Accurate precipitation forecasts are crucial for applications such as flood management, agricultural planning, water resource allocation, and weather warnings. Despite advances in numerical weather prediction (NWP) models, they still exhibit significant biases and uncertainties, especially at high spatial and temporal resolutions. To address these limitations, we explore uncertainty-aware deep learning models for post-processing daily cumulative quantitative precipitation forecasts to obtain forecast uncertainties that lead to a better trade-off between accuracy and reliability. Our study compares different state-of-the-art models, and we propose a variant of the well-known SDE-Net, called SDE U-Net, tailored to segmentation problems like ours. We evaluate its performance for both typical and intense precipitation events. Our results show that all deep learning models significantly outperform the average baseline NWP solution, with our implementation of the SDE U-Net showing the best trade-off between accuracy and reliability. Integrating these models, which account for uncertainty, into operational forecasting systems can improve decision-making and preparedness for weather-related events.
{"title":"Uncertainty-aware segmentation for rainfall prediction post processing","authors":"Simone Monaco, Luca Monaco, Daniele Apiletti","doi":"arxiv-2408.16792","DOIUrl":"https://doi.org/arxiv-2408.16792","url":null,"abstract":"Accurate precipitation forecasts are crucial for applications such as flood\u0000management, agricultural planning, water resource allocation, and weather\u0000warnings. Despite advances in numerical weather prediction (NWP) models, they\u0000still exhibit significant biases and uncertainties, especially at high spatial\u0000and temporal resolutions. To address these limitations, we explore\u0000uncertainty-aware deep learning models for post-processing daily cumulative\u0000quantitative precipitation forecasts to obtain forecast uncertainties that lead\u0000to a better trade-off between accuracy and reliability. Our study compares\u0000different state-of-the-art models, and we propose a variant of the well-known\u0000SDE-Net, called SDE U-Net, tailored to segmentation problems like ours. We\u0000evaluate its performance for both typical and intense precipitation events. Our results show that all deep learning models significantly outperform the\u0000average baseline NWP solution, with our implementation of the SDE U-Net showing\u0000the best trade-off between accuracy and reliability. Integrating these models,\u0000which account for uncertainty, into operational forecasting systems can improve\u0000decision-making and preparedness for weather-related events.","PeriodicalId":501166,"journal":{"name":"arXiv - PHYS - Atmospheric and Oceanic Physics","volume":"5 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2024-08-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142215410","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Computational models of atmospheric composition are not always physically consistent. For example, not all models respect fundamental conservation laws such as conservation of atoms in an interconnected chemical system. In well performing models, these nonphysical deviations are often ignored because they are frequently minor, and thus only need a small nudge to perfectly conserve mass. Here we introduce a method that anchors a prediction from any numerical model to physically consistent hard constraints, nudging concentrations to the nearest solution that respects the conservation laws. This closed-form model-agnostic correction uses a single matrix operation to minimally perturb the predicted concentrations to ensure that atoms are conserved to machine precision. To demonstrate this approach, we train a gradient boosting decision tree ensemble to emulate a small reference model of ozone photochemistry and test the effect of the correction on accurate but non-conservative predictions. The nudging approach minimally perturbs the already well-predicted results for most species, but decreases the accuracy of important oxidants, including radicals. We develop a weighted extension of this nudging approach that considers the uncertainty and magnitude of each species in the correction. This species-level weighting approach is essential to accurately predict important low concentration species such as radicals. We find that applying the uncertainty-weighted correction to the nonphysical predictions slightly improves overall accuracy, by nudging the predictions to a more likely mass-conserving solution.
