Accurate biodiversity and population monitoring is a requirement for effective conservation decision-making. Survey method bias is therefore a concern, particularly when research programs face logistical and cost limitations.

We employed point counts (PCs) and autonomous recording units (ARUs) to survey avian biodiversity within comparable, high elevation, temperate mountain habitats at opposite ends of the Americas: 9 mountains in British Columbia (BC), Canada and 10 in southern Chile. We compared detected species richness against multi-year species inventories and examined method-specific detection probability by family. By incorporating time costs, we assessed the performance and efficiency of single vs. combined methods.

Species accumulation curves indicate ARUs can capture ~93% of species present in BC but only ~58% in Chile, despite Chilean mountain communities being less diverse. The avian community, rather than landscape composition, appears to drive this dramatic difference. Chilean communities contain less-vocal species, which ARUs missed. Further, 6/13 families in BC were better detected by ARUs while 11/11 families in Chile were better detected by PCs. Where survey conditions differentially impacted method performance, PCs mostly varied over the morning and with canopy cover in BC, while ARUs mostly varied seasonally in Chile. Within a single year of monitoring, neither method alone was predicted to capture the full avian community, with the exception of ARUs in the alpine and subalpine of BC. PCs contributed little to detected diversity in BC, but including this method resulted in negligible increases in total time costs. Combining PCs with ARUs in Chile significantly increased species detections, again, for little cost.

Combined methods were among the most efficient and accurate approaches to capturing diversity. We recommend conducting point counts while ARUs are being deployed and retrieved in order to capture additional diversity with minimal additional effort and to flag methodological biases using a comparative framework.

Survey Locations

In Canada (2019), we surveyed nine mountains in the D'ze Kant (Bulkley)-Nechako and Kitimat-Stikine regions of British Columbia (BC; 1000–1801 m elevation; Fig. 1). In Chile (2018), we surveyed 10 mountains in La Araucanía and Los Ríos regions (1000–1700 m elevation). These mountains fall within the traditional unceded lands of the Wet'suwet'en, Gitxsan, and Tsimshian First Nations in BC and the Mapuche people in Chile. The farthest latitudinal and longitudinal distance among survey locations was 117 km and 106 km, respectively, in BC, and 178 km and 60 km, respectively, in Chile. Surveyed habitats across elevation gradients in both regions were classified as: montane habitat (≥50% tree cover, 1000 – 1557m a.s.l.); subalpine (≥5 – 50% tree cover, 1169 – 1658m a.s.l.); and alpine (0 – 5% tree cover, 1319 – 1801m a.s.l; Boyle & Martin, 2015).

All point count locations are listed in the associated file: "PointCountLocations_BC_CH.xlsx"

BC survey sites fall within five biogeoclimatic zones: Coastal Mountain Hemlock, Mountain Hemlock, Engelmann Spruce-Subalpine Fir, Boreal Altai Fescue Alpine, and Coastal Mountain-heather Alpine (British Columbia Ministry of Forests, Lands, Natural Resource Operations, and Rural Development, 2018). Montane habitat is primarily old growth conifer forest interspersed by avalanche chutes, producing age heterogeneity. The subalpine consists of woody shrubs, grasses, and perennial herbs with some tree cover, while the alpine is characterized by the presence of fescue grasses, heather, mosses, and lichens.

In Chile, montane habitats are dominated by old growth mixed broadleaf-conifer forests, with about 10% mid-successional forest. Subalpine habitat is a mix of highland herbaceous meadows, shrubs, and sparse patches of trees and/or krummholz. Perennial herbaceous plants, shrubs, few or no trees, and bare rock/scree characterize alpine habitat. Vegetation structure varies within- and among-mountains based on natural disturbances (i.e. volcanic eruptions) and/or land-use history (Caviedes & Ibarra, 2017).

Point Counts

Starting at sunrise, 95% of surveys were conducted within 5 hrs to encompass peak bird activity. The remaining 5% of surveys occurred 5–6 hrs after sunrise within subalpine and alpine habitats due to logistical constraints (total range: 04:53–11:24 hrs in BC and 05:51–12:18 hrs in Chile). Each mountain was surveyed from bottom to top (upslope) along transects with five designated point counts, 200m apart, within each of the three habitat types for a total of 15 point counts per mountain. In BC, steep topography meant that the subalpine on Thornhill Mountain fit only four point count locations and that the alpine on Nadina Mountain was inaccessible until July. Thus, BC had 129 point count sites: one fewer subalpine site and five fewer alpine sites in total.

