More about size-at-age and what determines it (reply to the previous post)
84 views
Skip to first unread message
asta...@gmail.com
May 25, 2015, 10:56:38 PM5/25/15
to atlantis-eco...@googlegroups.com
Hi,
sorry for the late reply but I was sorting out similar issues with fish and mammal growth myself (in the Baltic Sea model). The problem I had was similar to yours - the StN and ReN of some groups was just dropping trhoughout the simulation, yet they were not reported as starving. So here is my solution, which involved a range of things:
1) Increase mum_ and C_ rates ca 10fold. Increasing them separately only seems to achieve a small effect, and that is logical considering the grazing term formula, simplified here as Grazing=C/(1+C/mum). If you play with different values of mum and C, you can see that large effect in Grazing is only achieved when both mum and C are changed. The absolute mum and C values are similar in the model I am using
2) Reduce preference of rebuilding reserves over structure, pR_ parameter. It is now set to 2 for mammals and 3.5 for fish. This will help to reduce ReN/StN ratio when food is low, and starvation will then kick in. Also Kthreshm was increased (at 0.7 now).
3) Increase starvation mortality to make sure they die when food is scarse rather than become so small (for mammals is now set to 0.005, for fish 0.0001).
4) Increase availability of prey. The max availability values for some prey now are at 0.4 or 0.6. One must realise this is the max availability, assuming they overlap spatially, fit into the mouth and when refuge effect is taken into account.
Related to this point, I really encourage you to check the actual predation mortalities. I used the MortPerPred.txt file and my own R script to plot predation mortality of different prey per each predator. I am happy to share the script if anybody wants it. I attach a .pdf file that will show outputs in two different parameter settings (black and red). By plotting this you may realise that even though max availability is high, the predation is still very low, because e.g. there is little spatial overlap.
5) For mammals I also had to increase linear and quadratic mortality, as there is nothing controling them by predation. The values are in the rate 1e-6 to 1e-8. For fish I try to keep these mortalities in the range of 1e-15.
6) Modify gape limitation. I use the smooth feeding window and Kmax_coefft at about 8.5 to 10. The smaller the Kmax value, the 'smoother' the window is, which means that the availability of prey at the upper KUP limit is very low (so you might want to increase KUP).
7) Play with the recruitment parameters. I use BH recruitment (option 3) which makes recruitment dependent on the amount of spawn produced. Perhaps there are too many juveniles comming in? I also have an R script to produce length-at-age, numbers-at-age and biomass-at-age plots for vertebrate functional groups (see output attached) which might be useful to sort out what is going on.
8) Finally it is very useful to plot biomasses per box (from BoxBiomass.txt output). You may find very different things happening in different boxes (again output attached and again happy to share the very simple R code)
Finally, here are some ideas for Beth and Bec :) Would it be sensible to set some upper and lower limits to StN and ReN for some or all functional groups? This is especially true for mammals, as their size-at-age cannot change as much as fish size. If they are not getting enough food to grow to a certain size they should die from starvation rather than get smaller. Isn't this what happens in nature? Just an idea, for possible future developments. Perhaps it is already implemented in some way?
I appreciate comments and ideas from everyone. Surely many of you had to solve similar issues :)
regards,
Asta
Alexander Keth
May 28, 2015, 11:13:28 AM5/28/15
to atlantis-eco...@googlegroups.com
Hey,
here is a small summary of the things I have done in the past days which were inspired by your suggestions.
1. Considering previous settings I have to agree with you here. Increasing MUM and C alone has only limited impact on the model performance. I guess this is due to the fact that either MUM or C becomes a limiting factor when the other parameter is increased. However, it seems that our clearance rates are quite high at the moment therefore I made 4 runs with increasing MUMs (factor 2 (run 780), 5 (run 781), 10 (run 782) and 25 (run 783). I attached the plots of the relative weights. Starvation seems to be minimal using a multiplication factor of 2. However, young fish (especially agecl 2) still drop right at the beginning of the simulation. In addition agecl 1 gains too much weight at the moment. Do you think that the Recruit weights are to high (KWRR_ and KWSR_)? Given the rapid changes in relative weight after roughly 30y using higher multiplication factors (>= 5) those values may be a little to much. Nonetheless, it feels kind of satisfactory to see our fish groups increasing in weight ;)
2. I have not tuned the preference parameter for ReN and StN yet. At the moment we are using 3.5 for all mammals, 5 for the seabirds and 2/4 for the fish groups. Kthresm was set to 0.5 in all our previous runs. We will try some simulations using 0.7 as you suggested.
