Modeling utilization and density by landscape covariates

Basic ideas

• Default spatial models make simplifying assumptions about animal distribution and density in relation to landscape

  • Density is a Poisson process

  • State space is randomly (e.g., uniformly) utilized

• More realistic and interesting models allow for 

  • Non-homogeneous processes for density

  • Influence of landscape characteristics on utilization of state space (e.g, barriers, more/less attractive  habitat features).

Example - Coyote utilization/ density of Savannah River Site in relation to availability of hog carcasses

• Basic SCR data for coyotes

  • Detections of scats on transects located throughout SRS

  • Scats identified to individual coyote by DNA analysis

  • Detection and "trap" data discretized to 500-m resolution

• Modeling of state space

  • Needs to reflect entire state space (not just where encounters occurred)

  • Handled in secr by a habitat maske

Pre- dumping of carcasses  (Summer, Fall 2014)

Post- dumping of carcasses  (2015-2016; green = dumped, white = removed)

Analysis in SECR

• secr.data<-make.capthist(encounters, detectors, fmt = "trapID")

  • encounters  = 500-m resolution scat detections

  • detectors = 500-m "pixellated" transects

• habitat mask

  • Big enough to enclose SRS boundary

  • Partitioned into Treatment = 0 (carcasses dumped) and Treatment =1 (carcasses removed).

  • Spatial tools in R (Packages rgdal and sp required)

    • Read in shapefile containing treatment attribute

      • boundary <- readOGR(dsn)

    • Incorporate covariates into mask

      • mask.h<-make.mask(traps,buffer=8000,poly=bound)

      • mask.h<-addCovariates(mask.h,boundary)

• Analysis

  • Pre- and post- treatment comparison of models having and lacking treatment effect

  • without<-secr.fit(secr.data,mask,detectfn = "HN",binomN=0, method = 'BFGS')

  • with<-secr.fit(secr.data,model=list(D~Treatment),mask=mask.h,detectfn = "HN",binomN=0, method = 'BFGS',start=fit0)

Results