Estimation of Uncertainty and Location-specific Variation in Regional Food Web Interactions
Food-web interactions are commonly aggregated across sampling locations and represented by a single interaction strength for an entire region. This practice implicitly treats interaction strengths as fixed regional constants, even though ecological interactions may vary substantially among locations because of local environmental conditions, species abundances, and habitat structure. We argue that food-web interaction strengths should instead be viewed as spatially distributed latent variables whose regional distributions must be inferred from observations collected at a limited number of sites. To estimate this spatial variation, we develop a Bayesian framework based on transect observations from multiple locations within a region. Species-species interaction measurements at each location are modeled as Poisson random variables whose means are proportional to local interaction strengths. Preliminary analyses suggest that location-dependent interaction strengths may be simply approximated by a uniform distribution on an interval. Under this assumption, we estimate the endpoints of the interval and thereby characterize both the mean interaction strength and its spatial variability. For the case of three sampling locations, we derive graphical representations of the resulting estimates and their uncertainties. Among other results, we show that the average of three observed interaction frequencies systematically underestimates the true regional mean interaction strength under the uniform-distribution model. We apply the framework to avian frugivore-plant interaction data collected at three locations within each of six regions in Ecuador. The estimated distributions of interaction strength are used to test the hypothesis that species within the same genus spatially substitute for one another. The results do not support this hypothesis. Instead, species within the same genus tend to exhibit elevated interaction strengths in the same locations, indicating positive spatial association rather than substitution. We also find evidence that habitat fragmentation may reduce within-region variability in bird-plant interactions even when species diversity is maintained. These results suggest that explicitly modeling spatial variation can reveal ecological patterns that are obscured by conventional food-web representations.