Multi-scale Wild Boar Damage Prediction In The Key Area Of Siberian Tiger Distribution In Jilin Province

We identified the core distribution region of Siberian tiger(Panthera tigris altaica)in Jilin province based on the information collecting from snow survey,camera trap and wildlife damage compensation.Then we collated the wildlife compensation documents in the core tiger region and analyzed the type,intensity,frequency and spatiotemporal pattern of human-wild boar conflict.We predicted wild boar damage risk using Maximum Entropy Model,explored the efficiency of different model optimization methods via sample bias correction and model complexity tuning,and evaluated the power of several model testing parameters.We identified wild board damage risk with the optimal model,then analyzed the response pattern of environment variables and the scale effect in resources selection.We established general linear model to evaluate the relationship between wild boar damage cost and 3 variables:damage probability,population abundance of wild boar and predation pressure.At last,we proposed management suggestions of wild boar damage administration based on tiger distribution.Our research clarified that Dalongling is the core region of tiger distribution in Jilin province,where possessed 92.3% snow survey information,97.93% camera trap events and 95.8% damage records of tiger respectively.Tiger information collected by snow survey was much lower than camera trap.Tiger damage cases could effectively provide supportive data in monitoring gaps.Corn damaged represented 97.82% of crops damage cases caused by wild boar,and 46.65% of the total compensation cost.Wild boar showed positive selection for corn,negative for soybean and severely avoidance for rice.The annual variance of wild boar damage cases was inconsistent with population dynamic during 2014 to 2017.The raise on population abundance did not necessarily mean the increasing on wild boar damage.Both sample bias correction and model complexity tuning were effective in controling overfitting.The Regularization Multiplier of optimal model was higher than default setting and it was helpful to promote model performance using model complexity tuning and medium-level spatial filtering.AUCtrain was a worse parameter in model evaluation.Affected by pseudo-absent assumption,AUCtest decreased in the process of reducing overfitting.Both omission and commission error should be taking into account when the model was evaluated under MSS threshold.The optimal model was established with 1000m spatial filter,LQ Feature and 2 RM.Discriminated with MSStest threshold(0.29),The high damage risk region accounted for 33.1%of the total farmland area and model omission rate was 0.1679.Forest proportion(43.09%),forest edge distance(17.26%),road density(13.55%),mixed forest proportion(4.73%),slope(4.11%)and river density(3.46%)contributed most for the model.Forest proportion and mixed forest proportion were positive variables for damage risk,while forest edge distance,road density and slope were negative factors.Environment resources measured with home range scale contributed more influence for damage risk.The same resource might significantly change its response to wild boar damage probability when evaluating in different scale.Wild boar damage was significantly related with damage probability,population abundance and predation pressure in general linear model.Damage cost was positively related with damage probability,negatively with predation pressure,and unstablely responsed to population abundance.In low damage year,population density was a negative variable for damage cost,while became positive in general damage year.