You could use a uniform distribution RNG for the non-predictable random (the larger the minimum and maximum value, the least likely every number in the bounds will be generated) and for the predictable RNG you could use some other distribution (maybe a normal distribution) so that most of your RNG values lie within close distance to each other.
You could also go with another approach where you manually define the probabily of each event you are workign with. So if you want trees to be common in certain areas, then when you are in a forest area, the probability of a tree spawning in a cell should be increased vs. the probability a tree occurs in a cell of a desert. For enemies you could similarily define the probability of enemy encounters, and you could fine tune the probability based on things like sneaking vs. not sneaking, in an enemy base, etc.