"Delineation" vs. "Stratification"
"Delineation" of polygons is drawing a line around stand or polygon borders. There may be several of these stands with the same description, but each of them has a border that separates them from other areas.
"Stratification" is when you decide to process the data from several of these polygons as a group. You may decide that they have a different ratio in 3P sampling, or a different volume per acre for prism sampling. Traditionally, you gave the same answer to each polygon in a particular strata after combining the samples and processing the result. This never was necessary, but it was common.
In almost all cases, delineation is necessary. Stratification is optional. There are lots of methods that do not use stratification, and many of them are very efficient. There is no automatic need for stratification.
Forestry is a bit "strata crazy", and has been for a number of years. Many of the reasons to stratify started with processing data by hand and reporting the results. These no longer apply. One of the problems is that each strata "needs" (based on anxiety issues) some "minimum number of plots". In some cases that minimum number adds up fast, and gives a large total.
With modern GIS systems, you do not need to call things a strata or need to give them a special code just to combine data for reports. For example, you can simply ask the GIS to report the area and volume of stands having between 12 and 14 thousand BF per acre. Those stands do not need to be identified as "a strata" before the question is asked. Different volumes in the GIS can also rounded into the same map labels without storing those rounded numbers in the GIS itself.
It is far more important these days to be able to give every polygon a different description, and to grow them individually. The automatic urge to stratify often gets in the way of this.
Stratification is often called a "variance reduction" technique, and it can still be useful for that purpose if no estimate of the polygons is available before sampling. By sampling items with almost the same answer you can reduce your sampling error. The same advantage can be obtained by sampling to correct an initial estimate (which you often need anyway, to put the stands into strata). 3P sampling is a good example of this, and does not always need stratification.
From a sampling point of view, the only time stratification is the main alternative is when you know items are different, but cannot make an estimate of what the answer might be. Automated remote sensing sometimes give you this situation. Rounding estimates to put them into strata is almost always a bad idea.
"Delineation" is almost always
needed. "Stratification" is not always necessary or wise.
Stratification is not a bad technique, but when you design a sample you need
to ask yourself why you are doing it. Many of those reasons have not applied
for at least a decade.
Originally published October 2002
Regular Article Index