Data from: Elephant pathway use in a human-dominated landscape
Data files
May 22, 2024 version files 123.97 KB
-
conflict.csv
-
Data_analysis_1.csv
-
rain.csv
-
Rcode.txt
-
README.md
-
SURVEY_A_results.csv
-
SURVEY_B_results.csv
Abstract
Habitat loss and fragmentation are one of the biggest threats facing wildlife today. Understanding the role of wildlife pathways in connecting resource areas is key to maintain landscape connectivity, reduce the impacts of habitat loss and help address human-wildlife conflict. In this study, we used sign surveys and camera trapping to understand the fine scale movement of elephants moving between a protected area and agricultural zone in the Masai Mara, Kenya. We used Generalised Linear Models to determine factors driving high frequency of pathway use by elephants. Our results showed strong seasonal trends in pathway use, with peaks coinciding with the dry season. However, no correlations between rainfall and pathway use were found. Temporal patterns of pathway use indicate that elephants use risk avoidance strategies by moving between the two areas at times of low human disturbance. Spatial analysis revealed that the most frequently used pathways were closer to farms, saltlicks and forest and those that had a higher percentage of forest cover. Our models also showed a positive relationship between pathway use and the number of elephant crop raiding incidents, highlighting that pathways can play a role in human-elephant conflict. As habitat loss continues, pathways may become more important for linking resources. However, they are also likely to facilitate movement into farmland. The results from this study provide an opportunity for planned management activities to ensure connectivity and to mitigated conflict.
README: Elephant pathway use in a human-dominated landscape
https://doi.org/10.5061/dryad.ns1rn8q20
Data includes the final clean Excel sheets containing all the variable data that was imported into R for analysis. This data was used for Spearman’s Rank Correlation tests, a linear model and descriptive statistics.
Description of the data and file structure
The files 'SURVEY A_results' and 'SURVEY B_results' are Excel spreadsheets with a summary of the camera trap images from the pathways. Each row is one camera trap image with the processed data of the date, time, photo label, elephant group type, number of elephants and whether the elephants were traveling up or down the pathway.
The file 'Data_Analysis_1' is an Excel spreadsheet that has all the data used in the papers models. This dataset has the different pathway use variables that were tested. For example, distance to farmland, slope etc.
The file 'conflict' is an Excel spreadsheet with a summary of the number of human-elephant conflict incidents each month during the study period and the number of dung piles recorded on the pathways.
The file 'rain' is an Excel spreadsheet with a summary of monthly average rainfall and number of elephant dung piles recorded.
Code/Software
All R Code used in this study can be found in the file 'Rcode'.
Methods
We identified active pathways along the escarpment with the assistance of local rangers and farmers (Figure 2). We assumed pathways were in use if the path was devoid of vegetation (Blake and Inkamba-Nkulu, 2004), marked with elephant dung or footprints and showed signs of elephant browsing on the bordering vegetation (Von Gerhardt et al., 2014). Pathways that did not show any of these signs were not included in this study. We then mapped each pathway using a Garmin Etrek30 Global Positioning System (GPS). The GPS track was taken from the bottom of the escarpment on the border of the Masai Mara to the top of the escarpment. The end of the pathway was determined by the point at which the pathway widened and became open habitat. Habitat type was also recorded on each pathway using a classification system from Kindt et al., (2011). As each pathway went through a number of different habitats, we used a GPS to record the co-ordinate at which there was a change in habitat type. To determine seasonal pathway use, we conducted bi-weekly elephant dung surveys on each pathway from September 2014 to August 2015. During these surveys, on each pathway we counted dung piles along two predefined transects (one going down the pathway and one going down the pathway) to ensure we covered the pathway. Each dung pile represents a single individual. Dung was removed after each count to avoid recounting.
To determine temporal patterns of pathway use and elephant group type using the pathways, we placed 32 heat and motion camera traps (Bushnell Trophy Cam HD 2013) on 14 pathways during two sampling periods: September 2014 – October 2014 and February 2015 – August 2015 (Figure 2). We were unable to place cameras on all the pathways due to limited camera availability and the unsuitability of some pathways for camera trap placement; i.e. some pathways were too wide or too open to place camera traps. To ensure elephants were captured on the 14 chosen pathways, we placed cameras on the narrowest part of the pathway or sections where we knew elephants would cross (e.g. by small water bodies). To obtain suitable photographs of elephants for group type identification, the camera traps were mounted on trees or erected posts at varying heights between 1 – 3 metres depending on the pathway slope. The height of the camera > 1 metre was to ensure the best capture of the head, pinnae, and tusks of elephants (Smit et al., 2017). Each pathway had at least one camera facing up the escarpment and one facing down the escarpment to capture the frontal area of the elephant to aid identification of the elephants. If pathways split, two cameras were placed on each sub-pathway to ensure elephants were captured. Cameras were set at a 5-second trigger interval with three colour images taken per trigger event. At night time the camera used infrared and had the same trigger speed. We downloaded the images from memory cards and changed the camera batteries every three weeks.
