A new tool for power analysis of fixed plot data: Using simulations and mixed effects models to evaluate forest metrics

Since 2006, the National Park Service's Northeast Temperate Network (NETN) has been monitoring forest health in 10 national park units in the northeastern U.S. using a protocol adapted from the U.S. Forest Service Forest Inventory and Analysis Program. To ensure current methods are appropriate for monitoring long-term trends in forest composition, structure and function, we performed a power analysis of key forest metrics using data collected in each park and covering two four-year survey cycles. We determined statistical power by repeatedly generating bootstrapped datasets with specified percent change between survey cycles in the value of each metric, and then testing whether a mixed effects model detected a significant change. We applied effect sizes ranging from a 50% decline to a 50% increase in 5% increments. Power analyses indicated that, for most key forest metrics, our monitoring program met the target of detecting a 40% change in a metric over a 12-year period with 80% power and a Type I error rate of 0.10. Power also tended to improve with subsequent repeated surveys. Native species richness and live-tree basal area metrics consistently performed well for all parks. Average percent cover of plant groups performed better than quadrat frequency. Regeneration metrics performed best in parks with low or high regeneration rates. Coarse woody debris volume, snag abundance, and invasive species richness did not meet the trend detection target in multiple parks. In most cases, metrics that failed to meet the trend detection target in one or several parks had high proportions of zeros and relatively low overall values for the respective park. In cases where high metric variability was the reason for poor trend detection, results indicated that post-stratifying can sometimes improve power. We developed the power analysis tool in R to be applicable for a range of data types, including proportional and count data, and for any number of sampling areas (e.g., parks) and sampling units (e.g., plots). Our approach represents one of the few tools available that can assess the power to detect change over time using mixed effects models.