Predictive models of northern anchovy Engraulis mordax and Pacific sardine Sardinops sagax spawning habitat in the California Current

Abstract

We modeled presence of anchovy and sardine eggs in the California Current from 1984 to 2009 using predictors related to water mass properties and primary productivity.Spawning areas could be predicted with good discrimination for anchovy, as indicated by logistic models with an area under receiver-operating characteristic curve (AUC; i.e. probability of concordance) of 0.86, and acceptable discrimination for sardine, as indicated by an AUC of 0.79.Knowledge of 3-dimensional characteristics of the water column was particularly important for predicting anchovy spawning habitat, as indicated by 2 variables: depth at which maximum chlorophyll a (chl a) occurred and an index of geostrophic flow.These variables were not useful predictors in the sardine model.For both species, variables related to low productivity limited modeled spawning probability most frequently during the 1980s and 1990s, but low temperature was most frequently limiting during the 2000s.Egg abundance was extremely patchy for both species; models to predict numbers of eggs conditional on their presence had almost no additional predictive ability for either species.However, most eggs occurred in areas where predicted probabilities of egg presence were relatively high.About 80% of anchovy eggs were captured in 40% of the samples that had modeled capture probabilities ≥0.25.About 80% of sardine eggs were captured in 35% of the samples that had predicted probabilities of capture ≥0.25.These models may be useful for understanding interannual changes in populations of predators that depend on anchovies and sardine.They could also improve the precision of fisheries surveys if predictors were collected beforehand using unmanned gliders.