Combining potentially incompatible community datasets when harmonizing forest inventories in subarctic Alaska, USA

Abstract

Aims Plant responses to disturbances and environmental variation can manifest in communities as compositional nestedness (i.e., one community is a subset of another) and/or turnover (two communities represent different compositional gradient spaces). Yet, different sampling designs can artificially give an illusion of such compositional differences among two datasets, making it problematic to harmonize them in multi‐species analysis. We test the prediction that sampling differences which increase beta‐diversity components (nestedness and turnover) among two vegetation datasets will decrease their exchangeability. Location Boreal forests of Tanana River region, interior Alaska, USA. Methods We develop novel methods for comparing compositional variation among two datasets in nonmetric multidimensional scaling (NMDS) ordination. Resampled NMDS establishes internal sampling variability for each dataset independently, and reciprocal NMDS determines external exchangeability when the two are mutually exchanged. We first compare simulated data with specified beta‐diversity differences, then evaluate two forest inventories based on local vs regional sampling designs in Alaska's boreal forests. Results As simulated species turnover and nestedness increased, internal sampling variability remained essentially constant, but external exchangeability progressively declined. Species turnover (not nestedness) had the larger negative effect on exchangeability. Among the boreal forest inventories, internal sampling variability was relatively similar, and exchangeability was weakly moderate, but the regional inventory exhibited much better fit to broad‐scale environment. Species turnover (not nestedness) contributed the majority of beta‐diversity differences among the two forest inventories, suggesting that strong environmental gradients were unequally represented. Conclusions Species turnover alters multivariate outcomes more drastically than species nestedness. Therefore, combining two vegetation datasets may be inadvisable when turnover prevails. Instead, a multi‐scale perspective, with separate but complementary forest inventory analyses, can portray local and regional variation at appropriate scales. Our method is tractable for assessing exchangeability of potentially inconsistent sampling designs, like those that are common in synthesis studies and long‐term ecological monitoring.

Lack of congruence between terrestrial and epiphytic lichen strata in boreal forests

“…2018), the BNZ/CPC inventory was an intensified sample (more plots in a smaller area) conducted by cooperators at the Bonanza Creek LTER site and the University of Alaska-Fairbanks. The BNZ/CPC inventory aimed to use existing long-term records and infrastructure to measure the status and trends of multiple forest attributes in great detail (FIA 2012; Smith & Gray 2019).…”

“…2016) as 30-yr annual normals covering the period 1980–2010, which directly precedes the lichen sampling period. Finally, we obtained vascular vegetation abundances from a parallel study (Smith & Gray 2019): basal area of the two most dominant conifers ( Picea mariana and P. glauca ), basal area of the two most dominant hardwoods ( Populus tremuloides and Betula neoalaskana ), and percentage cover of the two most dominant moss functional groups (N-fixing feather mosses and Sphagnum peat mosses).…”

Evaluating and presenting uncertainty in model‐based unconstrained ordination

“…In contrast to constrained ordination methods where environmental features related to the sample units are used (Anderson & Willis, ; Birks, Peglar, & Austin, ; Økland, ), unconstrained ordination models do not include ancillary data about the sample units, but rather only use species (or other taxon) composition information to estimate the locations of sample units along compositional gradients. Traditionally, distance‐based methods have been used to determine the compositional gradients and locations of sample units (Legendre & Gallagher, ; Roberts, ); however, these methods require resampling‐based approaches for inference and uncertainty assessment (De Leeuw & Meulman, ; Heiser & Meulman, ; Jacoby & Armstrong, ; Smith & Gray, ). Recently, model‐based methods have been introduced for unconstrained ordination (Hui, Taskinen, Pledger, Foster, & Warton, ; Ovaskainen et al, ; Warton, Blanchet, et al, ; Warton, Foster, De'ath, Stoklosa, & Dunstan, ), which do permit uncertainty assessment.…”