{"title":"A nudge to the truth: atom conservation as a hard constraint in models of atmospheric composition using an uncertainty-weighted correction","authors":"Patrick Obin Sturm, Sam J. Silva","doi":"arxiv-2408.16109","DOIUrl":"https://doi.org/arxiv-2408.16109","url":null,"abstract":"Computational models of atmospheric composition are not always physically\u0000consistent. For example, not all models respect fundamental conservation laws\u0000such as conservation of atoms in an interconnected chemical system. In well\u0000performing models, these nonphysical deviations are often ignored because they\u0000are frequently minor, and thus only need a small nudge to perfectly conserve\u0000mass. Here we introduce a method that anchors a prediction from any numerical\u0000model to physically consistent hard constraints, nudging concentrations to the\u0000nearest solution that respects the conservation laws. This closed-form\u0000model-agnostic correction uses a single matrix operation to minimally perturb\u0000the predicted concentrations to ensure that atoms are conserved to machine\u0000precision. To demonstrate this approach, we train a gradient boosting decision\u0000tree ensemble to emulate a small reference model of ozone photochemistry and\u0000test the effect of the correction on accurate but non-conservative predictions.\u0000The nudging approach minimally perturbs the already well-predicted results for\u0000most species, but decreases the accuracy of important oxidants, including\u0000radicals. We develop a weighted extension of this nudging approach that\u0000considers the uncertainty and magnitude of each species in the correction. This\u0000species-level weighting approach is essential to accurately predict important\u0000low concentration species such as radicals. We find that applying the\u0000uncertainty-weighted correction to the nonphysical predictions slightly\u0000improves overall accuracy, by nudging the predictions to a more likely\u0000mass-conserving solution.","PeriodicalId":501166,"journal":{"name":"arXiv - PHYS - Atmospheric and Oceanic Physics","volume":"21 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2024-08-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142215414","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pritthijit Nath, Henry Moss, Emily Shuckburgh, Mark Webb
This study explores integrating reinforcement learning (RL) with idealised climate models to address key parameterisation challenges in climate science. Current climate models rely on complex mathematical parameterisations to represent sub-grid scale processes, which can introduce substantial uncertainties. RL offers capabilities to enhance these parameterisation schemes, including direct interaction, handling sparse or delayed feedback, continuous online learning, and long-term optimisation. We evaluate the performance of eight RL algorithms on two idealised environments: one for temperature bias correction, another for radiative-convective equilibrium (RCE) imitating real-world computational constraints. Results show different RL approaches excel in different climate scenarios with exploration algorithms performing better in bias correction, while exploitation algorithms proving more effective for RCE. These findings support the potential of RL-based parameterisation schemes to be integrated into global climate models, improving accuracy and efficiency in capturing complex climate dynamics. Overall, this work represents an important first step towards leveraging RL to enhance climate model accuracy, critical for improving climate understanding and predictions. Code accessible at https://github.com/p3jitnath/climate-rl.
{"title":"RAIN: Reinforcement Algorithms for Improving Numerical Weather and Climate Models","authors":"Pritthijit Nath, Henry Moss, Emily Shuckburgh, Mark Webb","doi":"arxiv-2408.16118","DOIUrl":"https://doi.org/arxiv-2408.16118","url":null,"abstract":"This study explores integrating reinforcement learning (RL) with idealised\u0000climate models to address key parameterisation challenges in climate science.\u0000Current climate models rely on complex mathematical parameterisations to\u0000represent sub-grid scale processes, which can introduce substantial\u0000uncertainties. RL offers capabilities to enhance these parameterisation\u0000schemes, including direct interaction, handling sparse or delayed feedback,\u0000continuous online learning, and long-term optimisation. We evaluate the\u0000performance of eight RL algorithms on two idealised environments: one for\u0000temperature bias correction, another for radiative-convective equilibrium (RCE)\u0000imitating real-world computational constraints. Results show different RL\u0000approaches excel in different climate scenarios with exploration algorithms\u0000performing better in bias correction, while exploitation algorithms proving\u0000more effective for RCE. These findings support the potential of RL-based\u0000parameterisation schemes to be integrated into global climate models, improving\u0000accuracy and efficiency in capturing complex climate dynamics. Overall, this\u0000work represents an important first step towards leveraging RL to enhance\u0000climate model accuracy, critical for improving climate understanding and\u0000predictions. Code accessible at https://github.com/p3jitnath/climate-rl.","PeriodicalId":501166,"journal":{"name":"arXiv - PHYS - Atmospheric and Oceanic Physics","volume":"98 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2024-08-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142215415","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Enno Tiemann, Shanyu Zhou, Alexander Kläser, Konrad Heidler, Rochelle Schneider, Xiao Xiang Zhu
Methane ($CH_4$) is a potent anthropogenic greenhouse gas, contributing 86 times more to global warming than Carbon Dioxide ($CO_2$) over 20 years, and it also acts as an air pollutant. Given its high radiative forcing potential and relatively short atmospheric lifetime (9$pm$1 years), methane has important implications for climate change, therefore, cutting methane emissions is crucial for effective climate change mitigation. This work expands existing information on operational methane point source detection sensors in the Short-Wave Infrared (SWIR) bands. It reviews the state-of-the-art for traditional as well as Machine Learning (ML) approaches. The architecture and data used in such ML models will be discussed separately for methane plume segmentation and emission rate estimation. Traditionally, experts rely on labor-intensive manually adjusted methods for methane detection. However, ML approaches offer greater scalability. Our analysis reveals that ML models outperform traditional methods, particularly those based on convolutional neural networks (CNN), which are based on the U-net and transformer architectures. These ML models extract valuable information from methane-sensitive spectral data, enabling a more accurate detection. Challenges arise when comparing these methods due to variations in data, sensor specifications, and evaluation metrics. To address this, we discuss existing datasets and metrics, providing an overview of available resources and identifying open research problems. Finally, we explore potential future advances in ML, emphasizing approaches for model comparability, large dataset creation, and the European Union's forthcoming methane strategy.
