New publication - How big does your study landscape need to be?

Choosing a landscape extent for study, monitoring, or conservation is difficult once you start considering disturbance processes - it's always possible that a fire, or windstorm, or whatever could come through and drastically change what you're looking at.  This could be fine, but generally when we select landscapes for those purposes we want something representative, something that incorporates disturbances as part of the community and ecosystem ecology of the region (after all, disturbance regimes are a part of the system too!).

This is critical in quantitative ecology at any scale, because few things disrupt a population, community, or ecosystem more than a major fire or a hurricane.  And if we want to truly understand something broader than a single location, we need to know how representative our study system actually is relative to the broader context of the given ecoregion (or system, or whatever).

But how big of a landscape do you need?  The old "rule of thumb" from the 1980's was somewhere between 5-10x the size of the "largest disturbance," but that's really fuzzy.  What you really need is a quantitative analysis of variability in disturbances - how big a landscape such that it doesn't matter where the landscape is, the "disturbance effect" is similar?  This is, in effect, asking how big a landscape extent is required to incorporate that disturbance-driven variability without causing major changes in landscape properties - basically a measure of variance between random landscapes.

So, together with Kurt Riitters at the US Forest Service and Jenn Constanza at NC State, Brian Buma quantified the variability between landscapes at a variety of scales in terms of two disturbance processes - proportion disturbed and contagion (a representation of the shape of disturbances).  This was done for all of North America at a 30m scale, using data from 2000-2014.  Because disturbance regimes vary by region, the data was stratified by ecoregion.

 As a first step, we quantified the actual percent disturbed for each ecoregion.  While we used wall-to-wall satellite data, it may also be desirable to set up a series of landscapes that approximate the ecoregion.  In that case, the number of landscapes matters, of course - the more landscapes the better you will be in terms of a representative set (in terms of disturbance area, in this case).  A shows the ecoregions in the boreal denoted by the oblong, northerly circle (large fires); B shows the ecoregions on the North Pacific coast (very low disturbance frequency, small events).  With larger landscapes, one gets a good idea of actual proportion disturbed more rapidly of course.  The actual map (bottom) is the true average, and a nice reference tool when writing proposals - what is the "normal" fraction of disturbed area in my study system?

As a first step, we quantified the actual percent disturbed for each ecoregion.  While we used wall-to-wall satellite data, it may also be desirable to set up a series of landscapes that approximate the ecoregion.  In that case, the number of landscapes matters, of course - the more landscapes the better you will be in terms of a representative set (in terms of disturbance area, in this case).  A shows the ecoregions in the boreal denoted by the oblong, northerly circle (large fires); B shows the ecoregions on the North Pacific coast (very low disturbance frequency, small events).  With larger landscapes, one gets a good idea of actual proportion disturbed more rapidly of course.  The actual map (bottom) is the true average, and a nice reference tool when writing proposals - what is the "normal" fraction of disturbed area in my study system?

The proportion of the landscape disturbed varies quite a bit at small landscape extents - unsurprisingly.  At a small extent, a random landscape might be within a burned area or in a location completely undisturbed.  At broad extents, however, most landscapes had fairly low variability.  In other words, it didn't matter where the disturbance occurred, it was incorporated into the landscape.  

Of course it depends where you are - some boreal ecoregions, with their big fires, still had high variance from landscape to landscape even at our largest extent.  And contagion takes longer to settle down.  But for many ecoregions, there are practical extents such that, at least under current disturbance regimes, you can be fairly confident that any future disturbance will be incorporated just fine, rather than completely shifting your landscape to something non-representative.

We also include a parallel analysis where we focus only on disturbed pixels themselves, useful to those researchers who are going to be looking at disturbance processes - how big an area around the disturbance do you need to get a real image of the broader ecoregion?

These results are critical in scaling and communicating the significance of ecological research that attempts to put these dynamic change processes into their broader ecoregion context. 

The results are in press:  Buma B, Costanza JK, Riitters K.  Determining the size of a complete disturbance landscape:  Multi-scale, continental analysis of forest change.  Environmental Modeling and Assessment.  In press.

 If you're concerned about your landscape being disturbed, then setting your landscape size larger than these values (in kilometers squared) will reduce the likelihood of a disruption.  The values represent the minimum landscape size to reduce the standard deviation between landscapes below 10% - in other words, it's fairly unlikely (though not impossible!) that a disturbance will make any given study landscape of that size, or larger, dramatically different from the rest of the ecoregion.   All summary statistics, for all ecoregions, included as Supplementary Data and available  here  and in the  paper .

If you're concerned about your landscape being disturbed, then setting your landscape size larger than these values (in kilometers squared) will reduce the likelihood of a disruption.  The values represent the minimum landscape size to reduce the standard deviation between landscapes below 10% - in other words, it's fairly unlikely (though not impossible!) that a disturbance will make any given study landscape of that size, or larger, dramatically different from the rest of the ecoregion.   All summary statistics, for all ecoregions, included as Supplementary Data and available here and in the paper.