Spatio-Temporal Habitat Models Using Boosted Regression Trees and Artificial Neural Networks

Southern flounder, Paralichthys lethostigma, is an important multi-million dollar commercial and recreational fishery, yet despite their economic and ecological importance, we have failed to manage a sustainable fishery for this species.  Due to declines, it is important for resource managers to understand and predict the future status of juvenile southern flounder. The main objectives of this study were to provide information needed for the fishery management plan of southern flounder by using statistical modeling techniques to understand how environmental factors influence the temporal and spatial patterns of juvenile southern flounder and to compare a relatively new modeling technique (Boosted Regression Trees; BRT) with a well accepted technique (Artificial Neural Network; ANN). Data were acquired from the Resource and Sport Harvest Monitoring Program conducted by Texas Parks and Wildlife Department.  BRT indicated juvenile southern flounder were associated with low temperatures, low salinity levels, and high dissolved oxygen.  Both spatio-temporal models consisted of high predictive performance with slight spatial differences. Both models suggest high probability of occurrence in Galveston Bay and East Matagorda Bay where as the ANN also indicated high probability of occurrence in Sabine Lake.  Our results provide valuable tools for fisheries managers to enhance management and ensure sustainability fisheries.  The results also identified a predictive framework for proactive approaches to ecosystem management.  These models will allow managers to more accurately conserve nursery habitats for the southern flounder fishery, by conserving appropriate habitat and understanding relationships between abiotic and biotic factors within those habitats.