Data from: The performance of drones and artificial intelligence for monitoring sage-grouse at leks
Data files
May 14, 2025 version files 112.84 KB
-
NmixDF.csv
4.55 KB
-
README.md
4.09 KB
-
SG_drone_data_longB.csv
96.02 KB
-
SG.drone.response.csv
8.18 KB
Abstract
We evaluated the effectiveness of drone-based survey protocols combined with an AI count model relative to traditional ground-based visual surveys for counting sage-grouse at leks. Drone-induced flushing of sage-grouse from leks occurred in 16% of flight attempts. Point-of-interest (POI) flight profiles outperformed linear flight profiles in counting accuracy for both AI and manual methods. POI flights provided more images and a larger field of view, resulting in counts similar to traditional ground-based visual (GBV) lek surveys, while linear flights consistently produced undercounts. Our custom AI counter (INDECS) yielded counts of sage-grouse similar to manual counts in POI surveys, but not in linear surveys. When integrated into modified N-mixture models, drone surveys with POI profiles yielded precise estimates of detection probabilities and abundance for all survey methods that resulted in similar inference to GBV surveys. Our results suggest that AI-enhanced drone surveys, particularly with POI flight profiles, offer a promising alternative to traditional surveys with reduced bias and improved consistency in sage-grouse population monitoring.
https://doi.org/10.5061/dryad.v41ns1s6v
Description of the data and file structure
From 19 March to 4 May in 2021 and 2022, we conducted ground-based visual surveys (GBV) at known sage-grouse leks using standard protocols, followed immediately by drone surveys. GBV surveys were carried out 30 minutes before and after sunrise, observing leks from a distance with binoculars. We used a Phantom 4 pro drone with a thermal camera for up to four aerial surveys per lek, using either point-of-interest (POI) or linear flight paths at different altitudes (150 ft and 200 ft). We manually counted sage-grouse from drone video footage and also used an AI algorithm, INDECS, to count birds from infrared videos. INDECS, built on the YOLO object detection framework, identified and counted sage-grouse in video frames. We analyzed the probability of drone-induced flushing of sage-grouse using a generalized linear mixed effects model, considering flight height and profile. The analyses were conducted in a Bayesian framework, assessing model convergence and fitting via posterior predictive checks.
"NA" values in data tables indicates "not available". Data were not collected or observations missing.
Files and variables
File: SG.drone.response.csv
Description: Data frame for analysis of sage-grouse lek flushes to drone flights.
Variables
- LekID: unique lek identifier
- Date: Date of the drone flight attempt (MM/DD/YYYY)
- Survey.No.: number of the attempt
- Flight.profile: Point of interest or linear flight profile
- Flight.Ht: altitude of drone during survey attempt in feet
- Birds.flushed: at least 2 birds flushed from lek during the attempt (Y/N)
- Distance: distance from the lek that drone was deployed for flushed leks in meters.
File: SG_drone_data_longB.csv
Description:
Variables
- Lek: unique lek name
- Lek.int: unique integer for each lek
- Visit: visit number for each survey
- Flight_ID: unique identifier for each lek flight with drone
- Year: year of the survey
- Date: calendar date of the surveys (MM/DD/YYYY)
- C_Time: Time of the survey count
- Sky: sky conditions (c= clear, pc=partly cloudy, f=fog, o=obscured/complete cloud cover)
- Distance_from_lek: Distance from lek in feet of drone launch
- Temp: in Celcius
- Wind_speed: kph
- Wind_direction: direction of wind
- Cloud_cover: percent cloud cover
- Moon: moon phase (% of full)
- Pressure: barometric pressure in hectopascals
- Dew_point: dew point (degrees C)
- Humidity: humidity (%)
- Profile: Survey profile type: Gr = ground survey, POI = point of interest, Linear = linear
- Height: flght height of drone in feet
- Count: Maximum count of sage-grouse during survey
- Type: Counter type: GB = manual count from ground, xAC = AI-based count from drone images, xMC = manual count from drone images
- GB: indicator variable for a ground based survey occurring before the drone flights
- GA: indicator variable for a ground based survey occurring after all drone flights
- AC: indicator variable for an AI-based counter drone flight
- MC: indicator variable for an drone survey with manual counts
File: NmixDF.csv
Description:
Variables
- Lek: unique lek name
- Lek.int: unique lek number
- GB1: GB1 - GB7 are counts of sage-grouse during each visit by ground-based observer
- Year: Year of the study
- Profile: POI or linear flight type
- Height: Drone flight height in feet
- Date1: Date1-Date7 are the dates of the replicate surveys of sage-grouse leks ((MM/DD/YYYY)
- Time1: Time1-Time7 are the times of the replicate surveys at sage-grouse leks (HH:MM)
- Wind1: Wind1-Wind7 are the wind speeds (kmh) during the replicate surveys.
- Hum1: Hum1-Hum7 contain the relative humidity (%) for replicate sage-grouse surveys
- Sky1: Sky1-Sky7 sky conditions during the replicate sage-grouse surveys (c= clear, pc=partly cloudy, f=fog, o=obscured/complete cloud cover).
Code/software
R version > 4.1.3.