BayFuse-DSP: A Bayesian fusion framework for deployment scale prediction in aerial forest firefighting

Rapid and reliable estimation of aerial firefighting deployment scale is essential for improving forest fire response and reducing fire impacts, particularly when only limited fire-scene information is available during the early stage of a forest fire. However, existing approaches remain limited: expert-based methods are often subjective, while data-driven methods typically depend heavily on historical data and offer limited interpretability. To address these limitations, this study proposes BayFuse-DSP, a Bayesian fusion framework for deployment scale prediction in aerial forest firefighting scenarios. In this framework, a fuzzy expert system converts expert knowledge into prior distributions, while a random forest model learns the relationship between scenario features and deployment scale from historical cases and estimates the likelihood function. The prior and likelihood are then dynamically fused through Bayesian inference to obtain the posterior distribution and the corresponding 95% credible interval. The framework is validated using 44 aerial forest firefighting cases from China. The results show that BayFuse-DSP outperforms standalone knowledge-driven methods, standalone data-driven methods, and external probabilistic regression baselines in terms of predictive accuracy, stability, and interpretability. BayFuse-DSP achieves an MAE of 0.891, an RMSE of 1.118, an R² of 0.707, and a 95% credible interval coverage rate of 97.7%. These findings indicate that BayFuse-DSP provides interpretable and uncertainty-aware decision support for early-stage aerial firefighting deployment, with potential applicability to disaster response planning and risk reduction in other emergency scenarios.