Understanding collocation mismatch is particularly relevant for atmospheric profiles obtained by radiosondes, as the balloons containing the measuring instruments tend to drift uncontrollably from their initial launch position. We propose a heteroskedastic functional regression model capable of explaining the relationship between collocation uncertainty and a set of environmental factors, height and distance between imperfectly collocated trajectories. Along this line, a five-fold decomposition of the total collocation uncertainty is proposed, giving both a profile budget and an integrated column budget. Considering the profiles as three-dimensional trajectories, we extend the model to include a trivariate smooth function that accounts for time and space mismatch. Results from a case study where we model collocation error of relative humidity and atmospheric pressure show that model fitting is improved once heteroskedasticity is taken into account.