Data-model-error compatibility

During data assimilation, differences between observations and their model counterparts should be consistent with the error statistics that govern how the model is to be corrected. The concept of incompatibility distance between observations and their model counterparts is introduced as a way of detecting inconsistencies, and formulae are presented for estimating the probability of encountering greater incompatibility. Observations can be examined one-by-one to insure that their confidence intervals are not widely separated from those of the model counterparts. They can be further examined in pairs to detect whether contrasts across fronts are consistent with assumptions about error correlations.