Stratifying the area in terms of factors that influence carbon stocks will normally reduce errors associated with project-scale estimates of carbon stocks. At a continental level, climate tends to explain more variation in soil organic carbon than any other single factor (Wang et al., 2010) but locally historic land use often has a dominant influence, and this may not be well reflected by current land use (e.g. Vagen et al., 2006). Stratifying on too many variables can rapidly become un-manageable in terms of the number of strata produced and in practice it is often adequate to stratify on at most several major ecological zones. The sentinel sites are typically large enough to capture variation in conditions at the landscape scale (e.g. valley bottoms, slopes, ridgetops).

More information on sampling can be obtained from: http://www.itc.nl/~rossiter/teach/stats/sampling_nrem.pdf