Sensitivity analyses can determine how robust model outcomes are to uncertainty in parameter estimates. The most basic sensitivity analyses hold all but one parameter in a model constant while quantifying the effects of altering a single parameter. However, if there is uncertainty in other model parameters then it is unclear at what constant values these parameters should be held while a single value is varied; this decision can have substantial effects on the outcome of the analysis. Also, in biological systems, complex interactions among factors mean that the effect of changing two or more variables can have a greater effect than the total effect of varying those factors independently. We used a sensitivity analysis approach that addresses these two issues for a mechanistic patch-based population model of Fremont cottonwood (Populus fremontii) that we developed for the Sacramento River Valley, CA. Fremont cottonwood is an important component of semi-arid riparian ecosystems throughout western North America, but its populations are in decline due to flow regulation and land conversion. Balancing human resource needs and riparian ecosystem function requires the efficient prioritization of research and policy efforts which can be aided by sensitivity analyses and a mechanistic understanding of the system.

Results/Conclusions

To identify the physical, biological and climatic factors that have the greatest effect on cottonwood populations, we ran simulations with 19,683 possible combinations of parameter estimates reflecting the ranges of uncertainty around each parameter in the model. Sensitivity was quantified by the amount of variance in patch occupancy explained by each factor and all interactions. Consistent with other managed river systems, our analyses show that Freemont cottonwood populations are highly sensitive to flow regime which affects habitat patch creation and seedling recruitment and survival. While flow regime is primarily determined by the volume of water in the system, an important policy concern, we also found that the model is sensitive to spatial variation in channel morphology. Estimates of this parameter could be improved with additional research. We also determined that a mechanistic understanding of competition between Fremont cottonwood and grasses and shrub willows would substantially improve model accuracy and that minimizing colonization by invasive competitors could be an important policy goal. Our sensitivity analyses also suggest that models of future population scenarios should incorporate regional climate change projections because changes in temperature and precipitation affect sensitive aspects of the system, including the timing of seed release and spring snowmelt runoff.