Coincident data with different values indicates measurement error. For semivariograms, the nugget is composed of microstructure and measurement error. Thus, if the dataset contains two or more different values for a single point, some measurement error exists, and the nugget cannot be zero.
Increase the nugget value of the semivariogram model so measurement error can be included. Alternatively, choose a method to handle coincident values (for example, use the mean value).