It's my impression that we differ from forest ecologists/ecologists at least in this manner. Forest researchers will randomly select plots to statistically sample a given experimental block (itself subject to levels of randomness), perform measurements, and with analysis make conclusions about the parts and the whole. Forest ecologists on the other hand, seem to walk into an area that to them seems to be an ideal example of what it is they are trying to define, and take measures of that 'community' so as to qualitatively summarize what it is they thought they'd seen to begin with.
This is an interesting observation. I am neither a forester, nor a forest ecologist, but I do have some knowledge of sampling and experimental design so I have no particular axes to grind. I don't think your observation is really a fair appraisal of the situation. There are problems with both approaches
If the analysis is based upon statistical analysis of random plots, that analysis is only correct to the degree in which the random sampling is representative of the whole. Is what you sampled in the random sampling typical or not? If an atypical of anomalous area or feature is included in one of the random plots, that feature will be over emphasized in the statistical analysis. If there is a large number of plots, these anomalous samples will be minimized by the weight of data from other sample plots. If there is only a few plots, then the data will be skewed. One example of this I have come across was an analysis of tree populations in the island in the Allegheny River Island Wilderness. One report, (I would need to look up the specifics), listed butternut as one of the major species on the islands representing a high percentage of the tree population. In fact from personal experience on these islands, I have found that butternut is present, but only sporadically and in patches. The random transect or plot happened to include one of these patches of butternut and the statistical analysis of the sampling led to an erroneous conclusion about the relative percentage the butternut in the population. Random plots work best where there are large numbers of samples, but can misrepresent the actual ground truth if only a limited number of plots are analyzed.
In the second scenario there is no doubt that selection of a site can lead to people finding what they want to find. But this is not always the case. In site selection, if done properly, there should have been developed a set of criteria for what sites to select. These can be developed by some random test plots, by general qualitative surveys of the entire area. The criteria would include what major characteristics or features are typically present, and what features are present but uncommon. Then a site site would be selected for detailed sampling and analysis based upon this set of selection criteria. This would include more detail than in the preliminary overview of the area. The site selected would include all of the major features found to be typical in the preliminary surveys, and minimize the atypical elements. The quality of the analysis would be vary based on how well the selection criteria was able to characterize a typical area within the forest. If only a limited number of sites are sampled, this has a potential to provide a better characterization of a typical site than does a random sampling, and it will allow a more detailed analysis focusing on one site rather than many sites. But it also has the potential to add observer bias as well. So neither methodology is without problems.
"I love science and it pains me to think that so many are terrified of the subject or feel that choosing science means you cannot also choose compassion, or the arts, or be awe by nature. Science is not meant to cure us of mystery, but to reinvent and revigorate it." by Robert M. Sapolsky