**Appendix B**

An assessment of the sensitivity of the metrics characterised in the present study to changes in threshold values has been performed for sample E1. The exact same data processing workflow was followed apart from the initial label seed definition. For the first test, a "normal" threshold range was selected for each label seed. For the second test, much smaller threshold ranges were selected for each of the label seeds, resulting in more of the voxels remaining to be assigned to a specific label by the watershed algorithm. The results from the two tests and their comparison are presented in Table A2. For the volume fraction values, the relative deviations observed are very small, below 1% of the measured values for AN and open porosity. For the morphological parameters, the relative deviations observed are greater, with values around 10%. However, as the standard deviations of the morphological parameters are generally high (~50%), relative deviations of only 10% of the measured values are deemed satisfactory. These results clearly demonstrate that the watershed segmentation of the filtered XCT data from AN prills is repeatable and that the associated data processing workflow developed here is robust.


**Table A2.** Results of the structural and morphological analysis of sample E1 based on two different segmentation strategies.

> 1 Deviation between Test 1 and test 2, with test 1 taken as reference.