甲烷（$CH_4$）是一种强效的人为温室气体，在 20 年内对全球变暖的贡献是二氧化碳（$CO_2$）的 86 倍，它也是一种空气污染物。鉴于甲烷的高辐射强迫潜力和相对较短的大气寿命（9 年/pm$1 年），甲烷对气候变化具有重要影响，因此，减少甲烷排放对有效减缓气候变化至关重要。这项工作扩展了短波红外（SWIR）波段甲烷点源探测传感器的现有信息。它回顾了最先进的传统方法和机器学习（ML）方法。将分别讨论甲烷羽流细分和排放率估算中使用的 ML 模型的架构和数据。传统上，专家们依靠劳动密集型的人工调整方法来检测甲烷。然而，ML 方法具有更大的可扩展性。我们的分析表明，ML 模型优于传统方法，特别是那些基于卷积神经网络（CNN）的方法，后者是基于 U 型网和变压器架构。这些 ML 模型能从甲烷敏感光谱数据中提取有价值的信息，从而实现更准确的检测。由于数据、传感器规格和评估指标的不同，在比较这些方法时会遇到挑战。为了解决这个问题，我们讨论了现有的数据集和指标，概述了可用资源，并指出了有待解决的研究问题。最后，我们探讨了 ML 未来的潜在发展，强调了模型可比性、大型数据集创建和欧盟即将推出的甲烷战略等方面的方法。
{"title":"Machine Learning for Methane Detection and Quantification from Space - A survey","authors":"Enno Tiemann, Shanyu Zhou, Alexander Kläser, Konrad Heidler, Rochelle Schneider, Xiao Xiang Zhu","doi":"arxiv-2408.15122","DOIUrl":"https://doi.org/arxiv-2408.15122","url":null,"abstract":"Methane ($CH_4$) is a potent anthropogenic greenhouse gas, contributing 86\u0000times more to global warming than Carbon Dioxide ($CO_2$) over 20 years, and it\u0000also acts as an air pollutant. Given its high radiative forcing potential and\u0000relatively short atmospheric lifetime (9$pm$1 years), methane has important\u0000implications for climate change, therefore, cutting methane emissions is\u0000crucial for effective climate change mitigation. This work expands existing\u0000information on operational methane point source detection sensors in the\u0000Short-Wave Infrared (SWIR) bands. It reviews the state-of-the-art for\u0000traditional as well as Machine Learning (ML) approaches. The architecture and\u0000data used in such ML models will be discussed separately for methane plume\u0000segmentation and emission rate estimation. Traditionally, experts rely on\u0000labor-intensive manually adjusted methods for methane detection. However, ML\u0000approaches offer greater scalability. Our analysis reveals that ML models\u0000outperform traditional methods, particularly those based on convolutional\u0000neural networks (CNN), which are based on the U-net and transformer\u0000architectures. These ML models extract valuable information from\u0000methane-sensitive spectral data, enabling a more accurate detection. Challenges\u0000arise when comparing these methods due to variations in data, sensor\u0000specifications, and evaluation metrics. To address this, we discuss existing\u0000datasets and metrics, providing an overview of available resources and\u0000identifying open research problems. Finally, we explore potential future\u0000advances in ML, emphasizing approaches for model comparability, large dataset\u0000creation, and the European Union's forthcoming methane strategy.","PeriodicalId":501166,"journal":{"name":"arXiv - PHYS - Atmospheric and Oceanic Physics","volume":"59 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2024-08-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142215441","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
With the increasing amount of renewable energy in the grid, long-term wind power forecasting for multiple decades becomes more critical. In these long-term forecasts, climate data is essential as it allows us to account for climate change. Yet the resolution of climate models is often very coarse. In this paper, we show that by including turbine locations when downscaling with Gaussian Processes, we can generate valuable aggregate wind power predictions despite the low resolution of the CMIP6 climate models. This work is a first step towards multi-decadal turbine-location-aware wind power forecasting using global climate model output.
