Two-step unsupervised clustering method on soil nanoscale secondary ion mass spectrometry (NanoSIMS) images to determine the spatial arrangement of minerals and organic matter

Abstract

In the soil, mineral particles, organic matter, and soil organisms are arranged in a complex architecture which can influence organic matter dynamics. Nanoscale secondary ion mass spectrometry (NanoSIMS) provides insights into the microscale arrangement of soil minerals and organic compounds at a resolution of approximately 50 nm. However, current NanoSIMS image processing lacks methods to analyze large datasets automatically and requires manual intervention. We developed a two-step unsupervised clustering method for batch analyses of NanoSIMS images. Our two-step method consists of K-Means clustering as the first step generating around 100 clusters, followed by hierarchical agglomerative clustering (HAC) re-grouping the K clusters into less than 10 cluster groups. The elbow method, HAC linkage method and gap statistics are used to automatically select the optimal clustering numbers. Subsequently, soil minerals and organic matter can be spatially segmented and identified as different species. Further, this method could apply to the spatial arrangement of mineral-dominated and organic matter-dominated parts or different mineral types. Moreover, this method enables the analyses of larger datasets with >100 NanoSIMS images providing insights of organo-mineral interactive hotspots or even micro function domains involved in soil organic matter dynamics.