Habitat Selection And Prediction Of The Spatial Distribution In Greater Horseshoe Bats, Rhinolophus Ferrumquinum At Multiple Spatial Scales

We studied habitat selection by greater horseshoe bats (Rhinolophus ferrumequinum) from the scale of landscape, forest stand and microhabitat in Luotong Mountain natural reserve in southwest of Changbai Mountain, and form a set of quantitative methods for animal habitat selection and spatial distribution prediction from landscape, forest stand and microhabitat scale. Combined remote sensing and Geographical Information System (GIS) technique, we predicted the potential spatial distribution from multiple spatial scales. The main conclusions were as follows:1. We assessed the seasonal activity of greater horseshoe bats in different habitats using acoustic survey. From relatively open habitat to relatively clutter habitat, there was an increase trend in the activity of greater horseshoe bats. The use frequency of different habitat was as follows: mixed woodland > cluttered broadleaved woodland > forest edge > pond > trail > oak woodland > stream > open broadleaved woodland > ridge line > rock > coniferous woodland > grassland > resident sites. We didn't detect any bats in farmland and treeline habitat. There were significant seasonal differences (p < 0.05) in pond, coniferous forest, cluttered broadleaved forest, open broadleaved forest, ridge, grassland and forest edge. There were high level activity of greater horseshoe bats in cluttered forest and trail in early summer, late summer, early autumn and late autumn, respectively. Except in late summer, the activity of greater horseshoe bats was affected by temperature and humidity in early summer, late autumn and late autumn, respectively (p < 0.05).2. In landscape scale, we used logistic regression model to study the effects of elevation, distance from the nearest stream, the length of terrestrial flyway, the length of riparian flyway, patch richness and edge density on habitat selection by greater horseshoe bats. We usedΔAICc and Akaike weight to assess and select the model with the lowestΔAICc. We constructed 31 habitat models including the variables in landscape scale. The model with the lowestΔAIC_c showed that elevation and distance from the neareast stream were the best predictors of the presence of greater horseshoe bats. In all the landscape variables, only the length of terrestrial flyway was significantly different in the presence sites and the absence sites (F = 4.787, p = 0.034).3. In forest stand scale, we studied the effects of forest type, forest cover, forest age, tree height and aspect on habitat selection by greater horseshoe bats, and constructed 31 habitat models including the variables in forest stand scale. Forest cover, aspect and forest type were the best predictor of habitat selection by greater horseshoe bats. The importance of forest cover was the highest (0.989), then aspect (0.633) and forest type (0.596), and the importance of height (0.424) and age (0.37) was low. In all the forest stand variables, only forest cover was significantly different in the presence sites and the absence sites (U = 40.0, p = 0.032). At the 52 presence sites of greater horseshoe bats, the aspect was south facing at 18 sites. There were no sites where the aspect was north or northwest. The results showed that eutropic aspect was important for foraging greater horseshoe bats.4. In microhabitat scale, we measured insect prey availability, vegetation structure, temperature and humidity data in different seasons in 75 sampling sites. These variables were extremely important to the activity of greater horseshoe bats. In trapped insects, total insect abundance in summer were higher than that in autumn (F = 504.054, p < 0.001). In late autumn, defoliation made the canopy cover lower than that in early summer, late summer and early autumn (F = 17.03, p < 0.001). We used Poisson Generalized Linear Model (GLM) to study the relationship between the activity of greater horseshoe bats and microhabitat variables, and found that the factors influencing habitat selection of greater horseshoe bats varied with seasons. In early summer, total insect abundance, shrub height and shrub density were the best predictors of habitat selection by greater horseshoe bats, thereinto insect resources were more important than shrub height and shrub density; in late summer, shrub height and density were the best predictors of habitat selection by greater horseshoe bats; in early autumn, total insect abundance and temperature were the best predictors of habitat selection by greater horseshoe bats; in late autumn, total insect abundance was the best predictor of habitat selection by greater horseshoe bats. This indicated that vegetation cover was more important than food when food resources were seasonally abundant, whereas food was more important than cover when food resources were seasonally scarce. These results revealed that there was a trade-off between the importance of food and cover for greater horseshoe bats.5. Fecal analysis was used to analyze the diet of greater horseshoe bats. Moths in the order Lepidoptera were the dominant component in the diet of greater horseshoe bats followed by Coleoptera. The percentage by volume of these insect groups changed with seasons. The volume percentage of order Lepidoptera was the highest in early summer (73.97 %) and late summer (51.15); the volume percentage of order Coleoptera was the highest in early autumn (42.03 %) and late autumn (54.03 %). Compared with other seasons, the volume percentage of order Hymenoptera has been improved (15.53 %) in early autumn, and the volume percentage of family Coccinellidae has been improved (12.2 %) in late autumn.6. GIS raster calculator was used to create the possibility map of distribution to predict the potential distribution area of greater horseshoe bat in multiple spatial scales. In landscape scale, we used elevation and distance from the nearest stream to create the potential distribution map. The results of leave-one-out cross-validation showed that the accuracy of the model in landscape scale was 62.5 %. Forest cover, aspect and forest type were used to create the potential distribution map in forest stand scale, and the accuracy of the model was 69 %.