Inferring Thunderstorm Occurrence from Vertical Profiles of Convection-Permitting Simulations: Physical Insights from a Physical Deep Learning Model
Uloženo v:
| Název: | Inferring Thunderstorm Occurrence from Vertical Profiles of Convection-Permitting Simulations: Physical Insights from a Physical Deep Learning Model |
|---|---|
| Autoři: | Vahid Yousefnia, Kianusch, Metzl, Christoph, Bölle, Tobias |
| Zdroj: | Artificial Intelligence for the Earth Systems. 4 |
| Publication Status: | Preprint |
| Informace o vydavateli: | American Meteorological Society, 2025. |
| Rok vydání: | 2025 |
| Témata: | FOS: Computer and information sciences, Numerical weather prediction, FOS: Physical sciences, Thunderstorms, Machine Learning (cs.LG), Machine Learning, Atmospheric and Oceanic Physics, Postprocessing, Machine learning, Atmospheric and Oceanic Physics (physics.ao-ph), Interpretability, I.2.6, J.2, Deep convection, Forecasting |
| Popis: | Thunderstorms have significant social and economic impacts due to heavy precipitation, hail, lightning, and strong winds, necessitating reliable forecasts. Thunderstorm forecasts based on numerical weather prediction (NWP) often rely on single-level surrogate predictors, like convective available potential energy and convective inhibition, derived from vertical profiles of three-dimensional atmospheric variables. In this study, we develop Signature-Based Approach of Identifying Lightning Activity using Machine Learning (SALAMA) 1D, a deep neural network which directly infers the probability of thunderstorm occurrence from vertical profiles of 10 atmospheric variables, bypassing single-level predictors. By training the model on convection-permitting NWP forecasts, we allow SALAMA 1D to flexibly identify convective patterns, with the goal of enhancing forecast accuracy. The model’s architecture is physically motivated: Sparse connections encourage interactions at similar height levels while keeping model size and inference times computationally efficient, whereas a shuffling mechanism prevents the model from learning nonphysical patterns tied to the vertical grid. SALAMA 1D is trained over central Europe with lightning observations as the ground truth. Comparative analysis against a baseline machine learning model that uses single-level predictors shows SALAMA 1D’s superior skill across various metrics and lead times of up to at least 11 h. Moreover, expanding the archive of forecasts from which training examples are sampled improves skill, even when the training set size remains constant. Finally, a sensitivity analysis using saliency maps indicates that our model relies on physically interpretable patterns consistent with established theoretical understanding, such as ice particle content near the tropopause, cloud cover, conditional instability, and low-level moisture. Significance Statement This work aims to improve thunderstorm forecasting by applying machine learning to vertical atmospheric profiles from numerical weather prediction. We developed a model that incorporates physical considerations, resulting in more accurate yet computationally efficient predictions compared to conventional methods. Additionally, the model provides insights into how it identifies thunderstorm occurrence, fostering interpretability and trust. Our research demonstrates how to enhance the skill of machine learning systems in severe weather forecasting, which is crucial for supporting timely, informed decision-making in situations that impact public safety and the economy. |
| Druh dokumentu: | Article |
| ISSN: | 2769-7525 |
| DOI: | 10.1175/aies-d-24-0096.1 |
| DOI: | 10.48550/arxiv.2409.20087 |
| Přístupová URL adresa: | http://arxiv.org/abs/2409.20087 https://elib.dlr.de/216616/ https://doi.org/10.1175/aies-d-24-0096.1 |
| Rights: | CC BY URL: http://www.ametsoc.org/PUBSReuseLicenses |
| Přístupové číslo: | edsair.doi.dedup.....88cdb7768881d1f9e57adcee2cb197d3 |
| Databáze: | OpenAIRE |
Nájsť tento článok vo Web of Science | Abstrakt: | Thunderstorms have significant social and economic impacts due to heavy precipitation, hail, lightning, and strong winds, necessitating reliable forecasts. Thunderstorm forecasts based on numerical weather prediction (NWP) often rely on single-level surrogate predictors, like convective available potential energy and convective inhibition, derived from vertical profiles of three-dimensional atmospheric variables. In this study, we develop Signature-Based Approach of Identifying Lightning Activity using Machine Learning (SALAMA) 1D, a deep neural network which directly infers the probability of thunderstorm occurrence from vertical profiles of 10 atmospheric variables, bypassing single-level predictors. By training the model on convection-permitting NWP forecasts, we allow SALAMA 1D to flexibly identify convective patterns, with the goal of enhancing forecast accuracy. The model’s architecture is physically motivated: Sparse connections encourage interactions at similar height levels while keeping model size and inference times computationally efficient, whereas a shuffling mechanism prevents the model from learning nonphysical patterns tied to the vertical grid. SALAMA 1D is trained over central Europe with lightning observations as the ground truth. Comparative analysis against a baseline machine learning model that uses single-level predictors shows SALAMA 1D’s superior skill across various metrics and lead times of up to at least 11 h. Moreover, expanding the archive of forecasts from which training examples are sampled improves skill, even when the training set size remains constant. Finally, a sensitivity analysis using saliency maps indicates that our model relies on physically interpretable patterns consistent with established theoretical understanding, such as ice particle content near the tropopause, cloud cover, conditional instability, and low-level moisture. Significance Statement This work aims to improve thunderstorm forecasting by applying machine learning to vertical atmospheric profiles from numerical weather prediction. We developed a model that incorporates physical considerations, resulting in more accurate yet computationally efficient predictions compared to conventional methods. Additionally, the model provides insights into how it identifies thunderstorm occurrence, fostering interpretability and trust. Our research demonstrates how to enhance the skill of machine learning systems in severe weather forecasting, which is crucial for supporting timely, informed decision-making in situations that impact public safety and the economy. |
|---|---|
| ISSN: | 27697525 |
| DOI: | 10.1175/aies-d-24-0096.1 |