The UN adjustment factor is applied to a national level, and then all subnational units are scaled accordingly. If you read the paper for GPW3, that gives the basic method for doing this, please see http://sedac.ciesin.columbia.edu/gpw/documentation.jspAlso, we can write the differences off, between the UN -adjusted and the unadjusted data, to noise or rounding error. The stats of the population count grid (unadjusted) for the areas that are not zero in the unadjusted grid and zero in the adjusted grid are: COUNT = 905129 (of over 35 million cells with population values) MAX = 1.4624 MEAN = 0.0005 These areas are very close to zero because small counts are being spread out across many grid cells. For these polys, a negative UN adjustment factor decreases the population counts small enough that they are being rounded to zero during the gridding process. If you are interested in finding areas that are ‘uninhabited’, I’d suggest using a density cutoff (< 1 or < 5 persons / square km, for example).