b'Education matters classifier such as Random Forests,summary, these methods, previously theCongcong Gai, Australian National and the value to be gained from andomain of computer and data scientists,University: Palaeomagnetic and rock assessment of the spatial distribution ofare now developed to be more widelymagnetic study of North Pacific and South class probabilities as opposed to viewingaccessible to mineral explorers. China Sea sediments.a final map as a solution in isolation.Marcus Haynes, Australian NationalMagnetism plays an important In the third and final core study, a rangeUniversity: A Bayesian reappraisalrole in marine science. On the one of training data sampling paradigmsof Australian crustal heat flow andhand, palaeomagnetism helps us to are tested in a data rich area locatedtemperature. understand ancient geomagnetic in the Domes region of the Centralfield behaviour that can also facilitate African Copper belt hosting the SentinelGeothermal data offers a uniquegeochronological studies. On the other (Ni) and Enterprise (Cu) deposits. Thisperspective from which to imagehand, rock magnetism can provide study simulates early and advancedthe Earth. However, geothermal datavaluable information concerning stage exploration project maturityis difficult to collect and this haspalaeoenvironmental and paleoclimatic in incorporating a priori geologicalnecessitated a reliance on industrialprocesses. However, magnetic particles Information. It culminates in the usedata collection during the explorationin marine sediments are influenced by of Random Forests to undertake anfor mineral and petroleum resources.multiple geological and environmental objective audit of the present companyResulting data quality issues have limitedprocesses, and commonly occur in geological map. Further to this,previous studies in their ability mapmixed magnetic mineral assemblages, unsupervised clustering is used in thethe information contained in the datawhich complicates magnetic parameter production of a geological map in theinto robust model inferences. In thisinterpretation. Therefore, to apply absence of training or constraint throughthesis I employ a Bayesian statisticalenvironmental magnetism in marine identifying the natural grouping of data.framework to address these issues,science, it is crucial to be able to The results of these studies highlight thepredominantly through the Bayesiandiscriminate magnetic signals in marine importance of proper sample balancingre-appraisal of Australian crustal heatsediments and to relate these signals to and explore the repercussions of limitedflow and temperature data. This has beenrelevant processes.and/or non-representative training data.achieved by defining sensitivity kernels The use of the information entropyfor surface heat flow determinations,In this study, detailed palaeomagnetic proxy is developed to identify whereexamining the sources and magnitudesand rock magnetic studies have been a classification may depart from theof noise in primary geothermal dataperformed on sediments from the North domain represented by training data. Thesets, and developing methods to assignPacific Ocean (NPO) and the South ranking of input data that is performedpetrophysical rock property a prioriChina Sea (SCS). The main conclusions in association with the Random Forestsdistributions with spatial variabilityof this work are as follows. Magnetite classification can be used to improvescaled appropriate to the length-scalesis the dominant magnetic mineral clustering results through optimisingof modelling. I combine these factors andthroughout SCS Hole U1431D, and its dataset selection. use them to infer the surface heat flowremanence is used to reconstruct a field across Australia. This demonstratesmagnetostratigraphic time framework Through the three core researchthe existence of a relationship betweenfor the past ~6.5 Ma. A single authigenic chapters, a set of practical considerationssurface heat flow determinations andgreigite-bearing interval is identified and recommendations for explorerslithospheric thickness. Remote sensingat 130.5-132.0 mbsf, whose age is are provided. It is demonstrated thatand Argo sea float data have beenestimated at ~2.53-2.55 Ma is associated Random Forests can provide an objectiveused to derive a national land-surfacewith oceanographic changes that audit and subsequent refinementtemperature coverage, providing thecoincided with intensification of northern of a pre-existing geological map.basis for detailed crustal temperaturehemisphere glaciation. Sedimentary The expression of uncertainty usingmodelling. This is supported by a newmagnetic parameters are affected mainly information entropy, and the assessmentframework for assessing the uncertaintiesby the East Asian monsoon and are used of class probabilities, can be used toinherent in bottom-hole temperatureto indicate monsoon evolution over the appraise the results from the machinemeasurements, enabling us to leveragepast 6.5 Ma. The summer and winter learning analyses. This includes validationvast industrial data sets to model crustalmonsoon were stable before 5 Ma. The in the case of complex outcomethermal fields. The significance ofsummer monsoon intensified from 5 Ma combinations, and generation of newthese outcomes is that they advancesto 3.8 Ma, when it started to weaken. insights. Ranking of input datasetsus on the pathway towards genuineThe winter monsoon weakened at 5 Ma, via Random Forests can enhancejoint-inversion involving geothermaland then intensified after 3.8 Ma, finally understanding of data and improvedata. The quantification of uncertaintystabilizing at 0.6 Ma. Spectral analyses both Random Forests classificationdistributions on geothermal dataindicate that there was a direct response results and improve clustering. With thesets provides the means to calculateof the summer monsoon to low-latitude proper selection of appropriate datasets,the degree with which a given earthinsolation between 6.5 and 3.2 Ma. Since clustering (for example immobile tracemodel is consistent with available3.2 Ma, both low-latitude insolation elements) and scaling can indeedgeothermal data. In doing so, this thesisand high-latitude global ice volume produce results that correspond well withestablishes the basis for future datahave influenced summer monsoon lithology. Studies presented in this thesisintegration projects to build detailedevolution. In NPO core NP02, three use data from current/active explorationmodels of Australian crustal structurevisible tephra layers are identified, and projects and methods are distilled toand composition, and through doingcan be correlated to the To-Of, Spfa-1, streamlined workflows using industryso, to better constrain the Australianand Kt-3 tephra layers. Together with a standard software and data formats. Inlithospheric heat budget. radiocarbon date, an initial age model DECEMBER 2021 PREVIEW 38'