Table and accompanying photographs for biogeomorphic classification of shorebird nesting sites on the U.S. Atlantic coast from March to September, 2016

iPlover users selected attributes for Geomorphic Setting, Substrate Type, Vegetation Type, and Vegetation Density that best characterized each nest or random point location. As such, there may be some variation in the assessment of a category between individual users. The reliability of the user-chosen classification was estimated as described below. The formal accuracy assessment described below was performed for data collected with the same protocols in two prior years (Sturdivant and others, 2016).
The accuracy of the classifications of Geomorphic Setting, Substrate Type, Vegetation Type, and Vegetation Density was evaluated using the classification of four subject-matter experts as validation (Thieler and others, 2016). The expert validation was performed on a 10 percent random sample of points collected using iPlover in 2014 and 2015 (k = 181, n = 1799). Attribute values were updated between 2014 and 2015 to provide greater specificity and conformance with biogeomorphic nomenclature. Thus, values in the 2014 dataset were first reclassified to 2015 values.
Subject-matter experts used the photograph and geolocation of each sample point to assign the biogeomorphic classification (across all four variables). Experts were selected based on their experience and history of relevant publications on the geology and geomorphology of coastal environments in the study domain, as well as demonstrated expertise in the interpretation of remotely sensed data. Truth values for each sample point were selected according to expert consensus.
The following percentages indicate the agreement between expert consensus and field user classification for all test points for which expert consensus could be determined: 58% for geomorphic setting (90 agreeing points out of 156 truth), 80% for substrate type (121 out of 151), 84% for vegetation type (141 out of 167), and 77% for vegetation density (118 out of 153) (see Table 2 and Supplemental Table 2 in Thieler and others, 2016). Of the 181 locations, an expert consensus could be reach in 151, 167, and 153 cases for substrate type, vegetation type, and vegetation density respectively. Systematic errors were more common for the substrate and vegetation variables compared to geomorphic setting (Table 3 in Thieler and others, 2016). What experts classified as Sand was frequently classified as Shell/Gravel/Cobble by iPlover field users and vice versa, accounting for 86% of the differences in this variable. Vegetation Type was frequently classified as Herbaceous by experts and as None by iPlover users and vice versa, accounting for 81% of the differences for this habitat variable. Finally, the most frequent differences in Vegetation Density were between None and Sparse (51% of differences) and between Sparse and Moderate (23%; Table 3 in Thieler and others, 2016). The most common differences between the classifications of iPlover field users and experts occurred at the "boundaries" of categorical habitat characteristics.
Experts and iPlover field users also disagreed frequently over whether a point fell in no vegetation or sparse vegetation or whether a point fell within sparse or moderate vegetation cover. Experts and iPlover field users were instructed to visually estimate vegetation density and did not use field-based techniques such as quadrat sampling to quantify density. Therefore, differences in classifications between experts and iPlover field users would be expected in places where vegetation density was close to 20%, separating Sparse from Moderate vegetation, or close to 90%, separating Moderate from Dense vegetation.