Improving bird abundance estimates in harvested forests with retention by limiting detection radius through sound truncation
Data files
Oct 18, 2024 version files 620.93 KB
-
AbundanceData.zip
145.06 KB
-
README.md
7.94 KB
-
TruncationData.zip
467.94 KB
Abstract
An inherent challenge with acoustically surveying birds is that the distance at which they can be detected depends on how far their song can be heard. We developed a distance-based sound detection space truncation method to correct for variable sampling radii due to surveying in forested or open conditions. The method was pivotal in evaluating bird responses to retention patches; without this methodological advancement, the impact of retention patches on songbird abundance was vastly underestimated. In the boreal forest, these patches of live trees are retained in regenerating harvested forests to provide ecological services for species adapted to natural disturbances. Although we did not verify our a priori assumption with ground observations, our findings suggest that limited-distance sampling better captures the effects of retention patches on bird use of harvested forests. When evaluated using unlimited distance surveys, retained trees had a negligible effect on bird abundance, whereas applying detection distance truncation highlighted the importance of retention on forest birds. We found that early to mid-seral forest songbirds benefited from retention patches, with notable increases in abundance after 10 years of regeneration. The size of retention patches, ranging from 0.1 to 1.2 hectares, did not have a linear relationship with bird abundance. Instead, edge effects stemming from the configuration of these patches emerged as key determinants of abundance for the majority of the species studied. Retention patches that were nearest to unharvested forests were used the most, compared to further into harvest areas. Our research not only highlights the underestimated impact of small-scale live tree retention on forest songbirds but also introduces a significant methodological innovation in the field of acoustic monitoring.
Description
Coda and data for a live-tree retention for songbirds recovery study, including distance truncation.
Authors
Isabelle Lebeuf-Taylor, Elly Knight, Erin Bayne
Date
September 2024
Experimental Methods
Retention in regenerating harvests
This study was conducted in harvested areas of boreal and foothills forests in Alberta, Canada. We selected 392 sites, including 246 with retention patches and 146 without residual trees within 150 m. Acoustic surveys were performed using SM2, SM3, or SM4 Autonomous Recording Units (ARUs) from Wildlife Acoustics Inc., placed at least 150 m from the harvest edge. ARUs recorded 1-minute segments every 20 minutes from 1 hour before to 4 hours after dawn, for a minimum of 3 good weather days during the migratory bird breeding season (May 25th to July 6th), primarily from 2021-2023. After filtering out recordings with excessive background noise, 10 random 1-minute recordings per site were selected for analysis. Expert transcribers processed these recordings using the WildTrax platform, identifying and tagging songs of six focal migratory songbird species. Signal strength (Sμ) for each tagged song was calculated using SoX audio processing software via WildTrax, expressed as the root mean square (RMS) of the decibel relative to full scale (dBFS), averaged between the left and right microphones.
Sound attenuation predictive dataset
To establish the relationship between song amplitude and distance, we conducted a playback experiment using 15 indicator bird species selected to represent acoustically similar groups. These species were chosen based on a cluster analysis of song characteristics (mean peak frequency, duration, bandwidth, and spectral entropy) extracted from 3951 songs of 79 species. Playbacks were performed during the 2023 field season, using a FoxPro speaker broadcasting at 90 dB sound pressure level. At each site, songs were played at three randomly selected distances between 20 m and 150 m along a homogeneous forest transect, with the speaker positioned at an average height of 1.75 m. ARUs (SM2, SM3, or SM4) continuously recorded the playbacks. A total of 2250 songs were broadcast across various sites. The recordings were then processed in WildTrax, where each detected song was tagged and its signal strength (Sμ) was extracted, excluding recordings with excessive background noise.
Contents
- Data Files
- R Scripts
- R Markdown Document
- Output Files
- Software Requirements
- Data Dictionary
- Analysis Workflow
- Additional Notes
1. Data Files
Abundance Data
Abundance files contain counts for the Olive-sided Flycatcher (OSFL), Red-eyed Vireo (REVI), White-throated Sparrow (WTSP), Ruby-crowned Kinglet (RCKI), Yellow-rumped Warbler (YRWA), and Tennessee Warbler (TEWA). Other columns are the location and visit datatime.