{"title":"Towards turbine-location-aware multi-decadal wind power predictions with CMIP6","authors":"Nina Effenberger, Nicole Ludwig","doi":"arxiv-2408.14889","DOIUrl":"https://doi.org/arxiv-2408.14889","url":null,"abstract":"With the increasing amount of renewable energy in the grid, long-term wind\u0000power forecasting for multiple decades becomes more critical. In these\u0000long-term forecasts, climate data is essential as it allows us to account for\u0000climate change. Yet the resolution of climate models is often very coarse. In\u0000this paper, we show that by including turbine locations when downscaling with\u0000Gaussian Processes, we can generate valuable aggregate wind power predictions\u0000despite the low resolution of the CMIP6 climate models. This work is a first\u0000step towards multi-decadal turbine-location-aware wind power forecasting using\u0000global climate model output.","PeriodicalId":501166,"journal":{"name":"arXiv - PHYS - Atmospheric and Oceanic Physics","volume":"42 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2024-08-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142215440","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Rapid deployment of offshore wind is expected within the coming decades to help meet climate goals. With offshore wind turbine lifetimes of 25-30 years, and new offshore leases spanning 60 years, it is vital to consider long-term changes in potential wind power resource at the farm planning stage. Such changes may arise from multiple sources, including climate change, and increasing wake-induced power losses. In this work, we investigate and compare these two sources of long-term change in wind power, for a case study consisting of 21 wind farms within the German Bight. Consistent with previous studies, we find a small but significant reduction in wind resource due to climate change by the end of the 21st century under the high-emission RCP8.5 scenario, compared with a historical period, with a mean power reduction (over an ensemble of seven climate models) of 2.1%. To assess the impact of wake-induced losses due to increasingly dense farm build-out, we model wakes within the German Bight region using an engineering wake model, under various stages of (planned) build-out corresponding to the years 2010-2027. By identifying clusters of wind farms, we decompose wake effects into long-range (inter-cluster), medium-range (intra-cluster) and short-range (intra-farm) effects. Inter-cluster wake-induced losses increase from 0 for the 2010 scenario to 2.5% for the 2027 scenario, with intra-cluster losses also increasing from 0 to 4.3%. Intra-farm losses are relatively constant, at around 13%. While the evolution of wake effects therefore outweighs the climate effect, and impacts over a shorter timescale, both factors are significant. We also find evidence of non-linear interactions between the climate and wake effects. Both climate change and evolving wake effects must therefore be considered within resource assessment and wind farm planning.
{"title":"The future of offshore wind power production: wake and climate impacts","authors":"Simon C Warder, Matthew D Piggott","doi":"arxiv-2408.14963","DOIUrl":"https://doi.org/arxiv-2408.14963","url":null,"abstract":"Rapid deployment of offshore wind is expected within the coming decades to\u0000help meet climate goals. With offshore wind turbine lifetimes of 25-30 years,\u0000and new offshore leases spanning 60 years, it is vital to consider long-term\u0000changes in potential wind power resource at the farm planning stage. Such\u0000changes may arise from multiple sources, including climate change, and\u0000increasing wake-induced power losses. In this work, we investigate and compare\u0000these two sources of long-term change in wind power, for a case study\u0000consisting of 21 wind farms within the German Bight. Consistent with previous\u0000studies, we find a small but significant reduction in wind resource due to\u0000climate change by the end of the 21st century under the high-emission RCP8.5\u0000scenario, compared with a historical period, with a mean power reduction (over\u0000an ensemble of seven climate models) of 2.1%. To assess the impact of\u0000wake-induced losses due to increasingly dense farm build-out, we model wakes\u0000within the German Bight region using an engineering wake model, under various\u0000stages of (planned) build-out corresponding to the years 2010-2027. By\u0000identifying clusters of wind farms, we decompose wake effects into long-range\u0000(inter-cluster), medium-range (intra-cluster) and short-range (intra-farm)\u0000effects. Inter-cluster wake-induced losses increase from 0 for the 2010\u0000scenario to 2.5% for the 2027 scenario, with intra-cluster losses also\u0000increasing from 0 to 4.3%. Intra-farm losses are relatively constant, at around\u000013%. While the evolution of wake effects therefore outweighs the climate\u0000effect, and impacts over a shorter timescale, both factors are significant. We\u0000also find evidence of non-linear interactions between the climate and wake\u0000effects. Both climate change and evolving wake effects must therefore be\u0000considered within resource assessment and wind farm planning.","PeriodicalId":501166,"journal":{"name":"arXiv - PHYS - Atmospheric and Oceanic Physics","volume":"2 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2024-08-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142215439","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
