Cost-effective ML approaches for atmospheric science

🌍 Climate scientists often face a trade-off: Global Climate Models (GCMs) are essential for long-term climate projections — but they operate at coarse spatial resolution, making them too crude for regional or local decision-making. To get fine-scale data, researchers use Regional Climate Models (RCMs). These add crucial spatial detail, but come at a very high computational cost, often requiring supercomputers to run for months. ➡️ A new paper introduces EnScale — a machine learning framework that offers an efficient and accurate alternative to running full RCM simulations. Instead of solving the complex physics from scratch, EnScale "learns" the relationship between GCMs and RCMs by training on existing paired datasets. It then generates high-resolution, realistic, and diverse regional climate fields directly from GCM inputs. What makes EnScale stand out? ✅ It uses a generative ML model trained with a statistically principled loss (energy score), enabling probabilistic outputs that reflect natural variability and uncertainty ✅ It is multivariate – it learns to generate temperature, precipitation, radiation, and wind jointly, preserving spatial and cross-variable coherence ✅ It is computationally lightweight – training and inference are up to 10–20× faster than state-of-the-art generative approaches ✅ It includes an extension (EnScale-t) for generating temporally consistent time series – a must for studying events like heatwaves or prolonged droughts This approach opens the door to faster, more flexible generation of regional climate scenarios, essential for risk assessment, infrastructure planning, and climate adaptation — especially where computational resources are limited. 📄 Read the full paper: EnScale: Temporally-consistent multivariate generative downscaling via proper scoring rules ---> https://lnkd.in/dQr5rmWU (code: https://lnkd.in/dQk_Jv8g) 👏 Congrats to the authors — a strong step forward for ML-based climate modeling! #climateAI #downscaling #generativeAI #machinelearning #climatescience #EnScale #RCM #GCM #ETHZurich #climatescenarios

Coupled chemistry-climate models (CCMs) are essential tools for understanding chemical variability in the climate system, but they are extraordinarily expensive to run. Eric Mei's recent paper shows that linear inverse models (LIMs) can be used to emulate CCMs at a fraction of the computational cost (laptop vs HPC). This opens up new opportunities for strongly-coupled chemistry-climate data assimilation, large ensembles, hypothesis testing, and cost/benefit analysis for nonlinear machine learning emulators of CCMs. In constrast to ML emulators, LIMs have transparent explainability, illustrated by the figure below showing the coupled time-evolving relationship between sea-surface temperature, ozone, and hydroxyl radical for the El Nino mode in the model. Link to the paper: https://lnkd.in/d4DcJHVZ Supported by Schmidt Futures

An end-to-end machine learning approach for weather prediction Weather forecasting is often guided by high-dimensional physics-based simulations that merge raw measurements with large-scale fluid dynamics models. While these numerical weather prediction pipelines have substantially improved over time, they rely on multi-stage procedures that can demand bespoke supercomputing resources. Scientists in atmospheric science, environmental risk, and related fields are increasingly investigating data-driven solutions as a way to reduce complexity, speed up predictions, and possibly discover new weather features not readily captured by traditional methods. Vaughan et al. present a machine learning system that replaces the entire weather forecasting pipeline, from raw observational data to final predictions. The authors’ model, termed Aardvark Weather, is built around an encoder–processor–decoder design that ingests satellite and in-situ measurements, then transforms them into global and local predictions. The encoder uses set-convolution layers and a vision-transformer backbone to handle sparse, off-grid input data. It provides an initial atmospheric state that is refined by an autoregressive processor, also powered by a vision transformer, which predicts future meteorological conditions at several lead times. Finally, a specialized decoder produces station-level forecasts for temperature and wind speed. During training, the system is pre-trained on reanalysis data to build robust representations, then fine-tuned end-to-end on smaller volumes of real-world observations. The authors’ approach demonstrates high forecast accuracy, with skillful predictions extending to nine or ten days for multiple variables. Their experiments indicate that this end-to-end system yields root mean squared errors competitive with or lower than those of established global numerical models on several benchmarks. The computational demands are also reported to be orders of magnitude less than standard pipelines, revealing the promise of a fully learnable framework. The researchers further show how fine-tuning can target specific variables or regions of interest, enabling customized yet efficient solutions. These findings open new possibilities in weather prediction, especially for domains or locations where conventional supercomputing-based approaches are hard to implement. Paper: https://lnkd.in/dZbVgC32 #weatherprediction #machinelearning #deeplearning #visiontransformers #climate #environmentalresearch #atmosphericscience #meteorology #computationalmethods #research #innovation #datadriven #earthobservation #AIforscience #scientificcomputing