Exploratory sensitivity analysis of a stream ecosystem model


Category:  Range Science
Linked Publication
Language: English
Author(s): Joseph H. Wlosinski
Description: Data collected by the Aquatic Section of the Desert Biome, U. S. International Biological Program, were used to create four data sets, each at a different level of resolution. Each data set was then used in a deterministic model to simulate a stream ecosystem. Each data set included four plant groups, two detrital categories, decomposers, nutrients, one fish species, and macroinvertebrates. The invertebrates were sampled in a manner that would allow simulating the ecosystem at four levels of resolution: 1) invertebrates were measured to size class within 15 taxonomic groups for a total of 37 variables, 2) size classes within taxonomic groups were combined to produce 15 variables, 3) similar taxonomic groups were combined to produce eight functional groups, and 4) the eight functional groups were combined and were simulated as one group. The number of parameters ranged from 399 at the lowest level of resolution to 4,212 at the highest level. Parameters were calibrated using data collected in 1971 and 1972. Data used for validating the model were collected in 1975 and 1976. Validation consisted of comparing model predictions to the 95 percent confidence intervals of over 1,000 measured values. Totals calculated for a number of state variables representing invertebrates at higher levels of resolution were compared with values predicted by models of lower resolution. Of 235 comparisons common to all four levels of resolution, 127, 142, 154, and 142 predicted values were within the confidence intervals of empirical data for the finest through coarsest levels of resolution. Results from a Turing test showed that 88.8 percent of 202 predicted values were found to be acceptable to four experts in the field of stream ecology. It is argued that the fluxes between variables, rather than estimates of the mass of variables, should be used for validating stream ecosystem models.