Porch: Estimating velocity and diffusion from tagging data 



703 



o 

 O 



Sample size 

 Figure 2 



Coefficients of error of the estimates for the parameters of the sinusoidal model. The graphs 

 on the left give the coefficients of error (CE's) under weak diffusivity and the graphs on the 

 right give the CE's under strong diffusivity. The three curves in each graph correspond to 

 zone B recovery probabilities of 0.0 (squares), 0.1 (triangles), and 1.0 (crosses). 



model vary (the lowest valley presumably occurring in 

 the vicinity of the true values of the parameters), sev- 

 eral of the valleys may be deep enough such that the 

 estimation routine finds a false global minimum. 



The convergence problem was eliminated when the 

 search was confined to a relatively restricted range 

 of periodicities and was supplied with good initial 

 guesses. In practice, adequate initial guesses and 

 relatively narrow ranges can usually be deduced even 

 from anecdotal data, therefore this should not prove 

 too serious a limitation to the method. In cases where 

 the periodicity is totally unknown, one should search 

 the entire feasible domain with as fine a resolution 

 as possible. 



Piece-wise discrete model 



The coefficients of error associated with each para- 

 meter were, for the most part, very low when the 

 diffusivity was low (Fig. 4). However, the estimates 

 pertaining to area 2 were highly biased and impre- 

 cise when the probability of recovering a tag was 0.0 

 in zone B. This is not unexpected given that exclud- 

 ing recoveries in zone B (x>0 km) all but eliminates 

 the possibility that any of the recovered tags would 

 ever have encountered area 2 (x> 1,000 km). Thus, 

 there is essentially no information on the advection 

 in area 2 and the estimation routine fails. A similar 

 result was not observed when the recovery probabil- 



