Lumber supply chain dynamics under wildfires across California

Abstract

The frequency and intensity of wildfires have escalated in recent decades; however, their growing destructiveness has emerged as the most alarming trend. The compounding effects of climate change and expanding human settlements into the wildland-urban interface (WUI) have made recent wildfires significantly more destructive than those in the past. Residential structures situated within the WUI are particularly vulnerable to wildfire damage, and their loss substantially complicates efforts to restore community functionality in the aftermath of wildfires. The majority of wildfire-damaged structures are single-family, wood-framed residences, underscoring the direct link between wildfire losses and added lumber demand during reconstruction. At the same time, wildfires are consuming vast forested areas, resulting in significant timber losses. Despite these interlinked dynamics, there remains a lack of integrated tools capable of quantifying the compounded effects of wildfire destruction on both lumber demand for reconstruction and long-term timber supply from forests under evolving wildfire risk. Developing such predictive capability is essential for anticipating future imbalances between the lumber demand and the forest's capacity to supply timber, thereby informing proactive mitigation strategies, sustainable forest management, and long-term resilience under a changing climate. In this study, a comprehensive modeling framework is developed to quantify wildfire-induced building damages, estimate post-fire rebuilding demand, and evaluate timber supply from forests under wildfire risk. California serves as the case study region, as it accounts for the majority of wildfire-related structural losses in the U.S., with over two-thirds of all destroyed buildings occurring within the state. The framework integrates multiple interconnected models. Two Artificial Neural Networks (ANNs) are developed: a Convolutional Neural Network (CNN) to predict the spatial probability of wildfire burn occurrence, and a Multilayer Perceptron (MLP) to estimate the probability of building damage conditional on burn exposure across regional extents. These two models are combined to estimate building-level wildfire damage by integrating burn and conditional damage probabilities, and the trained models are further applied to future climate scenarios to predict future burn probability and housing losses. On the supply side, a random forest model is developed to estimate the probability of timber harvest, which, together with wildfire and mortality probabilities, is incorporated into a Markov chain framework to simulate long-term timber supply dynamics under disturbance risks. Finally, a decision-making framework is constructed that integrates outputs from all models to assess long-term economic costs and resource sustainability across alternative mitigation and adaptation strategies. The findings revealed key drivers influencing both wildfire burn occurrence and building damage, including the spatial configuration of buildings and the characteristics of surrounding fuels. Projections indicated that burned areas are expected to expand under higher emission scenarios, leading to greater exposure of residential structures to wildfire risk. Simulations of timber supply under wildfire disturbance suggested a decline in yield toward the end of the century, highlighting growing dependence on imported lumber to meet domestic demand. Although mitigation strategies cannot fully offset this reliance, they can substantially reduce overall costs and ease ecological pressure on forests, underscoring their importance in promoting sustainable recovery and long-term resource resilience.