Field testing and enhancement of the diet-selection algorithm of POPMIX: A decision support system for multispecies stocking of rangeland

The reliability of predictions of diet composition from POPMIX, a decision support system for multispecies stocking of rangeland, was tested in post oak savanna in east central Texas. Diet samples were collected from esophageal-fistulated goats and cattle during 20-minute grazing bouts within 18 predefined grazing paths (50m x 3m) during each of 5 sampling periods.;Similarity between observed and predicted diets was highly variable for both cattle (22.2-96.9%) and goats (13.9-94.1%). Predictability declined with both declining availability of preferred forage species and increasing diet diversity. POPMIX consistently overpredicted the dietary proportions of preferred species and underpredicted the dietary proportions of undesirable species, particularly with cattle diets.;The grazing path results confirmed that forage species follow three broad patterns of selection. However, the species in the intermediate or desirable category were associated with variation in selection from preference to rejection across the grazing paths.;In POPMIX, proportions of each undesirable and desirable species in the diet are estimated via positive exponential and 1:1 relationships, respectively, with the corresponding field proportions. The field data gave little support for either the form or quantitative specification of these relationships. Expressing diet and field compositions on the basis of selection categories improved potential for development of predictive relationships.;A modified diet selection algorithm was designed in which predictions were a function of the field abundances of the preferred and undesirable selection categories. Eight sets of equations were evaluated in this algorithm with both the original grazing path diet data and three independent data sets. The modified algorithm reduced systematic bias in predictions. Testing with independent data sets indicated that the modified algorithm can adequately predict diet composition at the selection category level under most conditions. However, predictability was compromised by either highly selective use of the preferred category when in very low field abundance or by disproportionate use of species within a selection category. Although further testing is required, the modified algorithm should form the basis of a useful decision support system for multispecies stocking of rangeland.