Grassland fire dynamics are subject to myriad climatic, biological, and anthropogenic drivers, thresholds, and feedbacks and therefore do not conform to assumptions of statistical stationarity. The presence of non-stationarity in time series data leads to ambiguous results that can misinform regional-level fire management strategies. This study employs non-stationarity in time series data among multiple variables and multiple intensities using dynamic simulations of autoregressive distributed lag models to elucidate key drivers of climate and ecological change on burned grasslands in Xilingol, China. We used unit root methods to select appropriate estimation methods for further analysis. Using the model estimations, we developed scenarios emulating the effects of instantaneous changes (i.e., shocks) of some significant variables on climate and ecological change. Changes in mean monthly wind speed and maximum temperature produce complex responses on area burned, directly, and through feedback relationships. Our framework addresses interactions among multiple drivers to explain fire and ecosystem responses in grasslands, and how these may be understood and prioritized in different empirical contexts needed to formulate effective fire management policies.