Methods
This study utilizes data collected by the Alberta Biodiversity Monitoring Institute (ABMI) at 2510 sites in a 20 km systematic grid across the province of Alberta. Autonomous recording units (ARUs) and remote cameras are deployed at each site. The ARUs record bioacoustic data for 3-10 minutes at 8 different times during each 24-hour period, and the cameras are motion activated. The dataset for this study consists of sites where both recording data and camera data were available between May 1 and June 30 from 2015-2022. May 1 to June 30 is the breeding season for songbirds and is the time period during which birds can be most reliably detected with ARUs when present.
A single species, single season, occupancy model is used to model the relationship between cowbird occupancy and cattle presence in at sites with different habitat suitability index (HSI) values. The predictor variables are the presence or absence of cattle and the HSI, and the response variable is the probability of occupancy by cowbirds. This model assesses whether cowbirds are likely to occupy sites outside of what is considered suitable habitat in the absence of cattle.
Occupancy models use presence absence data to estimate the probability that a site will be occupied by the target species, while accounting for imperfect detection. The model uses the detection history (number of detections and non-detections at a site) to estimate a detection probability. This is an estimate of the likelihood of detecting the species when it is present. The model can also account for covariates that affect detection of the species, like time of year and time of day (birds are more likely to vocalize during breeding season, and early in the morning). These covariates are predictors of the detection probability. Site covariates can also be included, which describe characteristics of the site that are believed to influence the probability of occupancy by the target species. In this case, the site covariates are the presence of cattle, the HSI, how the two factors interact to affects the probability of occupancy by cowbirds. Occupancy modeling was done using the 'unmarked' package in R.
A generalized linear model (GLM) was used to test the relationship between cowbird presence or absence, habitat suitability, and cattle presence or absence. The response variable was the presence or absence of cowbirds, and the predictor variables were habitat suitability (HSI) and cattle presence or absence. The GLM was fitted using a binomial distribution with a logit link function. A GLM can handle binary and count data, making it suitable for modeling the presence/absence of a species. Additionally, a GLM allows for the inclusion of multiple predictor variables, such as habitat suitability and cattle presence/absence, and can model the relationship between these predictors and the binary response variable. Finally, GLMs allow for the use of different link functions to model the relationship between the predictors and response variable, and the logistic link function used in this analysis is appropriate for binary data.
The GLM and occupancy models can provide different insights into the relationship between predictor variables and the response variable. In general, the GLM is better suited for analyzing the relationship between a binary response variable and one or more predictor variables, while the occupancy model is better suited for analyzing the probability of detection of a species in a given habitat.
In the context of this study, the GLM was used to model the probability of cowbird presence as a function of habitat suitability and cattle presence/absence. The GLM can provide estimates of the effect size of each predictor variable on the response variable, as well as statistical significance tests for each variable.
The occupancy model, on the other hand, can provide estimates of the probability of detection of cowbirds in different habitats, as well as estimates of occupancy probability. This can provide insight into the factors that influence the detectability of the species, as well as the overall probability of occurrence. However, the occupancy model may not provide as detailed information on the effect size of individual predictor variables on the response variable as the GLM.
A single species, single season, occupancy model is used to model the relationship between cowbird occupancy and cattle presence in at sites with different habitat suitability index (HSI) values. The predictor variables are the presence or absence of cattle and the HSI, and the response variable is the probability of occupancy by cowbirds. This model assesses whether cowbirds are likely to occupy sites outside of what is considered suitable habitat in the absence of cattle.
Occupancy models use presence absence data to estimate the probability that a site will be occupied by the target species, while accounting for imperfect detection. The model uses the detection history (number of detections and non-detections at a site) to estimate a detection probability. This is an estimate of the likelihood of detecting the species when it is present. The model can also account for covariates that affect detection of the species, like time of year and time of day (birds are more likely to vocalize during breeding season, and early in the morning). These covariates are predictors of the detection probability. Site covariates can also be included, which describe characteristics of the site that are believed to influence the probability of occupancy by the target species. In this case, the site covariates are the presence of cattle, the HSI, how the two factors interact to affects the probability of occupancy by cowbirds. Occupancy modeling was done using the 'unmarked' package in R.
A generalized linear model (GLM) was used to test the relationship between cowbird presence or absence, habitat suitability, and cattle presence or absence. The response variable was the presence or absence of cowbirds, and the predictor variables were habitat suitability (HSI) and cattle presence or absence. The GLM was fitted using a binomial distribution with a logit link function. A GLM can handle binary and count data, making it suitable for modeling the presence/absence of a species. Additionally, a GLM allows for the inclusion of multiple predictor variables, such as habitat suitability and cattle presence/absence, and can model the relationship between these predictors and the binary response variable. Finally, GLMs allow for the use of different link functions to model the relationship between the predictors and response variable, and the logistic link function used in this analysis is appropriate for binary data.
The GLM and occupancy models can provide different insights into the relationship between predictor variables and the response variable. In general, the GLM is better suited for analyzing the relationship between a binary response variable and one or more predictor variables, while the occupancy model is better suited for analyzing the probability of detection of a species in a given habitat.
In the context of this study, the GLM was used to model the probability of cowbird presence as a function of habitat suitability and cattle presence/absence. The GLM can provide estimates of the effect size of each predictor variable on the response variable, as well as statistical significance tests for each variable.
The occupancy model, on the other hand, can provide estimates of the probability of detection of cowbirds in different habitats, as well as estimates of occupancy probability. This can provide insight into the factors that influence the detectability of the species, as well as the overall probability of occurrence. However, the occupancy model may not provide as detailed information on the effect size of individual predictor variables on the response variable as the GLM.