Global Journal of Human-Social Science, B: Geography, Environmental Science and Disaster Management, Volume 22 Issue 3

Figure 9: Images similar to the 12SU-LG sample image,' fine fraction of this research The soils with color distribution most similar to soil 12SU-LG' are soils 09MG-LA', 02AR-LG', 05PA-LA', 06CA-LA' and 01GO-LG'. It is noted that the closest soil (09MG-LA') coincides with the result of the cluster analysis, and, with the exception of soils 01GO-LG' and 02AR-LG', the other soils indicated as close are part of the same large group and subgroup (B and B-2). In view of the obtained results, the comparison of soils through images using the techniques of data science can be considered promising, since the obtained results were near the cluster obtained with cluster analysis. The results using these two techniques, even though the same parameters were not used, showed correspondence between the mineralogical characteristics visualized in the images (iron oxides, rock fragments, quartz grains, etc.) with the results of geotechnical tests. The few variations found can be attributed to the methods themselves, since each technique used can provide different results (distance measurements, linkage methods, image descriptor, etc.). As mentioned in the topic of Materials and Methods, for the acquisition of images, a Stereo Microscope Zeiss Discovery V8 available in the Paleontology laboratory of the Department of Geology of Federal University of Pernambuco was used. Although not trivial, it is believed that access to this type of equipment in universities is not very difficult, since it is an essential equipment in geology laboratories, biology and medicine. It can be noted that to use the Data Science tools presented here, it is not required that the images be obtained by this type of specific microscope, in fact, the important thing is that the images are of good quality and that they are obtained using minimum standards (process of obtaining samples, fraction of material, approximations, etc.) in order to obtain reliable results. IV. C onclusion The thirteen tropical soils of fine granulation examined in this study, originating from the northeastern region of Brazil, were characterized as materials with physical and mechanical behavior varying between clayey and sandy, having been labeled into different classes of fine soils according to the considered Classification systems (MCT, TRB and USCS). They presented high values of resilient modulus, low levels of permanent deformation, unsaturated behavior of the bimodal type in all soil-water characteristic curves, as well as chemical and mineralogical characteristics indicative of typically lateritic soils. Multivariate analyses, such as cluster analysis applied in this article, consider three or more variables to characterize the behavior of the analyzed object, so it is understood that several geotechnical parameters were used at the same time to form the groups with the most homogeneous characteristics. In this sense, other clusters were tested excluding, for example, the parameters of the two- dimensional models of resilient behavior (keeping only those associated with the composite model that presented the best framework for most soils), and it was observed that by excluding only the data associated with the model, or due to the deviation stress or the confining stress, there is no change in the dendrogram, however, by removing all the parameters associated with the two models, a dendrogram with another cluster Volume XXII Issue III Version I 22 ( ) Global Journal of Human Social Science - Year 2022 © 2022 Global Journals B Clustering of Fine-Grained Tropical Soils using Data Science Tools Applied to their Geotechnical Properties