Validation tests of a spatially explicit habitat effectiveness model for Rocky Mountain elk

We tested the validity of a spatially explicit habitat effectiveness model for Rocky Mountain elk (Cervus elaphus nelsoni). The model scored habitat e ffectiveness based on seasonal changes in the quality, quantity, and availa bility of forage. Seasonal forage potential scores were derived by integrat ing information on existing vegetation, site potential. historic disturbanc es, topography, and roads. The model generated maps of seasonal habitat eff ectiveness that were used to create utilization distributions (UD; i.e., 3- dimensional density estimates). We tested the elk habitat model using telem etry data collected on 5 cow elk sub-herds from 1993 to 1997 in Custer Stat e Park (CSP), South Dakota, USA. We computed fixed kernel UD from elk telem etry data and simulated random UD within the confines of each sub-herd boun dary. The degree of fit between elk UD and model predicted UD (elk-model UD ) and random UD and model predicted UD (random-model UD) was represented by sub-herd, season, and year using the Volume of Intersection test statistic (V.I. Index). There were no differences in V.I. Indices by year for elk-mo del (1993=0.59, 1994=0.54, 1995=0.60, 1996=0.57, 1997=0.57; F-4.70=0.93, P= 0.45) or random-model (1993=0.59, 1994=0.55, 1995=0.59, 1996=0.58, 1997=0.5 9; F-4.70=1.49, P=0.21) UD; thus, V.I. Indices were pooled across years. Tw o-way analysis of variance indicated that elk-model V.I. Indices did not di ffer lh sub-herd (B=0.50, Y=0.58, A=0.56, S=0.58, R=0.63; F-4.12=2.68, P=0. 08), season (Spring=0.55, Summer=0.55, Fall=0.60, Winter=0.58; F-3.12=0.80, P=0.52), or the interaction terms (B Spring=0.48, B Summer=0.52, B Fall=0. 52, B Winter=0.49, Y Spring=0.60, Y Summer=0.52, Y Fall=0.58, Y Winter=0.62 , A Spring=0.47, A Summer=0.58, A Fall=0.64, A Winter=0.54, S Spring=0.54, S Summer=0.49, S Fall=0.64 S Winter=0.65, R Spring=0.70, R Summer 0.64, R F all=0.60, R Winter=0.59; F-12.55=1.68, P=0.10). V.I. Indices for random-mod el UD did not differ by season (Spring=0.8, Summer=0.57, Fall=0.58, Winter= 0.59; F-3.12=0.56, P=0.65) or interaction term (B Spring=0.55, B Summer=0.5 8, B Fall=0.56, B Winter=0.54, Y Spring=0.57, Y Summer=0.53, Y Fall=0.53, Y Winter=0.57, A Spring=0.61, A Summer=0.63, A Fall=0.63, A Winter=0.64, S S pring=0.60, S Summer=0.49, S Fall=0.57, S Winter=0.59, R Spring=0.59, R Sum mer 0.60, R Fall=0.60, R Winter=0.59; F-12.55=1.44, P=0.17); however, diffe rences were noted among sub-herds (B=0.56, Y=0.55, A=0.63, S=0.56, R=0.60; F-4.12=4.48, P=0.02). V.I. Indices for elk-model UD differed from random-mo del UD (F-4.12=4.71, P=0.02); model performance was worse than random (i.e. , lower V.I.Indices) for 2 sub-herds (elk-model sub-herd B=0.50 vs. random- model sub-herd B=0.56 and elk-model sub-herd A=0.56 vs. random-model sub-he rd A=0.63). Lower V.I. Indices were observed for 2 sub-herds that occupied areas recently subjected to large-scale wildfires. For sub-herds not subjec ted to fire effects (he., greater loss of vegetation security cover), the m odel portrayed elk habitat use less consistently, as represented by greater variability (27-42% larger standard errors) in V.I. Indices, during summer . Conversely the model portrayed elk habitat use most consistent for the sa me 3 sub-herds during fall. These results suggest that the elk model perfor med more consistently during fall than any other season for sub-herds not a ffected by fire. We theorize that these seasonal trends in model performanc e were caused by elk use of vegetative security cover during fall, which wa s modeled, and use of topographic barriers for security cover during summer , which was not modeled. Conversely, poor and less consistent model performance was observed for the 2 sub-herds affected by fire, suggesting that topographic security cover w as more important. The elk habitat effectiveness model offers the flexibili ty to incorporate a multitude of habitat factors and, with further refineme nt, may be a useful alternative to elk models that do not incorporate forag e dynamics, topography or variable road effects.