Predictive models of grazing distribution are used to explore livestock – landscape interactions. However, conventional models based on local foraging rules or regression analyses have limited predictive value at larger scales commonly of interest to land managers. We developed a landscape scale model that matches the spatial scale of decision factors with the predicted distribution pattern.  The model was used to infer how animals integrate landscape factors. Predictive models of grazing suitability were generated for different seasons (Spring and Fall) and cattle breeds (Angus and Criollo) in a 2,425 ha pasture on the Jornada Experimental Range located in the northern Chihuahuan Desert. Models generate raster maps of grazing suitability using an adapted version of the Multi-Criteria Evaluation method from decision theory. Four factors: vegetation, distance from water, shrub density, and elevation were represented by 10 m grid maps. Independent GPS collars grazing positions data acquired at 5-minute intervals were used to calculate decision functions, which describe animal preference for different landscape attributes, and to evaluate how well the model fit the data.

Results/Conclusions

Models with three predictors produced the best fit of the data. Distance to water was the most important factor in all models. Animals showed higher preferences for areas within 4 to 6 km of permanent water depend upon breeds. Other factors were included with lower relative importance. Animal preferences were greater for areas with low shrub density and for low elevation sites where Pleuraphis mutica, Schleropogon brevifolius, and Sporobulus spp. dominate heavier texture soils. Suitable areas delimited by the models realistically represent the observed spatial pattern of grazing. Between 70 to 90 % of grazing positions occurred in cells classified by the model as suitable. Conversely, the number of suitable cells was 5 to 8 times larger than the observed grazed cells. These differences may have occurred because grazing positions correspond to a very low stocking rate (6 cows in a 2,425 ha pasture); under a realistic situation the number of grazed cells is expected to largely increase. According to the models the pasture is characterized by a single or few large grazing patches which includes about 80% of all suitable cells for grazing. This modeling approach can contribute to an understanding of how large animals perceive landscape heterogeneity.