During each 6-minute point count, birds were counted by sight and sound. Observers kept track of individual birds to minimize duplicate detections among point counts. Infinite radius detections were used to provide a fair comparison to ARU sampling; 95% of individuals in British Columbia and 99% of individuals in Chile were detected within 100m. Point counts that occurred near habitat transition zones did not record species that called  >100m away if they were clearly within adjacent habitats or if they were in that direction and were unlikely to be in the focal habitat, based on their ecology. Counts were repeated three times within each respective breeding season: between May 30 – July 16 in BC, and between November 7 – December 21 in Chile, to assess detection probability and address seasonal variation in detection. Repeated site visits were separated by ~2 weeks.

Acoustic recordings and analysis

Song Meter SM4 Autonomous Recording Units (ARUs; Wildlife Acoustics Inc. ©, Maynard, MA) with two omni-directional microphones were deployed at two point count sites within each habitat and away from habitat transition zones (6 ARUs/mountain, >400 m apart) in both BC and Chile. In BC, 36 units were deployed on all 9 mountains. As above, the alpine on Nadina Mountain was inaccessible resulting in ARUs being deployed at a total of 52 point count sites rather than 54. Units recorded for an average of 21 days within the BC breeding season: 10 – 20 days on six mountains, and 32 – 35 days on the remaining 3 mountains, between June 3 and July 15.  In Chile, 6 units were deployed on five mountains (30 point count sites in total) and recorded for an average of 6 days within the breeding season: 5 – 10 days each, between November 13 – December 28. ARUs recorded at a sampling rate of 24000 Hz in stereo wav format using default acoustic gain settings for the microphones. Units were mounted on a tree within several meters of the point count site, or on a PVC pipe at ~1.5 m height in the alpine. Units were programmed to record 30-mins on and 30-mins off, starting at sunrise and ending 5 hours after sunrise (5 x 30-minute recordings/day). From the full deployment period, we randomly sampled two (BC) or three (Chile) different days per point count site. In BC, four days were selected for the three mountains where the ARUs were deployed for a longer period (early and late breeding season). To be comparable with point counts, we randomly chose one, 6-minute interval to analyze from each of the 5 x 30-minute recordings on these days (5 x 6-minute point counts/day). Thus, within each survey day, ARUs were sampled within hourly windows from between 0–1 hr after dawn (“hour 0”) to between 4–5 hrs after dawn (“hour 4”). If any given hour(s) within the selected day had unfavourable conditions (wind or rain) that interfered with the audio, another day was selected randomly to obtain the missing time period(s). A total of 700 site-surveys in BC and 450 site-surveys in Chile were analysed.

Sound recordings were analyzed using Audacity® software (V2.3.0, Audacity Team, 2020). Three skilled observers reviewed all recordings: two in BC and one in Chile. All observers had experience conducting point counts in the same regions. In BC, both observers analyzed five of the same recordings to confirm detection consistency and conferred with each other on all recordings when species identification was uncertain. Spectrograms were scanned manually in stereo format as the observer listened to the recording. Species that were more difficult to identify were compared with recordings available on bioacoustics libraries such as the Xeno-Canto Foundation (2019) and/or sent to other skilled ornithologists.

Abiotic Variables

At each point count we recorded average temperature and wind speed using a Kestrel 3500 weather meter (Nielsen-Kellerman Company, PA, USA). We additionally scored wind as a categorical variable (Beaufort scale: 0-3) during point counts to allow for comparison with ARU wind scores that were assigned on the same scale based on interference with the audio recording. We also recorded percent canopy, understory (vegetation ~30cm in height), shrub, and ground cover (tundra vegetation, snow, rock, and dead trees) within a 50m radius of all point count sites. More canopy foliage in Chile was deciduous than in BC. Canopy cover value therefore increased with leaf-out during the season in Chile, while values in BC were static.

BC species are coded using four-letter alpha codes (for English common names); https://www.birdpop.org/pages/birdSpeciesCodes.php.

Chilean species are coded using six-letter alpha codes (for scientific names).

Species included in family-level groupings are listed in Table A1 of the associated manuscript.

Habitat is coded as: AL (alpine), SA (subalpine) and UM (upper montane)

Method is coded as: PC (human point count observation), ARU (autonomous recording unit)

Locations (lat, long) of point count sites are available in the associated file: "PointCountLocations_BC_CH.xlsx"

"spp_acc" files contain species precence/absence data formatted for species accumulation curves/Hill number calculations. Note: PC records were further subsetted to only those sites also surveyed by ARUs prior to analysis.

"Unmarked" files contain family-level precence/absence data and site and survey-level condition variables for program "Unmarked".