Error calibration from different sites

[steveh...@gmail.com]

Hi Chris,

Thanks again for all y our assistance so far, I deployed some stationary GPS loggers at three of our field sites and have been fitting error models. Our loggers don't record any DOP values, just an 'EHPE' value. I saw another conversation on here where you suggested to use this as a HDOP value. This works, and produces a better model than not using that parameter.

However, the error estimates produced (0.012) seem very small compared to the ones in the vignette, especially when some of the points that were recorded were > 50 meters from where the logger was sitting (as taken by a handheld GPS). When comparing the points from the loggers to this point, it is apparent that the x error is much greater than the y error. Is there a way to account for this in the uere.fit call?

When I checked to see if all the loggers were performing similarly, individual based models were preferred by AICc selection. Two of the sites had small estimates as above, but the other site had an estimate of 18.35. This didn't really align with my expectations, as the more rugged site that has more error wasn't the site with the higher estimate.

I found that I can fit a lm that find that there is a strong relationship between elevation error (the difference between the recorded altitude and elevation from a DEM), but this doesn't seem to be something I can use in the uere.fit?

I feel like I may not have collected enough data from these deployments, but we had  a shortage of loggers when I was last able to get out. We have more loggers now (and less animals due to drought related mortality), so I plan to put more out next time I get the chance, but I would be glad to hear your thoughts!

Thanks again for all your assistance,

Steve

[steveh...@gmail.com]

Here is the stationary logger data, it wouldn't let me upload on the first post.

[Christen Fleming]

Hi Steve,

You're looking at the summary of the uere.fit, but that's in units of meters per EHPE, and the EHPEs are about a thousand times larger than DOPs. Your median estimated error is actually 20 meters and the plots look appropriate if you plot the calibrated calibration data. So this seem to be working fine.

Right now, uere.fit will only fit a linear model with categorical modifiers (which how these things are supposed to work), so something like HDOP = covariate_1 * covariate_2 will work fine, but something like HDOP = covariate_1 * covariate_^beta would require running uere.fit inside an optimizer to estimate beta. Alternatively, one user was detrending the median locations and fitting a generalized gamma model, which is the same model but with a worse mean estimate (median instead of error weighted mean).

As far as the errors being elliptical, that can happen when part of the sky is blocked out by a mountain/hill/tree. However, GPS devices don't provide any kind of error-ellipse information, even though they theoretically could. So there's not much you can do about this, in practice.

Best,

Chris

[steveh...@gmail.com]

Awesome good to hear that the values seem reasonable.

Should I be worried that the different sites gave different estimated error? It probably isn't enough to make a huge difference, but I could go collect more stationary data if you think it would be prudent.

Steve

[Christen Fleming]

Hi Steve,

Ideally, you would want to model the differences or (coming at some point) model the variation. If you are curious about how much difference it can make in the end, then you can run your analysis with a few different calibrations and see what comes out. Small differences shouldn't' have a large effect.

Best,

Chris