Identifying Sources of Landscape Variation to Improve Predictions of Post-Fire Sagebrush Steppe Recovery

Sagebrush steppe ecosystems are endangered landscapes, threated by the annual grass-fire cycle where invasion by annual grasses drives larger fires and larger fires drive invasion. Despite extensive input of resources by land management agencies, restoration of these ecosystems is notoriously variable and difficult to predict. Understanding and accounting for variation is key to effectively allocating limited resources and having success in restoring burned sagebrush landscapes. I utilized Bayesian modeling to assess how variation in weather, seed dispersal, and topography/slope/landscape position affects understanding of post-fire sagebrush-steppe recovery and how we can best incorporate sources of variation into models predicting where plant communities will most successfully recover. We first asked how weather conditions directly after fire (in the first 4 years) during important phenological windows or during the antecedent five-years affected long-term vegetation trajectories and how inclusion of weather metrics affected the transferability of vegetation abundance models from one site to another. We found that annual grasses, perennial grasses, and sagebrush all responded differently to post-fire weather, with grasses more limited by post-fire precipitation and sagebrush more limited by post-fire temperatures. However, while including weather variables improved model transferability from one site to another for perennial and annual grass abundance (not for sagebrush), the chosen weather metrics did not matter. Next, we aimed to assess how sagebrush seed dispersal varies across large landscapes, such as megafires. We conducted a vertical seed trapping experiment and terminal velocity measurements in the lab and combined the data to parameterize a hierarchical Bayesian model that incorporated both an empirical and mechanistic component. We determined that seed dispersal is highly variable, even at a small scale. Our seed rain projections suggest that seed dispersal from natural recovery may pose severe seed limitations for large burned areas, although natural dispersal is likely to be extremely variable. Our novel data fusion approach to seed dispersal modeling has generalizable applications to estimating seed dispersal at larger scales for other species of concern. Finally, we asked how accuracy and precision of fractional vegetation cover estimates derived from several different satellite-derived products varied with plant cover type, scale, time, and topography in post-fire systems. We found that all gridded map products tested tended to overestimate very low cover and underestimate very high cover, although some products are more accurate than others. We also found that field-derived models of vegetation tend to agree more with satellite-derived models of vegetation at larger scale and less at a smaller scale. Finally, we found that annual herbaceous cover tends to be overestimated in higher elevation, more topographically diverse areas, whereas perennial herbaceous cover tends to be underestimated in these areas. Together these analyses provide a means by which to better understand variability and the reliability of post-fire vegetation recovery models. Incorporation of the sources of variability we have identified here will help refine future models of recovery, whether they are based on data sources from the field, lab, or remote-sensing.