Search

A two-stage super-resolution simulation method is proposed for street-scale air temperature and wind velocity, which considerably reduces the computation time while maintaining accuracy. The first stage employs a convolutional neural network (CNN) to correct large-scale flows above buildings in the input low-resolution (LR) simulation results. The second stage uses another CNN to reconstruct small-scale flows between buildings from the output of the first stage, resulting in high-resolution (HR) inferences. The CNNs are trained using HR simulation data for the second stage and their coarse-grained version for the first stage as the ground truth, where the HR simulations are conducted independently of the LR simulations used as input. This learning approach separates the spatial scales of inference in each stage. The effectiveness of the proposed method was evaluated using micrometeorological simulations in an actual urban area around Tokyo Station in Japan. The super-resolution simulation successfully inferred HR atmospheric flows, reducing errors by approximately 50% for air temperature and 60% for wind velocity compared to the LR simulations. Furthermore, the two-stage approach allowed for localized HR inferences, reducing GPU memory usage to as low as 12% during the training phase. The total wall-clock time for 60-min predictions was reduced to 6.83 min, which was approximately 3.32% of the HR simulation time (206 min). The proposed method demonstrates the potential for real-time micrometeorology predictions in urban areas with a combination of physics-based and data-driven models.

This study proposes a framework that concurrently optimizes the assimilation and generation processes of observations. This is implemented using a conditional variational autoencoder (CVAE) composed of an encoder and a decoder. The encoder performs assimilation by estimating the true state from the background state and observations at a given time, while the decoder serves as an observation operator and generates observations from the estimated true state. The loss function for the unsupervised learning of CVAEs, namely the evidence lower bound, can be regarded as a cost function for three-dimensional variational (3D-Var) assimilation, providing a theoretical basis for the use of CVAEs. Moreover, by leveraging the arbitrariness of the prior distribution, super-resolution data assimilation (SRDA) is developed within the proposed framework. Specifically, the background state is super-resolved by an SR neural-network model and is then used in training the CVAE. This approach enables the encoder to execute SR and DA simultaneously. The proposed SRDA is evaluated through numerical experiments on an idealized barotropic ocean jet system. The SR model can be trained using the observations as target data, despite most grid points being missing. This achievement is attributable to the translation equivariance of convolution. Comparisons with an ensemble Kalman filter demonstrate the superior performance and reduced computational cost of the SRDA. Importantly, the present SRDA can infer high-resolution forecasts at any time step through the combination of the SR model and CVAE. This research provides a theoretical foundation and motivation for the use of deep generative models, specifically CVAEs, in DA.

Atmospheric simulations for urban cities can be computationally intensive because of the need for high spatial resolution, such as a few meters, to accurately represent buildings and streets. Deep learning has recently gained attention across various physical sciences for its potential to reduce computational cost. Super-resolution is one such technique that enhances the resolution of data. This paper proposes a convolutional neural network (CNN) that super-resolves instantaneous snapshots of three-dimensional air temperature and wind velocity fields for urban micrometeorology. This super-resolution process requires not only an increase in spatial resolution but also the restoration of missing data caused by the difference in the building shapes that depend on the resolution. The proposed CNN incorporates gated convolution, which is an image inpainting technique that infers missing pixels. The CNN performance has been verified via supervised learning utilizing building-resolving micrometeorological simulations around Tokyo Station in Japan. The CNN successfully reconstructed the temperature and velocity fields around the high-resolution buildings, despite the missing data at lower altitudes due to the coarseness of the low-resolution buildings. This result implies that near-surface flows can be inferred from flows above buildings. This hypothesis was assessed via numerical experiments where all input values below a certain height were made missing. This research suggests the possibility that building-resolving micrometeorological simulations become more practical for urban cities with the aid of neural networks that enhance computational efficiency.

