accurately predict such representative features remains a grand challenge. Here, we investigate the skill of deep learning strategies in the prediction of extreme events in complex turbulent dynamical systems. Deep neural networks have been successfully applied to many imaging processing problems involving big data, and have recently shown potential for the study of dynamical systems. We propose to use a densely connected mixed-scale network model to capture the extreme events appearing in a truncated Korteweg-de Vries (tKdV) statistical framework, which creates anomalous skewed distributions consistent with recent laboratory experiments for shallow water waves across an abrupt depth change, where a remarkable statistical phase transition is generated by varying the inverse temperature parameter in the corresponding Gibbs invariant measures. The neural network is trained using data without knowing the explicit model dynamics, and the training data are only drawn from the near-Gaussian regime of the tKdV model solutions without the occurrence of large extreme values. A relative entropy loss function, together with empirical partition functions, is proposed for measuring the accuracy of the network output where the dominant structures in the turbulent field are emphasized. The optimized network is shown to gain uniformly high skill in accurately predicting the solutions in a wide variety of statistical regimes, including highly skewed extreme events. The technique is promising to be further applied to other complicated high-dimensional systems.