Fire-scarred trees provide a deep temporal record of historical fire activity, but identifying the mechanisms therein that controlled landscape fire patterns is not straightforward. We use a spatially correlated metric for fire co-occurrence between pairs of trees (the Sørensen distance variogram), with output from a neutral model for fire history, to infer the relative strength of top-down vs. bottom-up controls on historical fire regimes. An inverse modeling procedure finds combinations of neutral-model parameters that produce Sørensen distance variograms with statistical properties similar to those observed from two landscapes in eastern Washington, USA, with contrasting topography. We find the most parsimonious model structure that is able to replicate the observed patterns and the parameters of this model provide surrogates for the predominance of top-down vs. bottom-up controls. Simulations with relatively low spread probability produce irregular fire perimeters and variograms similar to those from the topographically complex landscape. With higher spread probabilities fires exhibit regular perimeters and variograms similar to those from the simpler landscape. We demonstrate that cross-scale properties of the fire-scar record, even without historical fuels and weather data, document how complex topography creates strong bottom-up controls on fire spread. This control is weaker in simpler topography, and may be compromised in a future climate with more severe weather events.