Fig 1.
The location of farms included in the present study where badger surveys and monitoring work were undertaken.
Table 1.
Description of farm level variables recorded during the present study.
Fig 2.
Histogram displaying the proportion of survey nights when badger visits were recorded on 62 farms (training data) where badger visits occurred.
Table 2.
Badger visits recorded on camera to different areas within the surveyed farms.
Fig 3.
Factors affecting the likelihood of badger visits (top) and the frequency of badger visits at farms where they were present (proportion of nights badgers seen, bottom). Values are average model coefficients (change in log odds) calculated for variables included in the top model set (≤ 6 AIC, Table A in S1 Appendix). Arrows indicate 95% confidence intervals. Model-averaged regression slopes are considered important if they are consistently positive or negative (i.e. their confidence intervals do not span zero). Continuous variables (sett density, cattle capacity, active sett distance and feed stores) were standardised (mean = 0, sd = 0.5) prior to analysis.
Fig 4.
Predicted probability of badgers being present (at least one observation on camera) in farmyards in relation to four farm level variables; cattle capacity, feed stores, badger sett density and distance to nearest active badger sett.
Bold lines represent the marginal predicted probability and dashed lines (or error bars) the standard deviation. Circles summarise the raw data, with the size of the grey point scaled to the number of observations in that group (smallest point = 1, largest point = 45).
Fig 5.
Predicted badger visitation rate (proportion of nights badgers observed) in relation to the maximum cattle capacity on the farm and the distance to the nearest active badger sett.
Bold lines represent the marginal predicted probability and dashed lines (or error bars) the standard deviation.
Fig 6.
Accuracy of model at predicting badger presence/absence at 40 farmyards used as a test data set.
Figure A is the ROC (receiver operator curve), which displays the true positive rate (proportion of farms with badgers present identified) vs the false positive rate (proportion of farms wrongly classified as having badgers present) for a varying cut off value. Figure B displays the percentage of farms correctly identified as having badgers present (sensitivity—grey line), badgers absent (specificity—black line) and total accuracy (dashed line), relative to the cut off used (farms with a predicted probability above this value are classed as having badgers present).
Table 3.
Confusion matrix displaying model predictions for badger presence/absence in farmyards (using a cut-off of 0.25) compared to observed survey results (based on camera sightings) at the 40 test farms where badger activity was monitored for 12 months.
Values for sensitivity (% of farms with badgers present correctly identified) and specificity (% of farms with badgers absent correctly identified), PPV (positive predictive value: % of farms where badgers were predicted as present where badgers were observed) and NPV (negative predictive value: % of farms where badgers were predicted as absent where badgers were not observed).
Fig 7.
Observed rate of badger visits to farmyards (proportion of nights of observation when visits took place) compared to the predicted rate (based on model parameters, Figs 3 and 5) at the 62 training farms analysed to produce the model (Fig A) and at the 40 test farms (Fig B).
Fig 8.
Interactive tool displaying an interface for entering farm characteristics.
Fig 9.
Example output from the interactive tool displaying the percentage risk score relative to other farms and an illustration of what this means using a 10X10 grid.