For decades, the science of restoring damaged ecosystems has been progressing as our understanding of post-disturbance succession has improved. Many studies of natural and man-made disturbances and the stages of their recovery have been conducted in grassland ecosystems. Incorporating what has been learned from these past studies when planning a restoration project may improve our ability to return a site to a functional state that approximates the pre-disturbance condition. Models that have been developed from past studies are a means to provide important insight into key processes and interactions that are directly applicable to restoration efforts. One of the hurdles that must be overcome is that the models developed to date have been either too complex to be accessible to land managers or they cannot be generalized to locations beyond where the original study was conducted. The application of Bayesian statistical modeling would allow land managers to take advantage of what has been learned from past studies to inform future restoration projects in a more intuitive, flexible, and less complicated form.
Results/Conclusions
To demonstrate how Bayesian analysis can be utilized in a restoration project, a dataset from the restoration of an abandoned farm was used to create a Bayesian Network Model. This dataset was collected over a three year period, from 2006 through 2008, and consists of seeding mixture and estimates of plant cover, soil carbon and nitrogen content, light penetration, average monthly precipitation, and average monthly temperature. The model’s accuracy was validated using two data sets similar to the one used to develop the model, but from different grassland systems. The model was used to predict the post-treatment ratios of relative cover of annual to perennial forbs and annual forbs to perennial grasses. These metrics were chosen because they are an indicator of the progress of regeneration on disturbed grassland systems. The model, when run forward, allows us to predict the outcome of the treatments applied. When run backward, the model allows a specific goal to be set and the treatment that will most likely achieve that goal is reported. This unique bi-directional capability provides a powerful tool for land mangers to evaluate the effectiveness of treatments and to plan their efforts in a way that maximizes the chance of success.
