Measuring the Impact of Explainable AI (XAI) Method on Decision-Makers' Trust and Response Decision Speed in Wildfire Spread Prediction Models

Artificial intelligence-based wildfire spread prediction models have emerged as critical decision support tools in wildfire management. However, the black-box nature of these models limits the trust that decision-makers place in model outputs and may lead to delays in response decisions. This study aims to measure how the integration of explainable artificial intelligence (XAI) methods into wildfire spread prediction models affects decision-makers' trust levels in model outputs and response decision speed. In this research, SHAP (SHapley Additive exPlanations), LIME (Local Interpretable Model-agnostic Explanations), and Grad-CAM (Gradient-weighted Class Activation Mapping) methods were applied to random forest, deep neural network, and convolutional neural network-based wildfire spread models. Within the experimental design framework, a controlled experiment was conducted with 48 emergency managers and wildfire response specialists; participants' responses to model outputs with and without XAI explanations were assessed through Likert scale trust scores and decision time measurements. The findings reveal that SHAP-based explanations increased trust levels by an average of 27.4% and reduced response decision time by 18.6%. Additionally, it was determined that explanation complexity beyond a certain threshold increased cognitive load and led to a decrease in trust levels. The study provides design principles and implementation recommendations for integrating XAI methods into wildfire management decision support systems.