search

PalEON-Project / stepps-calibration

STEPPS pollen-veg calibration model code and paper.
0 stars 1 forks source link

What is wrong with the Tamarack composition estimates? #15

Closed andydawson closed 9 years ago

andydawson commented 9 years ago

We seem to have a really hard time estimating Tamarack. I looked into this, and it looks like our estimates of local-ness (gamma) have very wide credible intervals. So I plotted the pollen proportions and the PLS proportions to see what was going on. Looking at these, I see why we have a hard time getting this one right - the spatial patterns of pollen and PLS are quite different.

PLS

pls_tamarack

Pollen pollen_tamarack

@SimonGoring or @IceAgeEcologist , any thoughts? It looks like we don't have any pollen samples from where Tamarack is most abundant.

SimonGoring commented 9 years ago

My guess is that tamarack is largely associated with wetlands, so it tends to show up stochastically in high amounts based on the fact that a wetland is sampled, not because of the regional frequency of wetlands. So you get some very high counts in southern MN, but not elsewhere if (for example) a lake was sampled rather than a wetland.

Further, the places where wetlands are highly concentrated (northern MN) are likely undersampled relative to the rest of the region because the vegetation signals might be thought of as 'uninteresting' relative to more southern records so places where it's likely to be locally sampled, and regionally abundant are underrepresented in the data.

IceAgeEcologist commented 9 years ago

Agreed with Simon. Tamarack is probably a very localized signal, with high abundances right next to particular wetlands. Sometimes an FIA plot is sitting on one of those wetlands, sometimes not. Ditto the pollen. So there is high local heterogeneity, which is being differently sampled by the FIA and pollen data, and so they are producing very different regional maps.

I think this will be a taxon worth discussing in the paper as an example of a taxon that doesn't work very well in the STEPPS model, and with an explanation of why. i.e. upscaling and spatialization isn't working well for Tamarack, but is behaving well for other tree taxa.

SimonGoring commented 9 years ago

It's worth pointing out that this is a problem that is common in pollen stuff. Aquatics don't really interpolate well either, for example Sphagnum.

On Wed, Jul 22, 2015 at 12:10 PM, Jack Williams notifications@github.com wrote:

Agreed with Simon. Tamarack is probably a very localized signal, with high abundances right next to particular wetlands. Sometimes an FIA plot is sitting on one of those wetlands, sometimes not. Ditto the pollen. So there is high local heterogeneity, which is being differently sampled by the FIA and pollen data, and so they are producing very different regional maps.

I think this will be a taxon worth discussing in the paper as an example of a taxon that doesn't work very well in the STEPPS model, and with an explanation of why. i.e. upscaling and spatialization isn't working well for Tamarack, but is behaving well for other tree taxa.

— Reply to this email directly or view it on GitHub https://github.com/PalEON-Project/stepps-calibration/issues/15#issuecomment-123794783 .

andydawson commented 9 years ago

Thanks guys - this is really cool stuff! I'll add mention of this in the paper.

paciorek commented 9 years ago

We also seem to be missing pollen sites from the locations where tamarack is abundant, so that doesn't help. The two areas with high tamarack are northern MN, west of the single site that does have a fair amount of tamarack pollen, plus a SW to NE ridge of high tamarack west of Duluth that is missing pollen sites.

chris

On Wed, Jul 22, 2015 at 10:15 AM, Andria Dawson notifications@github.com wrote:

Thanks guys - this is really cool stuff! I'll add mention of this in the paper.

— Reply to this email directly or view it on GitHub https://github.com/PalEON-Project/stepps-calibration/issues/15#issuecomment-123796044 .

andydawson commented 9 years ago

This issue has been addressed in the manuscript.