Short-term spatio-temporal forecasting of human-caused wildland fire occurrence: an errors in variables approach

Background Accurately predicting human-caused wildfires is critical for fire management. However, key predictors often contain measurement errors, which can compromise model performance. Aims We aim to demonstrate a methodology for incorporating uncertainty in short-term fire-weather forecasts into fine-scale, spatially explicit fire occurrence prediction (FOP) models, and show that doing so leads to improvements over an approach where the inherent uncertainty in forecasted fire-weather is ignored. Methods We build spatially explicit models to forecast daily human-caused wildfires 1–4 days ahead across Ontario’s Fire Region. An errors-in-variables approach is incorporated to correct for uncertainty in forecasted fine fuel moisture. Key results The proposed methodology produces short-term forecasts of the number of future fire occurrences along with prediction intervals to quantify uncertainty. Color-coded maps visualize output and highlight areas of concern. Our modelling approach consistently outperforms a comparable modelling approach that does not adjust for measurement error. Models that ignore uncertainty in forecasted fire weather tended to underpredict on average and were less precise. Conclusions Incorporating fire-weather forecasting uncertainty enhances short-term forecasts of human-caused wildfires. Implications This approach supports more reliable fire management planning and resource allocation, particularly under uncertain fire-weather conditions.