3. Regarding the starvation mortality we are already using a value of 0.0001 for all groups.
4. Based on a previous simulations with very high MUMs (base values multiplied by 100) and increasing C and avail we were able to reduce the starvation of some groups (runs 535-543 with C and avail multiplied by 1.5, 2, 5, 7.5, 10, 15, 25, 50 and 100. Only relative ReN is attached as StN looks basically the same). Even at moderate increases (factor 7.5) many groups showed no sign of starvation. I checked the predation mortalities in our current version using the MortPerPred-file. For most species the linear and quadratic mortalities are much higher than the predation mortalities. Therefore our model is basically externally controlled and predator-prey interactions only have a minor impact on biomass trajectories. Given the fact, that younger fish are mainly controlled via predation we either have to increase the feeding pressure in our model or decrease lin/quad mort. Unfortunaltey, availabilities are already at high values for most groups with common values between 0.5-0.9.
5. Our current lin/quad mortalities for fish are much higher than 1^10-15.
6. I tested the effect of the gape limitation using the following settings:
KUP was increased from 0.4 to 0.8 (in 0.1 steps) and KLP was increased from 0 to 0.3 (in 0.06 steps) resulting in 30 runs. All groups had the same values. The increase in KUP had almost no impact on biomass. In contrast the model was sensitive to changes in KLP. In additon, results highly differed between our base setting with individual KUPs and the runs where I used the same KUP for every group. This indicates that group specific changes in KUP can have an substantial effect on model results which makes calibration even more complicate :/
7. We are using the hockey-stick recruitment. I changed the carrying capacity (jack_b) for all groups yesterday. Jack_b was increased and decreased for all groups simultaneously. Unfortunantly results were equal for different parameter settings indicating (hopefully) some technical issue.
Cheers
Alex
here is a small summary of the things I have done in the past days which were inspired by your suggestions.
1. Considering previous settings I have to agree with you here. Increasing MUM and C alone has only limited impact on the model performance. I guess this is due to the fact that either MUM or C becomes a limiting factor when the other parameter is increased. However, it seems that our clearance rates are quite high at the moment therefore I made 4 runs with increasing MUMs (factor 2 (run 780), 5 (run 781), 10 (run 782) and 25 (run 783). I attached the plots of the relative weights. Starvation seems to be minimal using a multiplication factor of 2. However, young fish (especially agecl 2) still drop right at the beginning of the simulation. In addition agecl 1 gains too much weight at the moment. Do you think that the Recruit weights are to high (KWRR_ and KWSR_)? Given the rapid changes in relative weight after roughly 30y using higher multiplication factors (>= 5) those values may be a little to much. Nonetheless, it feels kind of satisfactory to see our fish groups increasing in weight ;)
2. I have not tuned the preference parameter for ReN and StN yet. At the moment we are using 3.5 for all mammals, 5 for the seabirds and 2/4 for the fish groups. Kthresm was set to 0.5 in all our previous runs. We will try some simulations using 0.7 as you suggested.
3. Regarding the starvation mortality we are already using a value of 0.0001 for all groups.
4. Based on a previous simulations with very high MUMs (base values multiplied by 100) and increasing C and avail we were able to reduce the starvation of some groups (runs 535-543 with C and avail multiplied by 1.5, 2, 5, 7.5, 10, 15, 25, 50 and 100. Only relative ReN is attached as StN looks basically the same). Even at moderate increases (factor 7.5) many groups showed no sign of starvation. I checked the predation mortalities in our current version using the MortPerPred-file. For most species the linear and quadratic mortalities are much higher than the predation mortalities. Therefore our model is basically externally controlled and predator-prey interactions only have a minor impact on biomass trajectories. Given the fact, that younger fish are mainly controlled via predation we either have to increase the feeding pressure in our model or decrease lin/quad mort. Unfortunaltey, availabilities are already at high values for most groups with common values between 0.5-0.9.
5. Our current lin/quad mortalities for fish are much higher than 1^10-15.
6. I tested the effect of the gape limitation using the following settings:
KUP was increased from 0.4 to 0.8 (in 0.1 steps) and KLP was increased from 0 to 0.3 (in 0.06 steps) resulting in 30 runs. All groups had the same values. The increase in KUP had almost no impact on biomass. In contrast the model was sensitive to changes in KLP. In additon, results highly differed between our base setting with individual KUPs and the runs where I used the same KUP for every group. This indicates that group specific changes in KUP can have an substantial effect on model results which makes calibration even more complicate :/
7. We are using the hockey-stick recruitment. I changed the carrying capacity (jack_b) for all groups yesterday. Jack_b was increased and decreased for all groups simultaneously. Unfortunantly results were equal for different parameter settings indicating (hopefully) some technical issue.
