The scale of forest disturbance

I had a couple questions related to study design in forests.  First, what is the general amount of disturbed area in a given ecoregion, and second, how big a landscape does one need before they can say they are looking at a good representation of the ecoregion as a whole?  The first is a (kind of) simple question - just figure out what percentage of the landscape is disturbed over some period of time (of course, this involves assumptions about grain size, definitions of a disturbance, etc).  This isn't even dependent on the extent of your landscape sampling, assuming you sample enough - it's just a feature of the landscape.

Percentage of each ecoregion (WWF definitions, from Olson et al. 2001) for the time period 2000-2014.  Grain size 30m.  Disturbance data from Hansen et al. 2013.

Percentage of each ecoregion (WWF definitions, from Olson et al. 2001) for the time period 2000-2014.  Grain size 30m.  Disturbance data from Hansen et al. 2013.

The second question is trickier.  Now we're not interested in average disturbed, but rather variance between sample landscapes within that ecoregion.  We want to know how big an area is required such that the area represents the ecoregion.  Obviously, a landscape the same size as an ecoregion would be ideal (study everything!) but that's not logistically feasible.  We want to know the extent required to represent that ecoregion.  We can get at that by looking at the extent/scale of investigation required to minimize the standard deviation between sample landscapes at that scale.  For this post, I arbitrarily set that "minimize" value as reducing the standard deviation between random landscapes to below 10% disturbed.

Turns out some ecoregions and their disturbance regimes require HUGE landscapes to adequately represent them, larger than the subsample extents we tried.  Those are not the surprising ones - the boreal, where fires are infrequent but big.  Other areas, more gap dominated areas, can be represented with smaller landscapes.  Ecoregions with hardly any forest can, of course, have very small study areas (since there's little forests to change anyway - something of a "duh" finding there...).  Of special interest are the highly human modified landscapes in the south and the northwest.

This is incredibly useful for monitoring disturbance frequency change, since now we know what size is a good representation of the ecoregion as a whole (another assumption warning - this assumes 2000-2014 is a good sample of disturbance frequencies over longer time periods).  It's also great for designing protected areas - these are the sizes of landscapes that can encompass a natural disturbance regime.

For more, come see the talk at IALE, read the paper (submitted soon), or contact Brian.  Work done in conjunction with NC State and the USFS.

Minimum landscape size within an ecoregion to represent the majority of variation within that ecoregion.  In otherwords, if your landscape of investigation is at least this big, you're likely getting a good representation of your landscape - at least in terms of forest disturbances.  *note the typo, should be >180 km2. Of course other things matter - overall percent forest to begin with, etc.  We also have data related to disturbance spatial shapes, etc - to be published soon!

Minimum landscape size within an ecoregion to represent the majority of variation within that ecoregion.  In otherwords, if your landscape of investigation is at least this big, you're likely getting a good representation of your landscape - at least in terms of forest disturbances.  *note the typo, should be >180 km2. Of course other things matter - overall percent forest to begin with, etc.  We also have data related to disturbance spatial shapes, etc - to be published soon!