Fire is a complex process involving interactions and feedbacks between biological, socioeconomic, and physical drivers across multiple spatial and temporal scales. This complexity limits our ability to incorporate fire into Earth system models and project future fire activity under climate change. Conceptual, empirical, and process models have identified the mechanisms and processes driving fire regimes, and provide a useful basis to consider future fire activity. However, these models generally deal with only one component of fire regimes, fire frequency, and do not incorporate feedbacks between fire, vegetation, and climate. They are thus unable to predict the location, severity or timing of fires, the socioecological impacts of fire regime change, or potential non-linear responses such as biome shifts into alternative stable states. Dynamic modeling experiments may facilitate more thorough investigations of fire-vegetation-climate feedbacks and interactions, but their success will depend on the development of dynamic global vegetation models (DGVMs) that more accurately represent biological drivers. This requires improvements in the representation of current vegetation, plant responses to fire, ecological dynamics, and land management to capture the mechanisms behind fire frequency, intensity, and timing. DGVMs with fire modules are promising tools to develop a globally consistent analysis of fire activity, but projecting future fire activity will ultimately require a transdisciplinary synthesis of the biological, atmospheric, and socioeconomic drivers of fire. This is an important goal because fire causes substantial economic disruption and contributes to future climate change through its influence on albedo and the capacity of the biosphere to store carbon.