In this work, we assess global offshore wind energy resources, wake-induced losses, array layout optimisation potential and climate change impacts. We first map global offshore ambient wind resource from reanalysis data. We estimate wake-induced losses using an engineering wake model, revealing that locations with low (high) resource typically experience larger (smaller) percentage losses. However, we further find that the specific wind speed distribution is important, with narrower distributions generally leading to greater losses. This is due to the overlap between the wind speed distribution and the high-sensitivity region of the turbine thrust and power curves. Broadly, this leads to much stronger wake-induced losses in the tropics (which experience the trade winds) than mid-latitudes. However, the tropics also experience a narrower wind direction distribution; our results demonstrate that this leads to greater potential for mitigation of wake effects via layout optimisation. Finally, we assess projected changes in wind resource and wake losses due to climate change under a high-emission scenario. Many regions are projected to decrease in ambient wind resources, and furthermore these regions will typically experience greater wake-induced losses, exacerbating the climate impact. These results highlight the different challenges and opportunities associated with exploiting offshore wind resources across the globe.
{"title":"Mapping global offshore wind resource: wake losses, optimisation potential and climate effects","authors":"Simon C Warder, Matthew D Piggott","doi":"arxiv-2408.15028","DOIUrl":"https://doi.org/arxiv-2408.15028","url":null,"abstract":"In this work, we assess global offshore wind energy resources, wake-induced\u0000losses, array layout optimisation potential and climate change impacts. We\u0000first map global offshore ambient wind resource from reanalysis data. We\u0000estimate wake-induced losses using an engineering wake model, revealing that\u0000locations with low (high) resource typically experience larger (smaller)\u0000percentage losses. However, we further find that the specific wind speed\u0000distribution is important, with narrower distributions generally leading to\u0000greater losses. This is due to the overlap between the wind speed distribution\u0000and the high-sensitivity region of the turbine thrust and power curves.\u0000Broadly, this leads to much stronger wake-induced losses in the tropics (which\u0000experience the trade winds) than mid-latitudes. However, the tropics also\u0000experience a narrower wind direction distribution; our results demonstrate that\u0000this leads to greater potential for mitigation of wake effects via layout\u0000optimisation. Finally, we assess projected changes in wind resource and wake\u0000losses due to climate change under a high-emission scenario. Many regions are\u0000projected to decrease in ambient wind resources, and furthermore these regions\u0000will typically experience greater wake-induced losses, exacerbating the climate\u0000impact. These results highlight the different challenges and opportunities\u0000associated with exploiting offshore wind resources across the globe.","PeriodicalId":501166,"journal":{"name":"arXiv - PHYS - Atmospheric and Oceanic Physics","volume":"109 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2024-08-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142215417","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
This work describes a process for efficiently fine-tuning the GraphCast data-driven forecast model to simulate another analysis system, here the Global Deterministic Prediction System (GDPS) of Environment and Climate Change Canada (ECCC). Using two years of training data (July 2019 -- December 2021) and 37 GPU-days of computation to tune the 37-level, quarter-degree version of GraphCast, the resulting model significantly outperforms both the unmodified GraphCast and operational forecast, showing significant forecast skill in the troposphere over lead times from 1 to 10 days. This fine-tuning is accomplished through abbreviating DeepMind's original training curriculum for GraphCast, relying on a shorter single-step forecast stage to accomplish the bulk of the adaptation work and consolidating the autoregressive stages into separate 12hr, 1d, 2d, and 3d stages with larger learning rates. Additionally, training over 3d forecasts is split into two sub-steps to conserve host memory while maintaining a strong correlation with training over the full period.
