search

PalEON-Project / stepps-calibration

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

White cells in map figures #13

Open andydawson opened 9 years ago

andydawson commented 9 years ago

I suspect that the ref set of coordinates I use for plotting has the water cells removed. Updated this to include ALL coordinates that are within the state land boundaries. It's entirely cosmetic, but should be fixed nonetheless.

andydawson commented 9 years ago

Actually I'm not sure if we want to plot these. Using the composition estimates from @paciorek , I filter to include all cells that are within the state boundaries of the UMW states, and then remove any cells with 100% water. What remains is the veg I use to calibrate. I think I should write this in the paper and leave those cells empty in the figures. We can predict pollen (and veg) there, but it's not really part of our original domain...

andydawson commented 9 years ago

@paciorek @SimonGoring Do we want to use the cells classified as being 100% water in the calibration model? They have composition estimates. In the current implementation they are not included in the domain for STEPPS at all. I'm not actually sure where the water classification comes from, but if we are pretty sure that the cells actually are 100% water we should probably leave them out.

In the UMW, there are 20 cells that are 100% water; 30 are 95% or more water.

paciorek commented 9 years ago

Yes, I agree that leaving out mostly water cells makes sense. Composition model just predicts everywhere, so it's as if we have "potential vegetation" even where there is water. But then I mask based on Simon's water layer - but this only leaves out Great Lakes and some very big lakes.

chris

On Fri, Aug 21, 2015 at 8:12 AM, Andria Dawson notifications@github.com wrote:

@paciorek https://github.com/paciorek @SimonGoring https://github.com/SimonGoring Do we want to use the cells classified as being 100% water in the calibration model? They have composition estimates. In the current implementation they are not included in the domain for STEPPS at all. I'm not actually sure where the water classification comes from, but if we are pretty sure that the cells actually are 100% water we should probably leave them out.

In the UMW, there are 20 cells that are 100% water; 30 are 95% or more water.

— Reply to this email directly or view it on GitHub https://github.com/PalEON-Project/stepps-calibration/issues/13#issuecomment-133457918 .

andydawson commented 9 years ago

OK, great. So are defining 'mostly water' as 100%? Or should the cutoff be something less than that?

SimonGoring commented 9 years ago

Just to be clear, did you say there are counts in cells that are supposed to be 100% water?

If there is less than 100% water then we should keep it in. Does the model account for tree # in assessing uncertainty?

A mask makes no difference particularly in the prediction stage. In calibration its important to know what's there, and the number of trees we have for the cell, right? On Aug 21, 2015 10:32, "Andria Dawson" notifications@github.com wrote:

OK, great. So are defining 'mostly water' as 100%? Or should the cutoff be something less than that?

— Reply to this email directly or view it on GitHub https://github.com/PalEON-Project/stepps-calibration/issues/13#issuecomment-133463880 .

andydawson commented 9 years ago

No, I didn't say that there PLS counts in 100% water cells - I don't know if there are or not. Somebody should check that there are no counts in the cells that the mask defines as being 100% water. Who wants to do this?

There are composition estimates in 100% water cells though, which is what I was referring to.

What do you mean tree #? You mean number of PLS tree counts per grid cell? Calibration model doesn't use the raw data, so definitely does not. As for the composition model, I'll let @paciorek answer that (but I think it does).

SimonGoring commented 9 years ago

Okay, no, there are no counts in all water cells. I just wanted to make sure I understood the distinction. On Aug 21, 2015 11:03, "Andria Dawson" notifications@github.com wrote:

No, I didn't say that there PLS counts in 100% water cells - I don't know if there are or not. Somebody should check that there are no counts in the cells that the mask defines as being 100% water. Who wants to do this?

There are composition estimates in 100% water cells though, which is what I was referring to.

What do you mean tree #? You mean number of PLS tree counts per grid cell? Calibration model doesn't use the raw data, so definitely does not. As for the composition model, I'll let @paciorek https://github.com/paciorek answer that (but I think it does).

— Reply to this email directly or view it on GitHub https://github.com/PalEON-Project/stepps-calibration/issues/13#issuecomment-133474540 .

paciorek commented 9 years ago

yes composition model accounts for tree counts in its uncertainty. It makes predictions in places with water as if the surveyors just didn't go there, so there was no data.

To be really fancy, I suppose we could have pollen production for a cell scale inversely based on %water, but not sure we want to go there at this point.

Is the question what to do if and when we rerun the calibration model? If so, I suppose we can scale based on %water ...

On Fri, Aug 21, 2015 at 9:13 AM, Simon notifications@github.com wrote:

Okay, no, there are no counts in all water cells. I just wanted to make sure I understood the distinction. On Aug 21, 2015 11:03, "Andria Dawson" notifications@github.com wrote:

No, I didn't say that there PLS counts in 100% water cells - I don't know if there are or not. Somebody should check that there are no counts in the cells that the mask defines as being 100% water. Who wants to do this?

There are composition estimates in 100% water cells though, which is what I was referring to.

What do you mean tree #? You mean number of PLS tree counts per grid cell? Calibration model doesn't use the raw data, so definitely does not. As for the composition model, I'll let @paciorek https://github.com/paciorek answer that (but I think it does).

— Reply to this email directly or view it on GitHub < https://github.com/PalEON-Project/stepps-calibration/issues/13#issuecomment-133474540

.

— Reply to this email directly or view it on GitHub https://github.com/PalEON-Project/stepps-calibration/issues/13#issuecomment-133478421 .

andydawson commented 9 years ago

I just wanted to confirm that we were all on the same page with respect to what we currently do. I am probably going to run calibration again for the paper (after the initial submission) but think we can save % water for prediction paper results if at all.

Just wondering, is the number of raw counts in a cell related to the proportion of water?

SimonGoring commented 9 years ago

That's not easy for me to figure out right away and it's complicated by the fact that there are lots of 'edge' cells where the number of trees is low, because (for example) Lake Michigan is there, but Lake Michigan doesn't get counted as a "lake" or "water" cell in the PLSS because it's too big.

Simon

On Fri, Aug 21, 2015 at 1:37 PM, Andria Dawson notifications@github.com wrote:

I just wanted to confirm that we were all on the same page with respect to what we currently do. I am probably going to run calibration again for the paper (after the initial submission) but think we can save % water for prediction paper results if at all.

Just wondering, is the number of raw counts in a cell related to the proportion of water?

— Reply to this email directly or view it on GitHub https://github.com/PalEON-Project/stepps-calibration/issues/13#issuecomment-133526524 .