Conclusions

Three conclusions may be drawn from this study. First, spatial correlation is a crucial contributor to uncertainty in map analyses that aggregate results over multiple mapping units. Ignoring this contribution inevitably leads to underestimates of variances and unwarranted statistical confidence in estimates. Unfortunately, the importance of this source of uncertainty is generally not known to those who conduct map-based analyses, and techniques for estimating its effects are generally unfamiliar. The second conclusion is that greater attention in the literature and academic environments should be given to uncertainty estimation. Third, estimation of uncertainty is not trivial, either conceptually or from a technical perspective. The necessity of decomposing very large matrices limits the size of regions that can be analyzed without high-speed computing facilities.