Using mixed models with the Fit output

[kyanan...@gmail.com]

Hi Chris,

I've been thoroughly enjoying the ctmm package since the Animove workshop in DC. I'm working with a dataset of tortoises which use farmland seasonally. I have used a Krige to make occurrence estimates of tortoises using farms and then extracted how many farms are within each occurrence estimate. I would also like to do some statistics on how much space the tortoises are using in farms and whether that has a relationship with things like sex, species, size etc., but I know that occurrence estimates are not the right tool for estimating space use. Some tortoises show range residency behaviour in farms but other do not and there is a big variation of Ne so there is only a subset of tortoises that akde seems appropriate for and I'm not sure I need to estimate home range area, just some sort of area of use. What exactly is the CIAreaML estimate? Is this for total area used by tortoises and only after doing the akde do you get home range i.e. the important area which doesn't include those exploratory bouts outside the core area they need?

Is it reasonable to do something like a linear mixed model (I have multiple seasons per tortoise) and test the ML Area estimate from the movement models generated in the Fit step with a weights argument for NeArea (e.g. null model<-lmer(CIAreaML~1 + (1|Tortoise), weights=NeArea,data=space,REML=F)) against other variables like sex, size etc.? If the area estimate is not yet through the home range filtering in the movement model and it is not technically home range area, how do I refer to it, as range area?

I have also just got new solar panel E-obs store on board tags. A calibration dataset was made a few years ago for an earlier non-solar version of the tags and I am going to make a calibration dataset for these new tags. I am planning to leave the tag in an open area, a wooded area and then under my house (tortoises sometimes go under platforms and dense shrubs) each for a few days at a time. Do you have any recommendations for making calibration data? I've set the tags to record once per hour all day. I'm also happy to share these data with you if it is useful for future ctmm versions.

Many thanks,

Kyana 

[Christen Fleming]

Hi Kyana,

summary() on the model fit returns a Gaussian area estimate at the specified quantile level.UD, while summary() on the AKDE returns a non-parametric area estimate at the specified quantile level.UD. If the distribution is not Gaussian, then summary() on the model fit still has meaning as (proportional to) variance in locations. Exploratory bouts get picked up in the tails of the distribution (higher quantiles).

It is more reasonable to feed the estimates (and their variances from the diagonal of slot COV) into a meta-analysis like with R package metafor, but at some point this kind of functionality will be in the package.

Different habitats is good for checking the veracity of the DOP/error estimates, though I expect the e-obs error estimates to be okay. Multiple devices is also a good idea.

N=24 locations is not very much for a calibration dataset, unless you have like 20 tags and they all behave identically. Relative error in the calibration parameter is like 1/sqrt(N), so you want a few hundred locations in total or per device, depending on how similar they behave.

I'm very interested in looking at more calibration data by October 13th, for a manuscript on telemetry error.

Best,

Chris

[Hayley Spina]

Hi Chris, 

Can you provide a bit more detail on how to use metafor for meta-analysis of home range sizes? For each individual home range, yi would be the direct home range size estimate? And what do you mean by "variances from the diagonal of slot COV " - is this something that can be pulled from the CIs of CoV[area], or calculated in some way? 

Thank you!
Hayley