bias toward aspen; no pine

[achubaty]

Currently, the results in most/all histograms show no pine (or other conifers) but med-to-high amounts of aspen.

Could be the result of the interaction between LANDIS and LandMine, where LANDIS produces more aspen post-disturbance in short-term and Landmine burns pine/spruce preferentially and doesn't know how to deal with changing vegetation types.

LANDIS:

• "turning off succession" (i.e., no dispersal)
• timing of resprouting (immediately) vs serotiny (delayed)
• dispersal parameters for pine (125m) vs aspen (5 km)
• check for possible errors in species naming/coding (maybe something is being missed)
• pine is under-estimated to begin with in the kNN data sets (though the overlay of proprietary data should improve the estimates). Try artificially increasing pine on the landscape to see effect on results.
• establishment probabilities are too low for pine?
• aspen longevity too high?

LandMine:

• rerun model without landmine to see if problem persists to same extent (not using LandMine isn't an option for LandWeb)
• rerun with equal burn probabilities
• run with longer fire return intervals (e.g., multiply by 2)

see also: LandWeb_verification, #98, #100, #101, #102, #103, #104, #105

[achubaty]

species coding

Does not appear to be an issue, although I'm not sure what species 0 corresponds to in the [leading] veg type maps. Is it NAs or "no species"?

code	species
0	?
1	Abies_sp
2	Pice_gla
3	Pice_mar
4	Pinu_sp
5	Popu_tre

[achubaty]

dispersal parameters

After discussing with Eliot, Ceres, and others, the parameter values are reasonable for Ward dispersal.

AMC: i'm sitting on a fix that makes seed dispersal in LBMR go from the edge of the pixel (I added half the pixel width to the disperasl distance params for each species) to see if it makes a difference with the deciduous bias we were seeing. so for it looks like it might but I need to test on larger area with more pine

EM: I don't think that is correct behavior.
Rather, you round based on distance
i.e., if seed disperses 20m,and pixel size is 100m,then 20/100 probability that is goes to next pixel.
That keeps the pro ability correct.
*probability

AMC: hmmm. and 125/100 means guaranteed dispersal next pixel?

EM: If you want to do what you suggested, you need to adjust the dispersal kernel too, because Ward relies on the fact that there is an average biomass in the pixel.
No. 125/100 is 0.25 probability in next
2nd
Also that should already be there.
Also Ward dispersal is backwards, from receiving pixel to source (adult trees) pixel
So, it is unlikely that it is responsible for any bias. It only takes one success for a given receiving cell for it to initiate a cohort
5000m vs 150m for parameter values
And biomass of deciduous VS conifer by the time of burn

UPDATE: after further discussion:

1. white spruce disperses further than default parameterization allows (use higher max dispersal distance for white spruce #96);
2. try some runs with double max dispersal distance to counteract effect of large fires (double the max dispersal distance for all species #105).

[achubaty]

species establishment probabilities

Eliot/Ceres adjusted these so that they are based on the proportion veg cover in the current conditions data. This change has dramatically improved the outputs by stabilizing the number of vegetated pixels (i.e., more pixels are regenerating, which prevents the massive die-off we were seeing with previous versions. However, note that the relative species proportions are still out of whack and are aspen-biased.

[achubaty]

doubling the fire return intervals

Dave has made it clear that this is not an option for LandWeb. Regardless, it's such an important assumption/input of the model that we need to evaluate it anyway.

The fire return intervals in some regions are really short given that they represent the mean number of years to burn an area equivalent to that region. LandMine models stand-replacing fires, so it’s no surprise that in regions with low FRI we see no old trees (or worse, no vegetation at all because it all burns up and never regenerates).

Also note, in areas with low FRI, the fire spread algorithm simply cannot find enough pixels to burn and it has to jump around a lot, which massively slows down computations. To address this, FRIs below 40 years are NAed, which results in no burning in those areas. This may be of particular concern for small FMAs where a large portion “doesn’t burn”.

[achubaty]

aspen longevity too high

Another set of runs where aspen longevity is shortened to 80 years are being run.

Note that this will affect the aspen growth curve, by compressing that curve along the x-axis -- it will die sooner but grow faster. This causes aspen to out-compete everything.

[achubaty]

"turning off succession"

Not particularly useful diagnostic, as this simply turns off all seed dispersal and regeneration. Because vegetation is not using a state-transition model, without seed dispersal, burnt pixels never recover their vegetation.