Performance comparison between spatial interpolation and GLM/GAM in estimating relative abundance indices through a simulation study

Generalized linear models (GLMs) and generalized additive models (GAMs) are commonly used to standardize catch rates as relative abundance indices in fisheries stock assessments. Spatial interpolation (SI) is an alternative way to estimate relative abundance indices but there have been no comparisons of the effectiveness of the two types of approaches. In the present study, the performances of GLMs, GAMs and SI were compared through a simulation study based on fishery independent surveys of yellow perch in Lake Erie in 1990, 1991, 1992, 2000, 2001, and 2003. Simulated scenarios were tested with sample sizes of 60, 120 and 180 drawn randomly from the survey data, and random errors variances of 0.5, 1 and 2 × the “true” estimate variances. For each combination of sample size and error, 100 simulations were calculated to estimate correlation between the “true” abundance and the estimated relative abundance indices from GLMs, GAMs and SI. The performances of all three methods improved with increasing sample sizes, but worsened with increasing magnitude of the simulated errors. SI performed better than GLMs and GAMs when the simulated errors were low, but SI was more sensitive than GLMs and GAMs to the magnitude of the simulated random errors. When simulated sampling covered the survey area incompletely, GLMs and GAMs performed better than SI.