Evaluating a sampling design: What site specific variability do you lose when stratifying with a national map?
The National Ecological Observatory Network’s (NEON) goal is to improve understanding and forecasting of ecological change over decades at continental scales (Schimel et al. 2011). The Terrestrial Observation System (TOS) in particular will quantify the impacts of change on terrestrial organisms and processes. Choosing where within a site to sample is important to help determine the types of questions that can be answered. Standardization across all systems and targeting dominant land cover types are key to NEON’s ability to provide comparable data from all locations. The twenty domains are clustered in regards to climatic and vegetation indices; the sites were chosen to represent the dominant land cover types of the domains; and the tower locations were selected to measure in areas of prevailing vegetation. Following these guidelines, the TOS sampling design stratified each site by the National Land Cover Database (NLCD). This broad level classification ensures that data collection is targeted at the dominant land cover types to increase our ability to detect changes over time by decreasing variability. However, at sites where previous research has produced finer scale maps,we must ask, what do we miss when using this coarse stratification?
At each NEON site up to thirty vegetation sampling plots were allocated based on the proportions of the dominant NLCD land cover types. We have evaulated how allocating plots based on NLCD stratification captures the site specific variability at locations where external research projects have created finer scale vegetation maps, geological maps, and management maps. An initial round of analysis involved five sites that span the diversity of NEON sampling: Toolik Field Station in Alaska, Yellowstone National Park in Montana, Harvard Forest in Massachusetts, Bartlett Experimental Forest in New Hampshire, and Ordway-Swisher Biological Station in Florida. Differences in the site specific map proportions and the proportion of NEON-allocated plots within those fractions, ranged from 1-30% with an average of 6% difference. Furthermore, the larger differences between the two allocation methods occurred in vegetation types where we already knew we had logistical constraints (steep slopes, seasonal flooding) for plot placement. Overall, the ability to capture a site’s variability with random plot selection from a coarse stratification indicates that standardized sampling across all sites does not limit NEON’s capacity to understand change at the site level as well.