Ecoforecasting in real time for commercial fisheries: the Atlantic white shrimp as a case study

Research paper by Sara P. Garcia, Lawrence B. DeLancey, Jonas S. Almeida, Robert W. Chapman

Indexed on: 11 Apr '07Published on: 11 Apr '07Published in: Marine Biology


Predictive modeling of natural resources has long relied on mechanistic descriptions incorporating various population attributes and to a lesser extent environmental conditions. A radical departure from this tradition is proposed, advocating the data-driven analysis and forecasting of population cycles from historical records, and using the Atlantic white shrimp, Litopenaeus setiferus, as a case study. The time series data were collected in the Charleston Harbor (32°47′00″N, 79°56′00″W), South Carolina, USA, and from the database of the National Marine Fisheries Service (http://www.st.nmfs.gov/st1/commercial/index.html), for the period between January 1986 and December 2004. Correlations between shrimp population cycles and environmental hydrological parameters were established by phase space reconstruction, a technique central to most nonlinear time series analysis methods. Predictive models of future shrimp population levels were built using feed-forward artificial neural networks, a well-known machine learning technique. From several attempted strategies, predicting the state commercial harvest from the sampling of populations in the Charleston Harbor conducted by the South Carolina Department of Natural Resources proved to be optimal, with an accuracy of 92% for 1-month and 79% for 3-month ahead predictions, as measured by the nonparametric and nonlinear Spearman’s correlation coefficient. In addition, the shrimp population levels seem to be more sensitively to changes in surface water temperature than salinity, but the latter is also an important consideration. These models also suggest that catch-per-unit-effort data are important indicators of commercial harvest and, thus, provide an important linkage between monitoring programs and commercial returns, enabling accurate predictions of natural resources to be made in near real time and extended beyond the critical time frames within which resource managers operate.