Abundance within 150m.csv: Counts are within 150m of ARU.Abundance within 250m.csv: Counts are within 250m of ARU.Abundance untruncated distance.csv: Counts are not distance truncated.
Covariates.csv
- location: Site name
- Year_since_logging: years between sampling and logging event. Determined with provincial harvest polygons, and verified with Landtrendr's NBR spectral analysis.
- latitude: latitude of site
- longitude: longitude of site
- RETN_m2: measured area of retention patch in m^2
- RETN_perimeter: measured perimeter of patch in m
- Veg_cat: Majority vegetation composition of tree stand within 150m of sampling unit, defined as >= 70% area occupied by a tree group. Categories are Pine (jack pine or lodgepole pine), deciduous (white birch, aspen, poplar), spruce (white spruce or Engelmann spruce), mixedwood (spruce understory, deciduous overstory).
- Dist_near_forest: the distance form the edge of the patch to the nearest forest of a minimum 80 years of age, in m.
ampl_dist_spp.csv
Contains amplitude measurements at various distances for different bird species.
- location: Site name
- species_code: 4-letter code of the species whose song was broadcast toward the recording unit
- forest: Forest composition type. Options are AS (aspen), BL (black spruce), MI (mixedwood), OP (open), PI (pine), SP (upland spruce).
- distance: distance at which song was ploayed, in m.
- left_amplitude: detected amplitude by the left mic, in dBFS RMS
- right_amplitude: detected amplitude by the right mic, in dBFS RMS
Counts to truncate.csv
Long format data of mean amplitude of detected songs in harvests, including traunction model parameters. Truncation model can be applied to this data.
- location: site name
- recording_date_time: datae and time of site visit
- species_code: 4-letter code of detetected species. Options are TEWA, OSFL, REVI, WTSP, RCKI, YRWA
- Year_since_logging: years between sampling and logging event.
- mean_amp: amplitude avergaed between left and right mic as dBFS RMS.
- SM2: whether ARU unit was an SM2 (1) or not (0).
locations_sm2_status.csv
Information about recording locations and SM2 recorder status.
- location: site name
- SM2: whether ARU unit was an SM2 (1) or not (0).
number_of_transcribed_recordings_per_visit.csv
- location: site name
- number_of_visits: number of processed visits per site
predicted_distance_amplitudes.csv
Output of playback experiment models.
- distance: known distance in m of played song
- *species_code: 4-letter code of the species whose song was broadcast toward the recording unit
- BinForest: binary forest presence (FO) or absence (OP)
- SM2: whether ARU unit was an SM2 (1) or not (0).
- lwr: 2.5% of predicted amplitude in dBFS RMS
- upr: 97.5% of predicted amplitude in dBFS RMS
- predicted: predicted amplitude, given model covariate of species at specific distance in dBFS RMS.
transcribed_tasks_all.csv
Recordings per site that are transcribed
- location: site name
- recording_date_time: ID of transcribed recordings
2. Scripts (Zenodo)
sound_attenuation_analysis.Rmd: R Markdown document to model sound attenuation.apply_perceptibility_truncation.ipynb: Jupyter notebook for applying perceptibility truncation to acoustic survey data.negBinom_models.ipynb: Jupyter notebook for modelling species-level relative abundance response to variable retention, after applying distance truncation.
