PARENT SESSION
Oral Session #26: Aquatic Ecology: Fish - Populations, life history, autecology. Presiding: M. Fox.
Tuesday, August 7, 2001. 8:00 AM to 12:00 PM. Hall of Ideas H.

How much of nothing is enough? Problems with zero abundance data in fish habitat models.

NIBBELINK, NATHAN¹, RAHEL, FRANK¹, ¹

ABSTRACT- Models of animal abundance are influenced by zero abundance data points. At the edge of the species geographic range, zeros may have a significant impact on model fit. We tested effects of zero abundance data on predictive models of brook trout abundance at the edge of their range. Mean July air temperature and stream size were used as predictor variables. Monte-Carlo methods were used to simulate baseline data and to add zeros near the edge of the species range and at four increasing distances outside the range. Linear regressions were performed for each of 100 simulations and six models. Simulations revealed that zero values near the edge of the temperature range increased adjusted r2 by up to 10%. The slope and intercept remained unchanged. Zeros had no effect on r2 in the stream size models. Brook trout abundance dropped off more abruptly with temperature so the zeros had more leverage in the temperature models. Zeros added at increasing distances outside the species range for both variables changed the slope and intercept by an average of 8% and 21%, respectively, but had little affect on r2. Therefore we suggest a two-stage approach (e.g. logistic then linear regression) should always be used when modeling species abundance near the edge of its range, thereby retaining valuable information from zero values, but preventing over-fitting or biased interpretations of coefficients.

KEY WORDS: animal abundance, linear regression, brook trout