Cannot see the random forests for the decision trees: selecting predictive models for restoration ecology

Authors: Barnard, David; GERMINO, M - Us Geological Survey (USGS); PILLIOD, D - Us Geological Survey (USGS); ARKLE, R - Us Geological Survey (USGS); APPLESTEIN, C - Us Geological Survey (USGS); DAVIDSON, B - Us Geological Survey (USGS); FISK, M - Us Geological Survey (USGS)

Submitted to: Restoration Ecology
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 4/19/2019
Publication Date: 5/28/2019
Citation: Barnard, D.M., Germino, M.J., Pilliod, D.S., Arkle, R.S., Applestein, C.V., Davidson, B.E., Fisk, M.R. 2019. Cannot see the random forests for the decision trees: selecting predictive models for restoration ecology. Restoration Ecology. 10(6):1-11. https://doi.org/10.1111/rec.12938.
DOI: https://doi.org/10.1111/rec.12938

Interpretive Summary: The ability to predict where ecosystem restoration treatments will be the most successful across varied landscapes is a priority for resource managers. Developing decision-support tools from field-collected and remotely-sensed datasets requires selecting a statistical approach, but there is limited guidance available to end-users to select from the multitude of approaches available. In this study, we use high resolution field sampling data of sagebrush reestablishment after wildfire, combined with remote sensing datasets, to test eleven different statistical modeling frameworks for greatest predictive accuracy. We tested machine learning, maximum likelihood, Bayesian, and ensemble approaches, and found predictive power to vary substantially among models. Moreover, model accuracy improved as the number of predictor variables included in the models decreased, suggesting that a few key landscape variables are responsible for the majority of variability in sagebrush recovery. The process of model comparison and refinement highlighted in this paper will aid future resource managers by guiding them through the process of model-selection to support management decisions.

Technical Abstract: Improving predictions of restoration outcomes is increasingly important to resource managers for accountability and adaptive management, yet there is limited guidance for selecting a predictive model from the multitude available. The goal of this paper was to identify an optimal predictive framework for restoration ecology using eleven modeling frameworks (including, machine learning, inferential, and ensemble approaches), and three data groups (field data, geographic data [GIS], and a combination thereof). We test this approach with a dataset from a large post-fire sagebrush reestablishment project in the Great Basin, USA. Predictive power varied among models and data groups, ranging from 58-79% accuracy. Finer scale field data generally had the greatest predictive power, although GIS data were present in the best models overall. An ensemble prediction computed from the ten models parameterized to field data was well above average for accuracy but was outperformed by others that prioritized model parsimony by selecting predictor variables based on rankings of their importance among all candidate models. The variation in predictive power among a suite of modeling frameworks underscores the importance of a model comparison and refinement approach that evaluates multiple models and data groups, and selects variables based on their contribution to predictive power. The enhanced understanding of factors influencing restoration outcomes accomplished by this framework has the potential to aid the adaptive management process for improving future restoration outcomes.