Multi-area model

Forums

RSS (Opens New Window)

Mark Maunder, modified 5 Years ago.

Multi-area model

Youngling Posts: 6 Join Date: 6/17/16 Recent Posts

Has anyone successfully use SS using multiple areas. We have tried a couple of times and and had the same type of issue where the biomass gets estimated very high. One diagnostic we tried did not help. We run the two-area model with no movement and one population went very high. We then rerun the model as a single area for each area separately by setting the catch for the other area to zero, setting the lambda to zero for the other areas data, and turned the parameters associated with the other area off. We got reasonable biomass estimated. One problem with SS is that it does not allow area differences in the Rinit parameter and the initial recruitment deviates (through temporal variation in the recruitment distribution parameter). So based on a suggestion by Rick, we extended the time period back 17 years and set the catch to the average level over the first few years of the original model time period. Since there is no data during this extended period it allows more flexibility due to recruitment deviates and the recruitment distribution parameter temporal deviates. However, we still got one of the populations estimated with high biomass.

The only spatial SS assessment we know of is the Canary assessment. Has anyone else used the area functionality of SS and did you get the same type of issue?

Thanks,

Mark

Ian Taylor, modified 5 Years ago.

RE: Multi-area model

Youngling Posts: 117 Join Date: 12/8/14 Recent Posts

In addition to Canary, the U.S. West Coast Yelloweye Rockfish assessment in 2009, updated in 2011 (available at http://www.pcouncil.org/wp-content/uploads/Yelloweye_2011_Assessment_Update.pdf) had 3 areas and seemed to produce estimates of area-specific biomass that were on a similar scale as the amount of area within a typical depth range for each area. That model didn't make any assumptions about catchability, but I assume that any prior assumptions on catchability will anchor the scale of the two areas.

Rishi Sharma and Adam Langley have worked on spatial models for Indian Ocean tunas using SS but I'm not sure that they behaved well enough to be used for management. Andre Punt has been tracking down spatial assessments (not necessarily using SS) so he may have additional models to add to the list.

Melissa Haltuch recently was exploring a two-area model in SS that also included different growth patterns for each area. The combination of growth patterns and areas does not seem to have been explored and tested as well as spatial models with homogenous growth, and Melissa faced a lot of problems and decided to pursue separate models.

If you haven't already done so, you might consider profiling over the parameters controlling the recruitment distribution to see what data sources are driving the out-of-balance estimates that you're getting.

Adam Langley, modified 5 Years ago.

RE: Multi-area model

Youngling Posts: 4 Join Date: 6/22/16 Recent Posts

As Ian mentioned, I have implemented a number of multi region models for the Indian Ocean tuna fisheries. I have also been using SS to configure multi region models for the assessment of a New Zealand Nemadactylus species (tarakihi).

My experience with these models is quite variable though. Unless the models incorporate some very informative data (esp abundance indices with considerable contrast and/or large tag release/recovery data sets), the models may perform as described i.e., estimating a large (unrealistic) biomass in one (or more) of the regions. In these cases it may be necessary to impose a range of constraints on the model such as sharing catchability qs and selectivity between regions for fisheries with comparable indices (e.g. method specific fishery CPUE). It may also be necessary to constrain the distribution of recruitment among the regions and the age of first movement to avoid the model estimating large cryptic biomasses in specific regions.

Some of the regional parameters are often highly correlated with other key parameters esp movement, recruitment distribution and selectivity. Again, it is very important to have informative data to estimate these parameters. Not just abundance but very good age composition data too. I always run parallel single region models for both the main individual regions of the model and for the entire model domain (removing the regional structure). This helps diagnose issues that may emerge in the spatially stratified models.

Adam