During the springs of 1995–1997, we studied birds and landscapes in the Chihuahuan Desert along part of the Texas–Mexico border. Our objectives were to assess bird–landscape relations and their interannual consistency and to identify ways to cope with associated uncertainties that undermine confidence in using such relations in conservation decision processes. Bird distributions were often significantly associated with landscape features, and many bird–landscape models were valid and useful for predictive purposes. Differences in early spring rainfall appeared to influence bird abundance, but there was no evidence that annual differences in bird abundance affected model consistency. Model consistency for richness (42%) was higher than mean model consistency for 26 focal species (mean 30%, range 0–67%), suggesting that relations involving individual species are, on average, more subject to factors that cause variation than are richness–landscape relations. Consistency of bird–landscape relations may be influenced by such factors as plant succession, exotic species invasion, bird species' tolerances for environmental variation, habitat occupancy patterns, and variation in food density or weather. The low model consistency that we observed for most species indicates the high variation in bird–landscape relations that managers and other decision makers may encounter.

The uncertainty of interannual variation in bird–landscape relations can be reduced by using projections of bird distributions from different annual models to determine the likely range of temporal and spatial variation in a species' distribution. Stochastic simulation models can be used to incorporate the uncertainty of random environmental variation into predictions of bird distributions based on bird–landscape relations and to provide probabilistic projections with which managers can weigh the costs and benefits of various decisions. Uncertainty about the true structure of bird–landscape relations (structural uncertainty) can be reduced by ensuring that models meet important statistical assumptions, designing studies with sufficient statistical power, validating the predictive ability of models, and improving model accuracy through continued field sampling and model fitting. Uncertainty associated with sampling variation (partial observability) can be reduced by ensuring that sample sizes are large enough to provide precise estimates of both bird and landscape parameters. By decreasing the uncertainty due to partial observability, managers will improve their ability to reduce structural uncertainty.