3. Software Requirements
- R (version 4.0.0 or later)
- Python (version 3.7 or later)
- Jupyter Notebook
- cmdstan (version 2.32.1)
R Packages:
- tidyverse
- glmmTMB
- ggplot2
- gridExtra
- DHARMa
- MuMIn
- caret
Python Packages:
- pandas
- numpy
- matplotlib
- seaborn
- scikit-learn
- arviz
- cmdstanpy
4. Analysis Workflow
- Data Preparation:
- Load and merge datasets
- Calculate mean amplitude
- Create binary forest variable
- Convert variables to appropriate types
- Model Fitting:
- Fit several glmmTMB models with different predictors
- Compare models using AICc
- Model Diagnostics:
- Check residuals using DHARMa
- Calculate R-squared values
- Predictions and Visualization:
- Generate predictions for open and forested habitats
- Create plots showing attenuation patterns
- Cross-validation:
- Perform 10-fold cross-validation
- Calculate RMSE and MAE
- Perceptibility Truncation:
- Apply distance-based truncation to abundance data
- Compare results at different truncation distances
- Negative Binomial Models:
- Fit negative binomial models for species abundance
- Compare models using WAIC
- Evaluate model fit using diffenret truncation distances using loo and plots
- Model evaluation with R-hat & trace plots
Study Area & Site Selection
We conducted the study in harvested areas in the boreal and foothills forests of Alberta (Figure 1). Dominant vegetation in and surrounding harvest areas consist of stands of jack pine (Pinus banksiana) or lodgepole pine (P. contorta), mixed-wood stands of upland spruce and deciduous trees, conifer stands of white spruce (Picea glauca) or Engelmann spruce (P. engelmannii), and deciduous stands of trembling aspen (Populous tremuloides), and to a lesser extent balsam poplar (P. balsamifera) and white birch (Betula papyrifera).
Using the ABMI Wall-to-Wall Human Footprint Inventory (ABMI 2021), we identified harvest areas. The year of harvest in these areas was determined by detecting the inter-annual drop in Normalised Burn Ratio (Hird et al. 2021), utilizing the Landtrendr Google Earth Engine pixel time-series tool (Kennedy et al. 2018). Tree stand types within 300m of the survey coordinates were extracted using the ABMI Alberta Wall-to-Wall Land Cover Map, and summarized by dominant stand if a type occupied >=70% of the area, categorizing them as deciduous, spruce, pine, or mixed-wood stands. A selection of retention patches ranging in size from 0.1-12,000 m2 was made evenly across harvest areas aged between 1-22 years old, ensuring a proportional representation of harvested forest types. Out of a total of 392 sites visited, 246 contained retention patches, and 146 were in harvest areas with no residual trees within 150 m. Retention patches and areas without retention were mostly identified using satellite imagery.
Acoustic Surveys & Processing
Acoustic surveys were conducted by affixing a single SM2, SM3, or SM4 ARU (Wildlife Acoustics Inc.) at least 150 m from the harvest edge. ARUs recorded 1-minute segments passively every 20 minutes from 1 hour before to 4 hours after dawn over a minimum of 3 good weather days during the migratory bird breeding season (May 25th to July 6th). Sites were mostly visited over the course of 3 years, from 2021-2023.
Recordings in which loud background noise due to inclement weather or industrial activity were identified visually by an observer and discarded to ensure bird songs were not drowned out by ambient noise and their amplitudes could be calculated. After filtering out noisy recordings, 10 random 1-minute recordings were selected from each study site, and their spectrograms were processed by expert transcribers using the WildTrax platform (https://wildtrax.ca). In a multi-visit framework, 3-5 surveys per site are more typical, but we opted to process more recordings per site to test our new distance truncation method. From 6,250 recordings, we identified every singing individual for each of the 6 focal species (Drake et al. 2016) and tagged the song of each individual in that minute that was the least masked by other birds or noises. This ensured that the tag's amplitude was related to the individual of interest.
For each song, signal strength (Sμ) was calculated with the SoX audio processing software (https://sox.sourceforge.net) via the bioacoustics platform Wildtrax (Wildtrax, Edmonton, AB, Canada). Signal strength, which can be understood as the volume at which a bird song is recorded by an ARU, is extracted from a SoX analysis of the temporal and frequency bounded song spectrogram tag as the root mean square (RMS) of the decibel relative to full scale (dBFS), averaged between the directional left and right microphones. RMS dBFS refers to the average power of the audio signal relative to the maximum possible level in the digital system, which is set at 0 dBFS to prevent digital clipping and ensure optimal audio quality. For example, -3 dBFS would indicate that the signal is 3 decibels lower than the maximum possible level.