Predicting anchovy recruitment in the southern Benguela ecosystem: developing an expert system using classification trees

We have developed a simple deterministic expert system to predict the approximate level of recruitment of anchovy in the southern Benguela ecosystem. Classification trees were used to help in constructing the model. Five predictors were incorporated to cover the main areas of the anchovy life cycle (spawning grounds, transport area and the nursery grounds): namely, suitable spawning habitat, gonad atresia, Cape Agulhas atmospheric Pressure (CAP) index, Cape Peninsula Sea Surface Temperature (SST) anomaly (CT4), and Hondeklip Bay SST anomaly (HB4). The CAP index and HB4 anomaly were found to be the most successful variables, highlighting the importance of efficient transport between the spawning grounds and the nursery area, and a productive nursery ground, respectively. The availability of suitable spawning habitat also ranked highly whereas gonad atresia and the CT4 anomaly carried reduced weightings. Recruitment for 16 of the 18 years (89%), including both test years, was correctly classified by the expert system. By providing an objective approach, classification trees are suitable for developing an efficient model and allow relative weighting of variables. We believe that if expert systems are regularly updated they may prove to be aids to management of South African anchovy, and provide a record of why management decisions were made.