# PlumeCityNet: Multi-Resolution Air Quality Forecasting

arXiv: https://arxiv.org/abs/2110.02661 · HTML: https://mlsanity.com/paper/2110.02661/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 9 of its 32 references that are in the map.
- First posted on arXiv 6 Oct 2021.

## Builds on (most cited first, 9 of 9)

- [U-Net: Convolutional Networks for Biomedical Image Segmentation](https://mlsanity.com/paper/1505.04597/) (2015)
- [Convolutional LSTM Network: A Machine Learning Approach for Precipitation Nowcasting](https://mlsanity.com/paper/1506.04214/) (2015)
- [MetNet: A Neural Weather Model for Precipitation Forecasting](https://mlsanity.com/paper/2003.12140/) (2020)
- [Machine Learning for Precipitation Nowcasting from Radar Images](https://mlsanity.com/paper/1912.12132/) (2019)
- [Deep Air Quality Forecasting Using Hybrid Deep Learning Framework](https://mlsanity.com/paper/1812.04783/) (2018)
- [Deep Air Learning: Interpolation, Prediction, and Feature Analysis of Fine-grained Air Quality](https://mlsanity.com/paper/1711.00939/) (2017)
- [HighAir: A Hierarchical Graph Neural Network-Based Air Quality Forecasting Method](https://mlsanity.com/paper/2101.04264/) (2021)
- [PlumeNet: Large-Scale Air Quality Forecasting Using A Convolutional LSTM Network](https://mlsanity.com/paper/2006.09204/) (2020)
- [DeepPlume: Very High Resolution Real-Time Air Quality Mapping](https://mlsanity.com/paper/2002.10394/) (2020)

## Abstract (arXiv)

This paper presents an engine able to forecast jointly the concentrations of the main pollutants harming people's health: nitrogen dioxide (NO2), ozone (O3) and particulate matter (PM2.5 and PM10, which are respectively the particles whose diameters are below 2.5um and 10um respectively). The engine is fed with air quality monitoring stations' measurements, weather forecasts, physical models' outputs and traffic estimates to produce forecasts up to 24 hours. The forecasts are produced with several spatial resolutions, from a few dozens of meters to dozens of kilometers, fitting several use-cases needing air quality data. We introduce the Scale-Unit block, which enables to integrate seamlessly all available inputs at a given resolution to return forecasts at the same resolution. Then, the engine is based on a U-Net architecture built with several of those blocks, giving it the ability to process inputs and to output predictions at different resolutions. We have implemented and evaluated the engine on the largest cities in Europe and the United States, and it clearly outperforms other prediction methods. In particular, the out-of-sample accuracy remains high, meaning that the engine can be used in cities which are not included in the training dataset. A valuable advantage of the engine is that it does not need much computing power: the forecasts can be built in a few minutes on a standard CPU. Thus, they can be updated very frequently, as soon as new air quality monitoring stations' measurements are available (generally every hour), which is not the case of physical models traditionally used for air quality forecasting.