One thing I’ve been working on lately involves statistical modelling of the probable distribution of permafrost at a specific project site in Yukon, Canada, where discontinuous permafrost is present. Permafrost is ground (soil and/or rock) that remains at or below 0C the whole year. In this area, presence of permafrost is patchy/sporadic. The following image is a predictive map for expected permafrost distribution:
The map shows approximate distribution of permafrost expected in different areas of the site, derived from a statistical analysis of past observations of frozen-unfrozen ground in relation to a number of other spatial themes (elevation, slope angle, slope aspect, ground curvature, soil type, etc). You can visually compare the different susceptibility levels (low, moderate, high) with the observed distribution of frozen/unfrozen ground. You will see the model is not perfect, but I think it is “pretty good.”
The basic idea is that in green areas, you can expect that about 15% of random test holes will encounter frozen ground (and 85 % unfrozen) during late summer. This increases to about 50 % for moderate (orange) and 85 % for high (red).
The interpretation uses a modified “weights of evidence” approach, which is a bi-variate statistical method comparing presence/absence of a specific characteristic (in this case frozen ground, assumed to be permafrost) individually with other spatial data. The predictive map involves the summation of individual weights calculated through those comparisons. I’ll not explain the method in detail here, one can easily google “weights of evidence method” to come up with an explanation.