Hi Andrew,

I think you might use as a template the sampling protocol for Patagonian toothfish (>50 years, > 150cm long "For the HIMI Patagonian toothfi sh trawl fi shery,

the sampling protocol for age determination

specifi es that a maximum sample size of 10 fi sh for

each 10 mm length bin be sampled for the range of

lengths between 400 and 1 000 mm for a given fi shing

season. For length bins outside this range, the

same restriction on sample size is made, but due

to the rarity of fi sh caught in these length bins, the

maximum sample size is usually not achieved." Candy et al., 2007, CCAMLR Science, 14: 43-66.

What you are suggesting is along the lines of a recent paper -- Coggins Jr, Lewis G., Daniel C. Gwinn, and Micheal S. Allen. "Evaluation of Age–Length Key Sample Sizes Required to Estimate Fish Total Mortality and Growth." Transactions of the American Fisheries Society 142.3 (2013): 832-840. They however suggest 500 animals.

Thanks Brett and Vladimir. The Candy et al. paper is interesting in that there weren't many large fish, our issue is that they are mostly lareg fish and yonger fish are underrepresented. This means that the impact on the von Bert curve is not at the top of the curve, but at the start (computing Lo I would suggest).

Hi Andrew

Is the species recreationally fished? You may be able to get samples by advertising to them to get involved and donate fish frames. It would not be stratified per se if completely voluntary, unless you can work with them to stratify sampling.

Thanks David, I've got the samples already - fishery independent, I just need to decide on how to subsample over 10 000 samples of a long-lived species, with 80% of the samples from larger individuals (>700 mm). Thanks Andrew

Hi Andrew

You might want to do some simulations to see what the impact of different sub-sampling schemes is on your estimates, which will also be affected by the fitting algorithm you choose - you might also want to think about which growth parameters you need to estimate with most confidence. There is also the issue of reader reliability and I guess my hunch would be that reader reliability would be lower for the older fish so again this might suggest more effort, possibly with a significant amount of re-reading or even using two readers if possible, would be needed at the upper end of your size/age distribution.

Clive, thanks for that. Will be using C14 to age the species (Neoceratodus forsteri) as it does not have a calcium based otolith (soft tissue only) and so we are using scales. We have validated the technique and have cross-checked with oxytetracycline marked fish - the oldest so far being 67 years old. So as far as absolute age, the older specimens will have a larger error associated with them compared to the younger ones, however the younger ones are cheaper to analyse and so we might be able to do more samples of the older fish. We need to be mindful of using enough younger fish to ensure that the Lo value is a reasonable estimate. Our age validation publication is at:

https://journals.uair.arizona.edu/index.php/radiocarbon/article/view/3606

Andrew,

It isn't clear to me exactly what your objective is. You said you need to determine the age structure of the species, but typically age structure is estimated for a particular area and/or portion of a population or stock (such as fish over a certain length, or fish vulnerable to a particular gear). Generally there are many orders of magnitude more age-0 and age-1 fish than adult fish, so you have to define the portion of the stock that your estimate applies to. You said your sampling was fishery-independent, but whatever gear you used would be selective and not likely to capture the smallest fish, which are typically most abundant. If you used a number of gears, all with different selectivity, then that selectivity would need to be accounted for in your estimate of age structure. You could subsample randomly from length ranges (as others suggested) and construct a length-age key to estimate the ages of non-aged fish. You would need to weight the age composition for each length interval by the original sample size for each interval. Since the fish are that long-lived, I'm guessing there is a lot of overlap in size at age, and the age-length key might not work any better than drawing a simple random subsample of 300 fish from the sample of 10000+. Hopefully, the age and length frequency distributions of the random subsample should be similar to that of the sample. If there are factors that affect age composition, such as time or area, you will need to stratify your subsample accordingly to make sure it is representative.

Incidentally, sampling from length intervals is good if you are also estimating growth parameters (which it sounds like you are). If your sample has few small fish you may get unrealistic estimates of tzero, which will bias the estimate of K. One possible fix is to sample age-0 fish for average length (L0), then plug that into the alternate vonB function with only two parameters to estimate: Lt = Linf-(Linf-L0)exp(-Kt).

Your validation paper was very interesting - I think it is one of the few I've seen that does bomb radiocarbon validation with scales. Looks like an interesting species and a great project. Wish you well.

Hi Scott, thanks for you response - great feedback!

Some response re your comments -

"Generally there are many orders of magnitude more age-0 and age-1 fish than adult fish, so you have to define the portion of the stock that your estimate applies to." Generally true in a fish species that has annual recruitment, but we have not found any small fish for many years suggesting that recruitment hasn't happened - the size structure is dominated by large (presumably larger fish) so the aim in the first instance is finding some basic age/length information and how old some of our younger fish are so we can see the river conditions when they successfully recruited. As this is a long-lived fish suspect that they are coping with highly variable recruitment over many decades - but this is something we need to confirm.

"If there are factors that affect age composition, such as time or area, you will need to stratify your subsample accordingly to make sure it is representative." Yes, we are looking at sites where we have modes of smaller fish identifiable in the length frequency graphs and will also target these modes to identify how old these fish are and what the river conditions were at these sites to provide the successful recruitment.

Agree we need some age-0 fish so that the curve is tight around a realistic t0 and that will be the challenge and might require us to target a particular set of small individuals.

Thanks again for your comments - really starting from scratch with this species and as this is a vulnerable species we have had to use scales as a non-destructive technique. It has taken many years, but really happy we have found a technique we can use to age this species and fill in some crucial knowledge gaps.

Cheers

Andrew