SVF variance too low

[Karina Li]

Hi All,

Has anyone had the issue of not being able to adjust the variance for the SVF guess high enough to reach the asymptote on their semivariogram? The fitted function does not look great (see attached images). Any suggestions for how to fix this? I fear that the resulting AKDE is severely underestimating the true home range for this animal.

Thanks!

-Karina

[Christen Fleming]

Hi Karina,

You can change the guess object parameter manually by assigning a larger value to that slot, and the slider will be centered on that value when you run ctmm.guess().

The discrepancy is likely from the location error model being substantial. This inflates the variogram above the theoretical semi-variance function of movement (which is without location error). In the past, I tried to include the error model in the theoretical SVF, but the results were not great.

In any case, the variance of the data will be decomposed into variance from location error and variance from movement. I would make sure that the location error parameters make sense in the end, and definitely do not freely estimate them (without calibration data or a strong prior).

Best,

Chris

[Karina Li]

Hi Chris,

Thanks so much for your response. I tried messing with the code for a few hours today, but couldn't figure out how to assign a larger value for the variance, as you instructed above. Everything I tried either resulted in an error or the same values on the sliders as I had previously. Specifically, I'm a bit stumped by the way the estimated variance is stored in a ctmm object/guess - I see that it is an object of class "covm", which I'm having a hard time interpreting. Apologies if the answer is obvious, I'm pretty new to R...

-Karina

[Christen Fleming]

Hi  Karina,

Try this:

GUESS$sigma <- 4.4*1000^2

Best,

Chris