years of each harmonic regression model and the forecast for 1982-83 are 

 shown for clarity. 



In contrast, models built from the trawl monitoring program data 

 accounted for more than half of the variation at only the NR station 

 (Fig. 23). The forecast percent error for this particular model was 



9.0- 



6.0- 



3 0- 



0.0- 



-3 0- _______ 



I ' I — - 



OCT DEC 

 1981 



FEB APR 



— 1 ' 1 — 



JUN AUG 

 1982 



OCT DEC FEB 



— I ' 1 — ■■ r 



APR JUN AUG 

 1983 



Figure 23. Harmonic regression forecast ( ) and 95% confidence 



limits ( ) for winter flounder at trawl station NR. 



Data from 1976-82 were used in building the model (O) 

 and compared to data from 1982-83 (*) . 



very high; the seasonal pattern was modeled adequately but the magnitude 

 of the catches was underestimated considerably for 1983. It remained 

 unclear whether the inadequacies of the trawl models were due to 

 inappropriate choices for the models, insufficient data, or the inherent 

 variability of the trawl data. Except for the NB station, models 

 developed from these data had terms including 72- and 12-mo periods as 

 well as shorter time intervals. The presence of 12-mo and shorter terms 

 in the models were not unreasonable and represented the yearly cycle of 

 abundance caused by patterns of movement and recruitment of juveniles 

 into the trawl catch. However, the 72-mo long-term harmonics may have 

 been an artifact of the data base and additional years of study will 

 probably be necessary to improve the models if the winter flounder has a 



54 



