Predicting the structure of larval fish assemblages by a hierarchical classification of meteorological and water column forcing factors

The first step in building predictive models of larval fish assemblages is to identify the main environmental parameters which influence their spatial and temporal structure. In this study, multivariate regression trees (MRT) were used to classify hierarchically the effects of large-scale meteorological factors and small-scale water column factors on pre-settlement larval fish assemblages at two sites in the lagoon at New Caledonia, southwest Pacific. The environmental conditions at one site were highly variable spatially and temporally, but varied little at the other. In spite of these differences, MRT models revealed that identical forcing factors influenced the structure of larval fish assemblages at both sites, with a similar hierarchy, but a different statistical efficiency. At a large spatial scale, the seasonal variabilities in sun hours and wind (speed and/or direction) explained 14% and 64% of the structure of larval fish assemblages at the sites of high and low variability, respectively. At a small spatial scale, the seasonal variability in mean surface water temperature, followed by the concentration in Chl a, explained 22% and 62% of the structure of assemblages at the sites of high and low variability, respectively. The Dufrane-Legendre index matched characteristic families of larvae to each set of environmental conditions, and illustrated the role of sheltered, Chl a enriched, coastal waters in producing a families-rich assemblage of fish larvae, some species of which are targeted by fishing. This study shows that it may be possible to use environmental data, and predictions computed from MRT to design spatially explicit models of larval fish distribution in coral-reef lagoons