AirCast: Improving Air Pollution Forecasting Through Multi-Variable Data Alignment

TL;DR

AirCast addresses the challenge of forecasting air pollution by integrating weather and air-quality data through a Vision Transformer with a multi-task head and a frequency-weighted MAE loss to handle heavy-tailed pollutant distributions. It introduces a spatially and temporally aligned dataset for the MENA region and employs variable tokenization with cross-variable aggregation to efficiently fuse multi-modal inputs. The approach yields improved PM forecasts (PM2.5, PM10, PM1) over baselines and demonstrates capability to anticipate extreme events, with near-surface variables driving most gains. The work contributes an open-source framework and dataset to advance multi-variable air pollution forecasting and regional forecasting capabilities.

Abstract

Air pollution remains a leading global health risk, exacerbated by rapid industrialization and urbanization, contributing significantly to morbidity and mortality rates. In this paper, we introduce AirCast, a novel multi-variable air pollution forecasting model, by combining weather and air quality variables. AirCast employs a multi-task head architecture that simultaneously forecasts atmospheric conditions and pollutant concentrations, improving its understanding of how weather patterns affect air quality. Predicting extreme pollution events is challenging due to their rare occurrence in historic data, resulting in a heavy-tailed distribution of pollution levels. To address this, we propose a novel Frequency-weighted Mean Absolute Error (fMAE) loss, adapted from the class-balanced loss for regression tasks. Informed from domain knowledge, we investigate the selection of key variables known to influence pollution levels. Additionally, we align existing weather and chemical datasets across spatial and temporal dimensions. AirCast's integrated approach, combining multi-task learning, frequency weighted loss and domain informed variable selection, enables more accurate pollution forecasts. Our source code and models are made public here (https://github.com/vishalned/AirCast.git)

Figures (6)

• Figure 1: This illustrates the architecture of the AirCast model, an extension of nguyen2023climaxfoundationmodelweather. The model integrates weather data from the ERA5 dataset and air quality data from the CAMS EAC4 dataset. The model is trained using regional data from the MENA region. The input variables are tokenized and aggregated, with a Vision Transformer (ViT) encoder, processing the combined weather and air quality inputs. A dual decoder head is employed, with one predicting weather variables and the other forecasting air quality variables. The predictions are compared with the ground truth at a certain lead time using the Frequency-Weighted MAE loss function.
• Figure 2: Skewed distribution of PM2.5. The y-axis corresponds to the frequency clipped at 200 (the maximum frequency is shown in each figure).
• Figure 3: Sample error plots for PM2.5 forecasting (prediction - ground truth). The unit is $kg m^{-3}$. The first and second plots are without and with the proposed fMAE loss function respectively.
• Figure 4: Extreme case visualizations of PM2.5 concentrations (Predictions - Ground Truth) for CAMS global forecasts and Aircast.
• Figure 5: Extreme case visualizations of PM2.5 concentrations (Predictions - Ground Truth) for Aircast and the CAMS global forecasts