Air Quality Arena: A Benchmark for Forecasting Air Pollution with Foundation Models

In July 2026, an international team of researchers introduced Air Quality Arena (AQA), a large open dataset and benchmark for air pollution forecasting. It covers 6 major pollutants, 7 countries on 4 continents, and more than 14,000 station-pollutant time series. The work is published as a preprint on arXiv, and the data is openly available.

Why a new benchmark was needed

Existing benchmarks for air quality forecasting are too narrow, both geographically and in the pollutants they cover, and they barely test modern time-series foundation models (TSFM) on real, large-scale data. Air Quality Arena fills that gap. Machine learning is increasingly used to forecast pollution levels, but without a single broad standard it is hard to compare models.

The problem is not merely academic. According to the authors' estimate, air pollution causes about 7.9 million premature deaths every year, which makes accurate forecasting a public health priority.

"Air pollution is estimated to cause 7.9 million premature deaths

annually, making accurate forecasting a critical public health priority," the arXiv preprint states.

What the dataset contains

Air Quality Arena has two parts: the data (AQA-Data) and the benchmark (AQA-Bench). The dataset compiles ground-based measurements of 6 major pollutants over a three-year period in 7 countries on 4 continents, for a total of more than 14,000 station-pollutant series. That is considerably broader than most earlier datasets, which were limited to a single region or to one or two substances.

• 6 major air pollutants
• 3-year observation period
• 7 countries on 4 continents
• More than 14,000 station-pollutant series
• 11 foundation models tested, plus classical baseline methods

AQA-Bench evaluates models on short-term forecasting, that is, predicting pollutant levels for the near future. The authors stress the diversity of countries in the sample, which makes it possible to test how universally a model performs across different conditions.

Why foundation models won

Time-series foundation models beat classical methods because they deliver accurate forecasts in zero-shot mode, without fine-tuning for a specific station or region. The authors ran 11 leading TSFMs and classical baseline models through AQA-Bench on the short-term pollution forecasting task. According to the study, the foundation models were consistently more accurate than the classical methods across all regions.

Zero-shot capability matters in practice: a monitoring system can be deployed in a new city that does not yet have a long observation history to train on.

The best result came from a cross-modal architecture that uses a vision foundation model, a model originally trained on images, to forecast time series. In other words, a "visual" neural network transfers what it learned from recognizing pictures to numerical measurement series.

"Time-series foundation models are effective zero-shot forecasters and

consistently outperform classical baseline methods," the authors note in the arXiv preprint.

What this means

Air quality forecasting is shifting toward general-purpose foundation models: a single pretrained model can be applied across different countries without local tuning. An open dataset of 14,000+ series gives researchers a common reference point and speeds up the development of early-warning systems for smog and pollution.

Frequently asked questions

What is Air Quality Arena?

It is an open dataset (AQA-Data) and benchmark (AQA-Bench) for air pollution forecasting: 6 pollutants, 3 years of observations, 7 countries on 4 continents, and more than 14,000 station-pollutant series.

Which model performed best?

The best performer was a cross-modal architecture that uses a vision foundation model to forecast time series. Overall, all the tested time-series foundation models outperformed classical methods in zero-shot mode.

How many models were tested in the benchmark?

The authors compared 11 leading time-series foundation models and several classical baseline methods on the short-term air quality forecasting task.