What is an elevational range?

Elevational distributions have long fascinated scientists, an interest that has burgeoned with studies of predicted upslope range shifts under climate change. However, this body of work has yielded conflicting results, perhaps due to varied conceptual and statistical approaches. Here, I explore how ecological processes and researcher decisions shape the patterns characterized by elevational ranges. I use community science data to illustrate 1) that elevational ranges include variation in abundance; 2) that elevational ranges are usually estimated, not observed directly; 3) that elevational ranges are dynamic across short distances and time intervals; and 4) that how we describe elevational ranges has consequences for inference of range shifts. I present a conceptual framework for understanding elevational ranges across multiple spatial scales, and propose elevational distributions are governed by scale-dependent processes. This hypothesis implies accurately quantifying elevational ranges and learning how they are formed or maintained requires matching questions to an appropriate scale domain. I provide a list of best practices for studying elevational ranges, and highlight promising directions for future research into these complex phenomena.

Methods
I described the U.S. range of Junco phaeonotus in May, June, and July using observations and sampling effort from the April 2023 release of eBird’s Basic Dataset (EBD). eBird is a community science platform where contributors submit ‘checklists’ noting the presence and/or abundance of bird species detected during an observation period of arbitrary duration and distance. eBird includes automated quality filters to initiate expert review of implausible or unusual records, removing them from its data products. I used the R package auk v.0.6.0 (Strimas-Mackey et al. 2018) to apply a series of filters and manipulations that were adapted from eBird’s best practices for species distribution modeling (Johnston et al. 2021; Strimas-Mackey et al. 2023) but more stringent in maximum checklist duration (<5 hours) and traveled distance (<2 km). After pairing checklists with elevation and slope aspect data and assigning each observation to a 50-meter elevational bin, I trained a random forest algorithm on detection / non-detection data using the R package ranger v.0.15.1 (Wright & Ziegler 2015), extracting the marginal effect (or “partial dependence”) of elevation on encounter rate. Additional details on data filtering and species distribution modeling are available in the Supporting Information.

To understand how different elevational range metrics influence our understanding of climate warming-driven distributional shifts, I used the empirical mean, maximum, and minimum elevation of all Junco phaeonotus observations to simulate hypothetical elevational ranges where relative abundance or encounter rate fit one of two different statistical distributions (normal and skew normal). In each instance, I used the corresponding density distribution function in base R v.4.3.0 or fGarch v.4031.90 to simulate the elevations of 1000 observations before and after a 150-meter upslope shift in the mean, removing all records falling above a hypothetical 3200-meter summit elevation, and again assigning each observation to a 50-meter elevational bin.

Additional details on data filtering and species distribution modeling are available in the paper's Supporting Information.