2024-10-30 | | Total: 5
In this study, we present the integration of a neural network-based parameterization into the global atmospheric model ARP-GEM1, leveraging the Python interface of the OASIS coupler. This approach facilitates the exchange of fields between the Fortran-based ARP-GEM1 model and a Python component responsible for neural network inference. As a proof-of-concept experiment, we trained a neural network to emulate the deep convection parameterization of ARP-GEM1. Using the flexible Fortran/Python interface, we have successfully replaced ARP-GEM1's deep convection scheme with a neural network emulator. To assess the performance of the neural network deep convection scheme, we have run a 5-years ARP-GEM1 simulation using the neural network emulator. The evaluation of averaged fields showed good agreement with output from an ARP-GEM1 simulation using the physics-based deep convection scheme. The Python component was deployed on a separate partition from the general circulation model, using GPUs to increase inference speed of the neural network.
Climate models serve as critical tools for evaluating the effects of climate change and projecting future climate scenarios. However, the reliance on numerical simulations of physical equations renders them computationally intensive and inefficient. While deep learning methodologies have made significant progress in weather forecasting, they are still unstable for climate emulation tasks. Here, we propose PACE, a lightweight 684K parameter Physics Informed Uncertainty Aware Climate Emulator. PACE emulates temperature and precipitation stably for 86 years while only being trained on greenhouse gas emissions data. We incorporate a fundamental physical law of advection-diffusion in PACE accounting for boundary conditions and empirically estimating the diffusion co-efficient and flow velocities from emissions data. PACE has been trained on 15 climate models provided by ClimateSet outperforming baselines across most of the climate models and advancing a new state of the art in a climate diagnostic task.
Tropical cyclones (TCs) are powerful, natural phenomena that can severely impact populations and infrastructure. Enhancing our understanding of the mechanisms driving their intensification is crucial for mitigating these impacts. To this end, researchers are pushing the boundaries of TC simulation resolution down to scales of just a few meters. However, higher resolution simulations come with significant computational challenges, increasing both time and energy costs. Adaptive mesh refinement (AMR) is a technique widely used in computational fluid dynamics but has seen limited application in atmospheric simulations. This study explores the use of h-adaptive grids using the spectral element discretization technique to accelerate TC simulations while allowing very high resolutions in certain parts of the domain. By applying AMR to a rapidly intensifying TC test case, we demonstrate that AMR can replicate the results of uniform grid simulations in terms of mean and local wind speed maxima while dramatically reducing computational costs. We show that AMR can speed up dry TC simulations by a factor of 2 - 13 for the set of tested refinement criteria. Additionally, we show that TC intensity changes as resolution is increased and that AMR can deliver high-resolution simulations at the cost of coarser static simulations. Our findings indicate that AMR and spectral element methods are promising tools for enhancing TC simulations.
In this paper, we conducted a detailed analysis of optical turbulence in the Atmospheric Surface Layer (ASL) at Muztagh-ata site during on-site testing. We utilized ultrasonic anemometers positioned on a 30-meter tower to collect and process data at five height levels, obtaining data from October 1, 2021 to the present. We investigated the behavior of optical turbulence parameters (\(C_n^2\) and seeing \(\varepsilon\)) in the ASL. Nighttime \(C_n^2\) primarily fluctuated in the range of \(10^{-16}\) to \(10^{-14}\), exhibiting an exponential decrease with height. During the day, it showed a \(h^{-0.82}\) dependency, while at night, it displayed a \(h^{-0.48}\) dependency. Additionally, we presented the distribution of seeing across different layers within the ASL, showing a gradual decrease with increasing height, with a median seeing of 0.24 arcseconds at nighttime and 0.48 arcseconds at daytime between 6-30m. We investigated the relationship between surface temperature inversion, seeing in the ASL, and wind speed at the site. Our results show that under temperature inversion conditions, seeing significantly improves and is often accompanied by low to moderate wind speeds, while high wind speeds are usually associated with poorer seeing. Preliminary calculations and observational results, combined with the high altitude and unique geographical location, suggest that Muztagh-ata site has the potential to be an outstanding optical astronomical observatory in the western plateau of china.
Cloud cover plays a critical role in weather prediction and impacts several sectors, including agriculture, solar power generation, and aviation. Despite advancements in numerical weather prediction (NWP) models, forecasting total cloud cover remains challenging due to the small-scale nature of cloud formation processes. In this study, we introduce CloudCast, a convolutional neural network (CNN) based on the U-Net architecture, designed to predict total cloud cover (TCC) up to five hours ahead. Trained on five years of satellite data, CloudCast significantly outperforms traditional NWP models and optical flow methods. Compared to a reference NWP model, CloudCast achieves a 24% lower mean absolute error and reduces multi-category prediction errors by 46%. The model demonstrates strong performance, particularly in capturing the large-scale structure of cloud cover in the first few forecast hours, though later predictions are subject to blurring and underestimation of cloud formation. An ablation study identified the optimal input features and loss functions, with MAE-based models performing the best. CloudCast has been integrated into the Finnish Meteorological Institute's operational nowcasting system, where it improves cloud cover forecasts used by public and private sector clients. While CloudCast is limited by a relatively short skillful lead time of about three hours, future work aims to extend this through more complex network architectures and higher-resolution data. CloudCast code is available at https://github.com/fmidev/cloudcast.