ASSESSING THE DATA REQUIREMENTS OF SPATIALLY EXPLICIT DISPERSAL MODELS

We evaluated the consequences of parameter errors for predictions of s patially explicit population models. We examined a simple model for or ganisms dispersing in a fragmented landscape and assessed how errors i n three model input parameters propagate into errors in model predicti ons: (1) misclassification of habitat suitability (landscape error); ( 2) incorrect estimation of how far a disperser can travel (mobility er ror); and (3) incorrect estimation of the mortality rate of dispersers (dispersal-mortality error). The two-dimensional landscape through wh ich organisms dispersed was filled with patches of various shapes (squ are, linear, and elbow) and sizes (4, 9, and 16 cells), and we allowed the overall proportion of suitable habitat in the landscape (2, 8, 16 , and 24%) to vary among runs. A single run consisted of 400 individua ls dispersing through the landscape until they found suitable habitat patches, and the output was a frequency distribution of the number of steps taken before a patch was found (n = 400 individuals). In the err or-free model, dispersal success increased with the percentage of the landscape that was composed of suitable habitat and was greater in lan dscapes filled with more small patches than in those with fewer large patches. Errors in dispersal-mortality parameters resulted in the grea test prediction errors (25-90%), followed by mobility errors (2-60%) a nd landscape errors (<1-17%). In general, prediction errors were highe r in landscapes with a lower percentage of suitable habitat, precisely the type of habitat characterizing most species of conservation conce rn. Our results point to the need for better empirical estimates of er rors in dispersal parameters. In addition, our results suggest that le ss detailed models would improve the match between the complexity of t he model and the quality of available data.