AtmoBench: A Large-Scale Multi-Region Benchmark for Air Quality Forecasting with Time Series Foundation Models

Rishi Bharadwaj, Manik Gupta, Pandarasamy Arjunan

Keywords: air quality forecasting, time series foundation models, benchmark, dataset, climate

TL;DR: We present AtmoBench, a large-scale multi-region multi-pollutant dataset and benchmark, spanning 7 countries and 4 continents, with more than 14,000 station-pollutant series.

Abstract:

Air pollution causes an estimated 7.9 million premature deaths annually, making accurate forecasting a critical public health priority. Machine learning is increasingly being applied to forecast air pollution levels, yet existing benchmarks remain narrow in both geographic scope and pollutant coverage, and fail to evaluate the latest generation of time series foundation models (TSFMs) on real world, large scale data. We present AtmoBench, a large scale multi-country and multi-pollutant dataset and benchmark to address this gap.
AtmoBench covers 6 major pollutants over a three year period across 7 diverse
countries and 4 continents, with more than 14,000 station-pollutant series, aiming to provide a comprehensive benchmark for air quality
tasks. We benchmark this dataset across 11 leading time series foundation models
and classical baselines to assess performance on short-term air quality
forecasting. Our results demonstrate that TSFMs are effective zero-shot forecasters and consistently outperform classical baselines, with our top-performing model employing a cross-modal architecture that leverages a vision foundation model for time series forecasting.

Submission Number: 103