Cheers
Alex
--
You received this message because you are subscribed to the Google Groups "Atlantis ecosystem model" group.
To unsubscribe from this group and stop receiving emails from it, send an email to atlantis-ecosystem...@googlegroups.com.
To post to this group, send email to atlantis-eco...@googlegroups.com.
To view this discussion on the web, visit https://groups.google.com/d/msgid/atlantis-ecosystem-model/31862b18-f409-4c7f-a25c-4e2143f3ecf1%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.
-- <°)))>< ><(((°> <°)))>< ><(((°> <°)))>< ><(((°> Alexander Keth Institute for Hydrobiology and Fisheries Science University of Hamburg Olbersweg 24 22767 Hamburg Germany Tel.: +49 (0)40 428 38 6708 Fax: +49 (0)40 428 38 6618 <°)))>< ><(((°> <°)))>< ><(((°> <°)))>< ><(((°>
beth.fulton
May 28, 2015, 7:50:37 PM5/28/15
to atlantis-eco...@googlegroups.com, alexand...@uni-hamburg.de
G'day,
While the exploration you're doing is good for your understanding - for instance learning that the lower end of the window is particularly important if things are ending up on the small side of things and so effectively "hiding" form predators by staying outside their feeding window. Predation windows are an important trait in reality that's why it proves to be such a central mechanism in Atlantis too.
I would caution though that I still think your issues is still with the number of individuals not the individual ability to consume or grow. While you've made some extra progress (which is good) you want to make sure that the mum etc don't get too high because once you start depleting numbers through perturbation you will find individuals getting too big instead.
In terms of KWRR etc make them the size you want successful settlers to be - so not necessarily the very tiny larval size of brand new settlers who typically don't survive in numbers but the ones that are a little bit larger and can really start to grow (make sure they have access to sufficient invertebrate/plankton prey etc to facilitate that.
Where ever possible keep mQ < M for all but the very top predators or you will in effect end up with multiple simultaneous single species models ;)
How bad are the numbers if you zero out the mL and mQ?
For the jack knife issues is that a code issue my end I need to sort or an oops at your end?
Cheers
Beth
asta...@gmail.com
May 28, 2015, 9:02:45 PM5/28/15
to atlantis-eco...@googlegroups.com, bndf...@gmail.com
Hi,
Beth (of course!) made a very good point, about not increasing mums too much much. I just had the same issue myself - while I thought I sorted mammal sizes and solved the fact that they were getting too small, once more prey was available they started growing far too big. So I had to decrease mums and C again. Indeed regulating numbers through predation is the key.
Alex, the availabilities you mentioned are high, but if linear mortality is high then high availabilities may not help much, as the animals are dying off by themselves instead of being eaten. Also, how many species can predators eat? Do you have high availabilities for only some groups and zero for many others? I suppose predation mortality on some key prey groups could easily be in the range of 0.01-0.1. If all your predation mortalities are low (<0.0001) then it seems that predator-prey interactions are too weak. Try to reduce mL and mQ by some orders of magnitude and you may be amazed how the system responds, and animals suddenly start eating each other :)
I also agree that feeding windows are very important parameters. Especially if you use hard feeding window (which I would not recommend at all), because just a tiny change in size can mean a jump from very high to zero predation. And yes, they are indeed very important in nature, and therefore require careful thought and perhaps some sensitivity analyses. It seems to me that using smooth feeding window makes things a bit less sensitive, as availability does not change in a knife-edge manner, but more slowly.
regards,
Asta
Beth (of course!) made a very good point, about not increasing mums too much much. I just had the same issue myself - while I thought I sorted mammal sizes and solved the fact that they were getting too small, once more prey was available they started growing far too big. So I had to decrease mums and C again. Indeed regulating numbers through predation is the key.
Alex, the availabilities you mentioned are high, but if linear mortality is high then high availabilities may not help much, as the animals are dying off by themselves instead of being eaten. Also, how many species can predators eat? Do you have high availabilities for only some groups and zero for many others? I suppose predation mortality on some key prey groups could easily be in the range of 0.01-0.1. If all your predation mortalities are low (<0.0001) then it seems that predator-prey interactions are too weak. Try to reduce mL and mQ by some orders of magnitude and you may be amazed how the system responds, and animals suddenly start eating each other :)
I also agree that feeding windows are very important parameters. Especially if you use hard feeding window (which I would not recommend at all), because just a tiny change in size can mean a jump from very high to zero predation. And yes, they are indeed very important in nature, and therefore require careful thought and perhaps some sensitivity analyses. It seems to me that using smooth feeding window makes things a bit less sensitive, as availability does not change in a knife-edge manner, but more slowly.
regards,
Asta
Alexander Keth
Jun 8, 2015, 6:35:58 AM6/8/15
to atlantis-eco...@googlegroups.com
Hello guys,
sorry it took so long for me to give you an update.
