Figure 2. Pest abundance (total number of individuals per trap night) in relation to data collection date. Data were collected at agricultural fields in Champaign County, IL, in 2021. Red line represents a ‘best-fitting-line’.
Statistical analysis
Bat activity
Bat activity (i.e., number of bat calls per detector per night per phonic group, N = 2471, mean ± SD = 17 ± 37, range = 1 – 470) is a count variable and, as such, we used a negative binomial generalized linear mixed effects model with quadratic parameterization (glmmTMB) to analyze the data (Brooks et al., 2017; Hardin and Hilbe, 2007). We included a two-way interaction between distance to forest and relative pest abundance, a two-way interaction between distance to forest and phonic group, and a two-way interaction between phonic group and relative pest abundance as main predictor variables. We also included daily minimum temperature (Tmin) and daily precipitation (Precip) as fixed covariates as these environmental factors have been shown to significantly affect bat activity (Gorman et al., 2021). Lastly, we included a quadratic effect of date as a fixed effect, as we expected bat activity to increase as pups become volant, until mid-August, and decline towards late-September, as bats migrate to hibernation sites. We included detector (N = 50) nested within transect line ID (N = 10) as random effects to control for potential spatial and temporal autocorrelation biases. Statistical tests with a p-value lower than 0.05 were considered statistically significant. We report means and standard errors where appropriate.
Bat species diversity
We recognized several species within our dataset. Bat diversity was defined as the number of species detected per detector per night. However, in our study area Myotis species – apart from MYLU – were generally rarely identified through automated ID algorithms, so we combined all Myotis species with the exception of MYLU into one category ‘Myotis spp.’ Consequently, bat species diversity in our dataset ranged from 1 to 8 (i.e., EPFU, LANO, LACI, LANO, NYHU, PESU, MYLU, and Myotis spp ).
We used a similar negative binomial generalized linear mixed effects model with a quadratic parameterization to analyze bat diversity (N = 1698 detector-nights). We included a two-way interaction between distance to forest and relative pest abundance as main predictor variables. We also included the same covariates in this model: daily minimum temperature (Tmin), daily precipitation (Precip), and a quadratic effect of date as a fixed effect. The random effects were detector (N = 50) nested within transect line ID (N = 10).
Results
Bat Activity
Bat activity or bat calls per detector-night (calls/night) decreased significantly with increasing distance from forest (χ2= 16.37, df = 1, p < 0.001), such that bat activity decreased 56% from the forest edge (11.4 ± 1.8 calls/night) to 4000m away from the forest edge (5.0 ± 1.0 calls/night; Fig. 3A). Bat activity decreased less with increasing distance to forest when relative pest abundance was high (i.e., 1), compared to when relative pest abundance was low (i.e., 0; χ2 = 10.04, df = 1, p = 0.002; Fig. 3A). The relation between bat activity and distance to forest was not affected by phonic group (χ2 = 5.21, df = 2, p = 0.074; Fig. 3B). However, bat activity varied among phonic groups (χ2 = 2167.74, df = 2, p < 0.001) such that, on average, low-frequency bats were most active (20.7 ± 2.8 calls/night), mid-frequency bats had less activity (4.2 ± 0.6 calls/night), and high-frequency bats had the least activity (1.2 ± 0.2 calls/night) throughout the study. Bat activity decreased significantly with increasing relative pest abundance (χ2 = 14.22, df = 1, p < 0.001), but this effect varied by phonic group (χ2 = 28.07, df = 2, p < 0.001); while activity of low-frequency bats decreased with increasing relative pest abundance, activity of med- and low-frequency bats did not vary with increased relative pest abundance (Fig. 3C).
Lastly, bat activity increased with increasing daily minimum temperatures (χ2 = 137.67, df = 1, p < 0.001) and decreased with increasing daily precipitation (χ2 = 4.26, df = 1, p = 0.039). Modeled bat activity was also highest in the beginning of July (χ2 = 136.44, df = 2, p < 0.001).