We then created a database of the camera trap images and recorded the site (pathway name), the position of the camera trap (up or down), the type of photo (e.g. wildlife, people or false trigger), the wildlife species in the photo and the date and time the image was taken. Specifically, for the elephant images, we recorded additional information including: (1) the direction in which the elephant was travelling; (2) the number of elephants in the image and; (3) the group type. Group type was classified as (i) bull groups, (ii) female-led family groups or (ii) family + bull groups. Group type was determined by sexing elephants based on their genitalia (if visible), body size, shape of their head and length and configuration of tusk size (Moss, 1996). During the period in which an elephant group crossed a camera, depending on the size of the group, many images were captured. Thus, to avoid double counting elephant groups, we developed a Python script to select images from our database that were taken more than 15 minutes apart. This time marker was determined after reviewing all the images and calculating the average time between each independent group. Group type was then determined by reviewing the series of images within the 15-minute time frame.
To understand the relationship between pathway use and crop raiding, we collected data on elephant incursions into agricultural fields in the Trans Mara District from September 2014 to September 2015 (see Tiller et al., 2021).
Analysing patterns of pathway usage
All the data analysis was carried out using the statistical software R (R Core Team, 2016). We assessed the seasonal patterns of pathway use by totalling the number of dung piles counted across all pathways for each month and averaging rainfall readings from weather stations across the Trans Mara and Masai Mara National Reserve. We then carried out a Spearman’s Rank Correlation test with dung counts and rainfall. Due to the potentially delayed effects of rainfall on the ripening of crops and greening of vegetation, we also ran correlations between dung counts and rainfall from the previous month.
To look at the temporal patterns of elephant groups travelling up the pathways into the Trans Mara and down the pathways into the Masai Mara, we sorted the camera trap data into time and direction. Images were grouped into time stamps of 24 one-hour intervals so that we had a frequency distribution of each elephants travelling up and elephants travelling down the pathways.
Analysing the factors determining pathway use
To determine whether high pathway activity predicted high human-elephant conflict, we fitted a linear model based on the number of crop raiding incidents per month as our response variable and the total number of dung piles across all pathways as our predictor variable.
We used descriptive statistics to summarise the number of elephant detections on each pathway. To investigate the factors driving high elephant pathway use, we used four predictor variables: distance of pathway to nearest farm, continuous forest outside the pathway, saltlicks and the percentage of forest cover along pathways. To measure the percentage of forest on each pathway, we calculated the length of the pathway and then used the GPS co-ordinates from the habitat survey to work out the proportion of the pathway that we had classified as forest. We mapped farmland and forest cover and locations of saltlicks by on-screen digitising of 5 m CNES/Astrium satellite imagery from 2015 using QGIS (QGIS Development Team, 2015). We then calculated the distance from the end of each pathway to the nearest farm, continuous forest in the Trans Mara and saltlicks using the Generate Near Table function in ArcMap 10.4.
We then carried out exploratory analysis including graphical inspection, correlation matrices and bivariate tests. Variance inflation factors (VIFs) and Pearson correlation coefficient (r) were used to test for collinearity amongst the predictor variables and we found no evidence of collinearity between our predictor variables (VIF < 5; r < 0.7; Dormann et al., 2013). Exploratory modelling identified persistent over dispersion between the response and predictor variables. Thus, to overcome this problem, we fitted a GLM with a negative binomial error structure and all predictor variables were scaled to have a mean of zero and a standard deviation of 0.5 (Gelman, 2008). For model selection, we used a model averaging approach (Burnham and Anderson, 2002) using the MuMin package (Barton, 2016), which examines average parameter estimates, standard errors and confidence intervals of the predictor variables. We constructed a global model containing all predictor variables and applied the Dredge function to produce a model set containing all possible model permutations. Models were then ranked based on their AICc (Akaike Information Criterion) score, where the lowest score signifies the most parsimonious model, and calculated the delta AICc (the difference in AICc between each model and the best preforming model). We then averaged parameter estimates for models with a delta AICc <2, as this suggests a similar level of support among models (Burnham and Anderson, 2002). The relative importance (RI) of explanatory variables was then calculated by summing the Akaike weights across all models in which the variable was present, resulting in an estimate of probability that the variable of interest features in the best model. Finally, we applied a goodness of fit test to the model set to determine if the models fitted the data well.