Joe,

Thanks for your interest in MGET. As you probably know, the MGET Fit GAM tool wraps the gam() function of the R mgcv package. gam() has a parameter called weights. MGET does not currently expose this parameter, although it should not be difficult for MGET to do that. But before we discuss that further, let’s discuss what you’re aiming to do a bit more.

I have done quite a bit of density modeling of cetaceans and other large vertebrates by fitting GAMs to line transect survey data. The traditional approach for those models is to split survey transects into segments, tally the number of sightings made on each segment, and also obtain some habitat covariates such as such as SST, depth of the sea floor, etc. For each segment we also estimate the area effectively surveyed (km^2) as 2wL where L is the length of the segment and w is the “effective strip half width” estimated with distance sampling methodology. Then, for the GAM, we use the number of sightings on the segment as the response variable, the habitat stuff as covariates (and/or space and/or time, depending on the model), and use the effective area of the segment as a model offset. Because this is a count model, we use a distribution and link function appropriate for counts or rates. Most of the time, we’ll use either the Tweedie or negative binomial distribution, as our data are usually overdispersed, and a log link function, which is appropriate for those distributions. Because we’re using a log link, we log-transform the offset in the GAM formula as well. So a final model formulation might look like this:

Abundance ~ offset(log(Area)) + s(SST) + s(Depth)

Then, when predicting, we have to supply both the covariates (SST, Depth) and the offset (Area). If we want to estimate density (number of sightings / km^2), we set the offset to 1.

It sounds like your situation might be similar: you want to estimate CPUE, and you have records that have the amount of catch and the number of trawls, along with some covariates. In that case, it seems reasonable to do this:

Catch ~ offset(log(Trawls)) + s(SST) + s(Depth)

Does that make sense? If so, then you can do it using the extant MGET tools (I can help you parameterize them).

Best,

Jason