{"title":"Efficient fine-tuning of 37-level GraphCast with the Canadian global deterministic analysis","authors":"Christopher Subich","doi":"arxiv-2408.14587","DOIUrl":"https://doi.org/arxiv-2408.14587","url":null,"abstract":"This work describes a process for efficiently fine-tuning the GraphCast\u0000data-driven forecast model to simulate another analysis system, here the Global\u0000Deterministic Prediction System (GDPS) of Environment and Climate Change Canada\u0000(ECCC). Using two years of training data (July 2019 -- December 2021) and 37\u0000GPU-days of computation to tune the 37-level, quarter-degree version of\u0000GraphCast, the resulting model significantly outperforms both the unmodified\u0000GraphCast and operational forecast, showing significant forecast skill in the\u0000troposphere over lead times from 1 to 10 days. This fine-tuning is accomplished\u0000through abbreviating DeepMind's original training curriculum for GraphCast,\u0000relying on a shorter single-step forecast stage to accomplish the bulk of the\u0000adaptation work and consolidating the autoregressive stages into separate 12hr,\u00001d, 2d, and 3d stages with larger learning rates. Additionally, training over\u00003d forecasts is split into two sub-steps to conserve host memory while\u0000maintaining a strong correlation with training over the full period.","PeriodicalId":501166,"journal":{"name":"arXiv - PHYS - Atmospheric and Oceanic Physics","volume":"18 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2024-08-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142215442","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Fabrizio Falasca, Aurora Basinski-Ferris, Laure Zanna, Ming Zhao
The energy surplus resulting from radiative forcing causes warming of the Earth system. This initial warming drives a myriad of changes including in sea surface temperatures (SSTs), leading to different radiative feedbacks. The relationship between the radiative feedbacks and the pattern of SST changes is referred to as the "pattern effect". The current approach to study the pattern effect relies on diagnosing the response of atmosphere-only models to perturbations in the SST boundary condition. Here, we argue that the fluctuation-dissipation relation (FDR), together with coarse-graining procedures, is a computationally cheap and theoretically grounded alternative to model experiments. We introduce a protocol to study the pattern effect and present its application in a state-of-the-art coupled climate model. By focusing on the coupled dynamics, we unveil the role of the slow ocean component in setting the pattern effect. We present a new "sensitivity map", representing a first, qualitative prediction of the response of the average top-of-the-atmosphere (TOA) radiative flux to perturbations in the SST field. We find negative sensitivity throughout the tropics, in contrast to the current understanding of a positive-negative dipole of sensitivity in the tropical Pacific. Considering only the shortest time scales, the response is dominated by the fast atmospheric variability and we recover results in qualitative agreement with the literature. Therefore, the difference between our results and previous studies, largely comes from including the atmosphere-ocean coupling. The framework offers a conceptually novel perspective on the pattern effect: feedbacks in the coupled system are encoded in a temporally and spatially dependent response operator, rather than time-independent maps as for previous studies.
{"title":"Diagnosing the pattern effect in the atmosphere-ocean coupled system through linear response theory","authors":"Fabrizio Falasca, Aurora Basinski-Ferris, Laure Zanna, Ming Zhao","doi":"arxiv-2408.12585","DOIUrl":"https://doi.org/arxiv-2408.12585","url":null,"abstract":"The energy surplus resulting from radiative forcing causes warming of the\u0000Earth system. This initial warming drives a myriad of changes including in sea\u0000surface temperatures (SSTs), leading to different radiative feedbacks. The\u0000relationship between the radiative feedbacks and the pattern of SST changes is\u0000referred to as the \"pattern effect\". The current approach to study the pattern\u0000effect relies on diagnosing the response of atmosphere-only models to\u0000perturbations in the SST boundary condition. Here, we argue that the\u0000fluctuation-dissipation relation (FDR), together with coarse-graining\u0000procedures, is a computationally cheap and theoretically grounded alternative\u0000to model experiments. We introduce a protocol to study the pattern effect and\u0000present its application in a state-of-the-art coupled climate model. By\u0000focusing on the coupled dynamics, we unveil the role of the slow ocean\u0000component in setting the pattern effect. We present a new \"sensitivity map\",\u0000representing a first, qualitative prediction of the response of the average\u0000top-of-the-atmosphere (TOA) radiative flux to perturbations in the SST field.