Evaluating spatial autocorrelation and depletion in pitfall-trap studies of environmental gradients