Two-stage boosted regression tree model to characterize southern flounder distribution in Texas estuaries at varying population sizes

Abstract

Linking trends in fish population abundance to environmental characteristics is often difficult because fish use a variety of habitats throughout their ontogeny and may exhibit large interannual fluctuations in abundance. We developed a two-stage boosted regression tree model to investigate spatiotemporal patterns of Southern Flounder Paralichthys lethostigma abundance and distribution in Texas estuaries. We used a 36-year fishery-independent data set (1977-2012) to correlate distribution with environmental conditions and seasonal or long-term changes in abundance. Adult Southern Flounder were sampled with gill nets using a random-stratified design. Predictions of abundance were made to grids of environmental data to identify “hot spots” as well as seasonal or decadal shifts in distribution. Models were fit using cross validation, and variance was estimated using nonparametric bootstrapping. Depth, temperature, distance to a tidal inlet, and salinity were the primary environmental determinants of Southern Flounder distribution and abundance. Because distribution and response to environmental conditions can depend on population abundance, we also developed a standardized index of annual abundance using the same two-stage boosted regression tree model. The index identifies a long-term decline in abundance punctuated by recent recovery (possibly in response to management actions). Mapped results identify a coastwide decline in abundance between 1980-1984 and 2005-2009 but the magnitude varied substantially, suggesting disproportionate changes in abundance across the study area.