Deep learning has recently gained attention in the atmospheric and oceanic sciences for its potential to improve the accuracy of numerical simulations or to reduce computational costs. Super-resolution is one such technique for high-resolution inference from low-resolution data. This paper proposes a new scheme, called four-dimensional super-resolution data assimilation (4D-SRDA). This framework calculates the time evolution of a system from low-resolution simulations using a physics-based model, while a trained neural network simultaneously performs data assimilation and spatio-temporal super-resolution. The use of low-resolution simulations without ensemble members reduces the computational cost of obtaining inferences at high spatio-temporal resolution. In 4D-SRDA, physics-based simulations and neural-network inferences are performed alternately, possibly causing a domain shift, i.e., a statistical difference between the training and test data, especially in offline training. Domain shifts can reduce the accuracy of inference. To mitigate this risk, we developed super-resolution mixup (SR-mixup)--a data augmentation method for domain generalization. SR-mixup creates a linear combination of randomly sampled inputs, resulting in synthetic data with a different distribution from the original data. The proposed methods were validated using an idealized barotropic ocean jet with supervised learning. The results suggest that the combination of 4D-SRDA and SR-mixup is effective for robust inference cycles. This study highlights the potential of super-resolution and domain-generalization techniques, in the field of data assimilation, especially for the integration of physics-based and data-driven models.

This paper investigates the super-resolution (SR) of velocity fields in two-dimensional fluids from the viewpoint of rotational equivariance. SR refers to techniques that estimate high-resolution images from those in low resolution and has lately been applied in fluid mechanics. The rotational equivariance of SR models is defined as the property in which the super-resolved velocity field is rotated according to a rotation of the input, which leads to the inference covariant to the orientation of fluid systems. Generally, the covariance in physics is related to symmetries. To clarify a relationship to symmetries, the rotational consistency of datasets for SR is newly introduced as the invariance of pairs of low- and high-resolution velocity fields with respect to rotation. This consistency is sufficient and necessary for SR models to acquire rotational equivariance from large datasets with supervised learning. Such a large dataset is not required when rotational equivariance is imposed on SR models through weight sharing of convolution kernels as prior knowledge. Even if a fluid system has rotational symmetry, this symmetry may not carry over to a velocity dataset, which is not rotationally consistent. This inconsistency can occur when the rotation does not commute with the generation of low-resolution velocity fields. These theoretical suggestions are supported by the results from numerical experiments, where two existing convolutional neural networks (CNNs) are converted into rotationally equivariant CNNs and the inferences of the four CNNs are compared after the supervised training.

Effect of surface tension reduction on wind-wave growth is investigated using direct numerical simulation (DNS) of air-water two-phase turbulent flow. The incompressible Navier-Stokes equations for air and water sides are solved using an arbitrary Lagrangian-Eulerian method with boundary-fitted moving grids. The wave growth of finite-amplitude and non-breaking gravity--capillary waves, whose wavelength is less than 0.07 m, is simulated for two cases of different surface tensions under a low--wind speed condition of several meters per second. The results show that significant wave height for the smaller surface tension case increases faster than that for the larger surface tension case. Energy fluxes for gravity and capillary wave scales reveal that, when the surface tension is reduced, the energy transfer from the significant gravity waves to capillary waves decreases and the significant waves accumulate more energy supplied by wind. This results in faster wave growth for the smaller surface tension case. Effect on the scalar transfer across the air-water interface is also investigated. The results show that the scalar transfer coefficient on the water side decreases due to the surface tension reduction. The decrease is caused by suppression of turbulence in the water side. In order to support the conjecture, the surface tension effect is compared with laboratory experiments in a small wind-wave tank.

The present paper proposes a super-resolution (SR) model based on a convolutional neural network and applies it to the near-surface temperature in urban areas. The SR model incorporates a skip connection, a channel attention mechanism, and separated feature extractors for the inputs of temperature, building height, downward shortwave radiation, and horizontal velocity. We train the SR model with sets of low-resolution (LR) and high-resolution (HR) images from building-resolving large-eddy simulations (LESs) in a city, where the horizontal resolutions of LR and HR are 20 and 5 m, respectively. The generalization capability of the SR model is confirmed with LESs in another city. The estimated HR temperature fields are more accurate than those of the bicubic interpolation and image SR model that takes only the temperature as its input. Except for the temperature input, the building height is the most important to reconstruct the HR temperature and enables the SR model to reduce errors in temperature near building boundaries. The SR model considers the appropriate boundary for each building from its height information. The analysis of attention weights indicates that the importance of the building height increases as the downward shortwave radiation becomes larger. The contrast between sun and shade is strengthened with the increase in solar radiation, which may affect the temperature distribution. The short inference time suggests the potential of the proposed SR model to facilitate a real-time HR prediction in metropolitan areas by combining it with an LR building-resolving LES model., Comment: accepted by Building and Environment

