Kriging gives you two maps: a surface, and its own uncertainty. At the centre of our three gauges it reads 30.2 mm, give or take 19.

Kriging uses the data twice. First it measures how alike the values are at each distance and fits a curve, the semivariogram, with a range, a sill and a nugget. Then it uses that curve to weight the neighbors, and it reports a standard error for every prediction. Cross validation tells you whether those error bars are honest.