The quantification of microstructural parameters of glass-bearing samples: electron backscatter diffraction mapping versus backscatter electron imaging

Quantifying phase proportions (Φ), grain size, crystal area number density (N_A), and surface area to volume ratio (S_v^P) is essential to understanding igneous crystallisation. Electron backscatter diffraction (EBSD) offers advantages over backscattered electron (BSE)-based image analysis: it segments grains based on crystal structure and orientation, and involves a smaller interaction volume. However, EBSD can be time-consuming, and grain reconstructions depend on post-processing parameters and workflows, especially in glass-rich samples. We evaluated strengths and weaknesses of microstructural analysis of glass-rich samples using EBSD.

Crystallisation experiments synthesised and analysed via BSE imaging by Pontesilli et al. (2019) were re-analysed using EBSD. The samples are two synthetic trachybasaltic glasses, one nominally anhydrous, the other with 2 wt% added H₂O, heated to a superliquidus temperature of 1300°C at 400 MPa and fO₂ close to NNO+2 in a piston cylinder apparatus. After 30 minutes the samples were cooled at 80°C min^-1 to 1100°C (considerably below their liquidus temperatures) and held there for 30 minutes before quenching. The samples contain dendritic to skeletal clinopyroxene (Cpx) crystals in a glass matrix, clustered with smaller skeletal to anhedral titanomagnetite (Tmt) grains.

Standardising the confidence index of EBSD pixels to the highest value for each grain strongly influences quantification results. Standardisation leads to overestimation of crystallinity and grain size, but delivers better estimates of N_A and S_v^P values. For non-standardised scans, varying the minimum confidence index threshold used for cleaning affects all microstructural parameters studied, whereas varying minimum grain size threshold and step size strongly affects only N_A and S_v^P. Varying boundary smoothing only affects S_v^P.

For Cpx, EBSD and BSE are in excellent agreement for Φ_Cpx in both samples and S_v^P_Cpx in the hydrous sample, while EBSD-derived S_v^P_Cpx is 50% higher than the BSE-derived value for the anhydrous sample. For both samples, EBSD-derived maximum Cpx length is ~ 100% higher than the BSE result. For Tmt, EBSD systematically finds slightly elevated Φ_Tmt and Tmt grain size, and for the anhydrous sample only, significantly higher Tmt N_A.

Despite larger maximum lengths, calculated Cpx growth rates from EBSD are within 10% of BSE-derived values, because the calculation employs the square root of length times width. The large differences in N_A and S_v^P found for the anhydrous sample derive from its finer, dendritic microstructure, and the smaller (200 nm) step size of the EBSD scans compared to BSE imaging. For the more euhedral and coarser-grained hydrous sample, BSE and EBSD return similar results, and an EBSD step size of 1 µm is sufficient. The systematically larger Tmt sizes obtained from EBSD are overestimates due to signal from Tmt below the sample surface.

In conclusion, care must be taken applying EBSD to glass-rich samples. Thresholds must be carefully chosen by comparing reconstructed grains and image quality maps, different processing workflows are required to obtain different microstructural parameters, and phase-specific over-/underestimates of parameters may occur. EBSD delivers most improvement for microstructures with sub-micrometer length scales.

Pontesilli et al. (2019), Chem Geol 510:113-129. 10.1016/j.chemgeo.2019.02.015

Funded by the Austrian Science Fund (FWF): P 33227-N