\u0000We find negative sensitivity throughout the tropics, in contrast to the current\u0000understanding of a positive-negative dipole of sensitivity in the tropical\u0000Pacific. Considering only the shortest time scales, the response is dominated\u0000by the fast atmospheric variability and we recover results in qualitative\u0000agreement with the literature. Therefore, the difference between our results\u0000and previous studies, largely comes from including the atmosphere-ocean\u0000coupling. The framework offers a conceptually novel perspective on the pattern\u0000effect: feedbacks in the coupled system are encoded in a temporally and\u0000spatially dependent response operator, rather than time-independent maps as for\u0000previous studies.","PeriodicalId":501166,"journal":{"name":"arXiv - PHYS - Atmospheric and Oceanic Physics","volume":"21 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2024-08-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142215444","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Aishwarya Chakravarthy, Dhiman Mondal, John Barrett, Chet Ruszczyk, Pedro Elosegui
Antarctic ice shelves play a vital role in preserving the physical conditions of the Antarctic cryosphere and the Southern Ocean, and beyond. By serving as a buttressing force, ice shelves prevent sea-level rise by restraining the flow of continental ice and glaciers to the sea. Sea-level rise impacts the global environment in multiple ways, including flooding habitats, eroding coastlines, and contaminating soil and groundwater. It is therefore essential to monitor the stability of Antarctic ice shelves, for which a variety of complementary data sources is required. We have developed RIS-Vis, a novel data visualization platform to monitor Antarctic ice shelves. Although focused on the Ross Ice Shelf (RIS), RIS-Vis could be readily scaled to monitor other ice shelves around Antarctica, and elsewhere. Currently, RIS-Vis is capable of analyzing and visualizing seismic, geodetic, and weather data to provide meaningful information for Antarctic cryosphere research. RIS-Vis was built using Python libraries including Obspy, APScheduler, and the Plotly Dash framework, and uses SQLite as the backing database. Visualizations developed on RIS-Vis include filtered seismic waveforms, spectrograms, and power spectral densities, geodetic-based ice-shelf flow, and meteorological variables such as atmospheric temperature and pressure. The dashboard visualization platform abstracts away the time-intensive analysis process of raw data and allows scientists to better concentrate on RIS science.
{"title":"RIS-Vis: A Novel Visualization Platform for Seismic, Geodetic, and Weather Data Relevant to Antarctic Cryosphere Science","authors":"Aishwarya Chakravarthy, Dhiman Mondal, John Barrett, Chet Ruszczyk, Pedro Elosegui","doi":"arxiv-2408.12106","DOIUrl":"https://doi.org/arxiv-2408.12106","url":null,"abstract":"Antarctic ice shelves play a vital role in preserving the physical conditions\u0000of the Antarctic cryosphere and the Southern Ocean, and beyond. By serving as a\u0000buttressing force, ice shelves prevent sea-level rise by restraining the flow\u0000of continental ice and glaciers to the sea. Sea-level rise impacts the global\u0000environment in multiple ways, including flooding habitats, eroding coastlines,\u0000and contaminating soil and groundwater. It is therefore essential to monitor\u0000the stability of Antarctic ice shelves, for which a variety of complementary\u0000data sources is required. We have developed RIS-Vis, a novel data visualization\u0000platform to monitor Antarctic ice shelves. Although focused on the Ross Ice\u0000Shelf (RIS), RIS-Vis could be readily scaled to monitor other ice shelves\u0000around Antarctica, and elsewhere. Currently, RIS-Vis is capable of analyzing\u0000and visualizing seismic, geodetic, and weather data to provide meaningful\u0000information for Antarctic cryosphere research. RIS-Vis was built using Python\u0000libraries including Obspy, APScheduler, and the Plotly Dash framework, and uses\u0000SQLite as the backing database. Visualizations developed on RIS-Vis include\u0000filtered seismic waveforms, spectrograms, and power spectral densities,\u0000geodetic-based ice-shelf flow, and meteorological variables such as atmospheric\u0000temperature and pressure. The dashboard visualization platform abstracts away\u0000the time-intensive analysis process of raw data and allows scientists to better\u0000concentrate on RIS science.","PeriodicalId":501166,"journal":{"name":"arXiv - PHYS - Atmospheric and Oceanic Physics","volume":"98 1","pages":""},"PeriodicalIF":0.0,"publicationDate":"2024-08-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142215446","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：