In order to calibrate our numbers as you suggested we had to rethink our current SSB/R settings. Therefore we had to extract SSB/R data from both single species and multi species assessments (which took ages due to very strange datastructures... ;)). In addition we intend to transition from jackknife to Beverton & Holt recruitment since parameter interpretation is much more simple (carrying capacity depends on both parameters using jackknife.). Therefore, we basically have to start from scratch regarding recruitment. Furthermore I added new plotting routines (inspired by Asta) which give an overview of the mortalities, predation mortalities and SSB/R. All this took quite some time. Hopefully we will be able to start our first scenarios today and share the results with you tomorrow.
Regarding the SSB/R calibrations some questions came up:
1. Where do the first recruits come from? As far as I know they are not forced via the init_file. However, recruits are present right at the beginning of the simulation.
2. Can we analyze the SSB/R relationships using the "output[...]YOY.txt" and "output[...]SSB.txt" files?
3. I scanned the sourcecode for the calculations of SSB and YOY (Young of Year = Recruits). Am I right, that both are reported in t for the whole area?
Cheers
Alex
sorry it took so long for me to give you an update.
In order to calibrate our numbers as you suggested we had to rethink our current SSB/R settings. Therefore we had to extract SSB/R data from both single species and multi species assessments (which took ages due to very strange datastructures... ;)). In addition we intend to transition from jackknife to Beverton & Holt recruitment since parameter interpretation is much more simple (carrying capacity depends on both parameters using jackknife.). Therefore, we basically have to start from scratch regarding recruitment. Furthermore I added new plotting routines (inspired by Asta) which give an overview of the mortalities, predation mortalities and SSB/R. All this took quite some time. Hopefully we will be able to start our first scenarios today and share the results with you tomorrow.
Regarding the SSB/R calibrations some questions came up:
1. Where do the first recruits come from? As far as I know they are not forced via the init_file. However, recruits are present right at the beginning of the simulation.
2. Can we analyze the SSB/R relationships using the "output[...]YOY.txt" and "output[...]SSB.txt" files?
3. I scanned the sourcecode for the calculations of SSB and YOY (Young of Year = Recruits). Am I right, that both are reported in t for the whole area?
Cheers
Alex
--
You received this message because you are subscribed to the Google Groups "Atlantis ecosystem model" group.
To unsubscribe from this group and stop receiving emails from it, send an email to atlantis-ecosystem...@googlegroups.com.
To post to this group, send email to atlantis-eco...@googlegroups.com.
To view this discussion on the web, visit https://groups.google.com/d/msgid/atlantis-ecosystem-model/c3328a74-06ce-4e57-bc66-4ccadbe4ffd7%40googlegroups.com.
For more options, visit https://groups.google.com/d/optout.
Beth....@csiro.au
Jun 29, 2015, 9:24:24 PM6/29/15
to alexand...@uni-hamburg.de, atlantis-eco...@googlegroups.com
G'day,
Again sorry to be slow
1) On start up the model checks to see if there should be "recruits in the water" from spawning the previous calendar year. If yes it does a quick version of theca calculations that happen at spawning using current adults to get the first recruits - its the easiest plug or you get a whole in recruitment.
2) The files should help get an idea of recruitment vs SSB (either that or report strait to a log file if you want it in detail)
3) They report per stock, but as most models have 1 stock per functional group yes it will be for the entire model domain simultaneously
Cheers
Beth
> To view this discussion on the web, visit https://groups.google.com/d/msgid/atlantis-ecosystem-model/5575700B.9000802%40uni-hamburg.de.
Again sorry to be slow
1) On start up the model checks to see if there should be "recruits in the water" from spawning the previous calendar year. If yes it does a quick version of theca calculations that happen at spawning using current adults to get the first recruits - its the easiest plug or you get a whole in recruitment.
2) The files should help get an idea of recruitment vs SSB (either that or report strait to a log file if you want it in detail)
3) They report per stock, but as most models have 1 stock per functional group yes it will be for the entire model domain simultaneously
Cheers
Beth
Reply all
Reply to author
Forward
0 new messages