Simulating the Air Quality Impact of Prescribed Fires Using Graph Neural Network-Based PM$_{2.5}$ Forecasts

TL;DR

This paper addresses forecasting $PM_{2.5}$ in California at hourly resolution under wildfire and prescribed-fire scenarios by developing a spatio-temporal Graph Neural Network (GNN) that performs robust predictions with sparse sensor data. A two-step approach separates ambient versus fire-derived pollution, enabling fire-specific PM$_{2.5}$ forecasts, and a novel prescribed-fire simulation framework integrates transposed FRP emissions with GNN forecasts. The GNN outperforms LSTM and MLP baselines, and the two prescribed-fire experiments demonstrate both timing optimization (finding March as favorable) and a counterfactual trade-off showing that prescribed burns can modestly raise short-term PM$_{2.5}$ but substantially reduce longer-term wildfire-induced pollution. The framework provides actionable insights for land managers and public health planning, offering a computationally efficient alternative to chemistry-transport models for exploring air-quality impacts of prescribed fires. Future work will incorporate physics-based elements to further refine the fire-physics component within the GNN.

Abstract

The increasing size and severity of wildfires across the western United States have generated dangerous levels of PM$_{2.5}$ concentrations in recent years. In a changing climate, expanding the use of prescribed fires is widely considered to be the most robust fire mitigation strategy. However, reliably forecasting the potential air quality impact from prescribed fires, which is critical in planning the prescribed fires' location and time, at hourly to daily time scales remains a challenging problem. In this paper, we introduce a spatial-temporal graph neural network (GNN) based forecasting model for hourly PM$_{2.5}$ predictions across California. Using a two-step approach, we leverage our forecasting model to estimate the PM$_{2.5}$ contribution of wildfires. Integrating the GNN-based PM$_{2.5}$ forecasting model with prescribed fire simulations, we propose a novel framework to forecast the PM$_{2.5}$ pollution of prescribed fires. This framework helps determine March as the optimal month for implementing prescribed fires in California and quantifies the potential air quality trade-offs involved in conducting more prescribed fires outside the fire season.

Figures (9)

• Figure 1: FRP aggregated around a given radius for each PM$_{2.5}$ monitor location using wind and distance information
• Figure 2: Graph neural network (GNN) used in our PM$_{2.5}$ forecasting model considers PM$_{2.5}$ monitors as nodes in the graph and produces node-level predictions
• Figure 3: PM$_{2.5}$ predictions one hour into the future from a temporal subset of testing results for example sites. The GNN (column one), LSTM, and MLP all use WIDW FRP variables, while the GNN with IDW FRP (column two) uses IDW FRP variables
• Figure 4: Conceptual diagram of the methodology for distinguishing fire-specific and ambient PM$_{2.5}$ concentrations
• Figure 5: Ambient and fire-specific PM$_{2.5}$ predictions one hour into the future from a temporal subset of testing results for example sites