Storm trainings, consisting of line-shaped aggregates of cumulonimbi, bring persistent local heavy rains, often causing devastating floods and landslides. Weather control techniques could in theory help prevent such disasters, but so far successful weather control has been limited to local rain initiation or the diffusion of local clouds. No successful strategies have been proposed for the control of mesoscale storms. Here we show that a strategic use of consumer air conditioners, which can typically remove about 1kg of moisture from the air per hour when run in dehumidification mode, and which are installed in large numbers in big cities, can modulate a storm downstream. We numerically reproduced a storm training that affected the Hiroshima metropolis in Japan in 2014, and conducted experiments to test the sensitivity of the storm to the initial moisture field near the surface. We propose an empirically-derived formula for a control efficiency parameter, which can be used to estimate the impact of moisture removal on the rainfall accumulation. It reveals that removal of half a kiloton of moisture, which could be achieved within half an hour in a city with a population of one million since more than one air conditioner is installed per capita in Japan, could lead to a significant reduction of the total rainfall accumulation over a 100 km2 area of heavy rain during the storm event. Conversely, our results indicate that some summertime storms occurring inside or near a metropolis could be strengthened by the excess moisture discharged from large numbers of air conditioners used for cooling rooms. We anticipate our results, which reveal that human activity can have a significant impact on storms, will be a starting point for considering the coupling of weather and the economy, and will contribute to the development of a sustainable society.

We propose a super-resolution (SR) simulation system that consists of a physics-based meteorological simulation and an SR method based on a deep convolutional neural network (CNN). The CNN is trained using pairs of high-resolution (HR) and low-resolution (LR) images created from meteorological simulation results for different resolutions so that it can map LR simulation images to HR ones. The proposed SR simulation system, which performs LR simulations, can provide HR prediction results in much shorter operating cycles than those required for corresponding HR simulation prediction system. We apply the SR simulation system to urban micrometeorology, which is strongly affected by buildings and human activity. Urban micrometeorology simulations that need to resolve urban buildings are computationally costly and thus cannot be used for operational real-time predictions even when run on supercomputers. We performed HR micrometeorology simulations on a supercomputer to obtain datasets for training the CNN in the SR method. It is shown that the proposed SR method can be used with a spatial scaling factor of 4 and that it outperforms conventional interpolation methods by a large margin. It is also shown that the proposed SR simulation system has the potential to be used for operational urban micrometeorology predictions.

A wavelet-based method for compression of three-dimensional simulation data is presented and its software framework is described. It uses wavelet decomposition and subsequent range coding with quantization suitable for floating-point data. The effectiveness of this method is demonstrated by applying it to example numerical tests, ranging from idealized configurations to realistic global-scale simulations.

Understanding rare events in turbulence provides a basis for the science of extreme weather, for which the atmosphere is modeled by Navier-Stokes equations (NSEs). In solutions of NSEs for isotropic fluids, various quantities, such as fluid velocities, roughly follow Gaussian distributions, where extreme events are prominent only in small-scale quantities associated with the dissipation-dominating length scale or anomalous scaling regime. Using numerical simulations, this study reveals another universal promotion mechanism at much larger scales if three-dimensional fluids accompany strong two-dimensional anisotropies, as is the case in the atmosphere. The dimensional crossover between two and three dimensions generates prominent fat-tailed non-Gaussian distributions with intermittency accompanied by colossal chain-like structures with densely populated self-organized vortices (serpentinely organized vortices (SOV)). The promotion is caused by a sudden increase of the available phase space at the crossover length scale. Since the discovered intermittency can involve much larger energies than those in the conventional intermittency in small spatial scales, it governs extreme events and chaotic unpredictability in the synoptic weather system.