Shales form distinctive peaks in natural gamma downhole logs, which are used to manually identify the location of ore boundaries. These shales form important lithology markers that can potentially be automatically identified using Gaussian processes. This approach is trialled using data from a typical iron ore mine in Western Australia which contains the Marra Mamba sequence of the Hamersley province.The method uses Gaussian processes (GPs) with a single length scale squared exponential covariance function. A library of 8 m sections, where natural gamma measurements were taken at 10 cm intervals down the hole, was used to train the GP model. This library was iteratively improved and was tested using natural gamma logs that were not included in the library._x000D_
Areas that were misclassified or were not clearly classifi ed were added to the library. The new library was then used to train the model and improve the results. The trials used a double gamma peak from the NS3 and NS4 shales at the boundary between the Newman 1 and Newman 2 ore zonesin the Newman member and the AS1 and AS2 shales at the base of the West Angelas member. The results show that both boundaries can be identifi ed with an accuracy of over 99 per cent. The classification has the highest accuracy on undeformed shales. This method provides a fast, accurate and objective identifi cation of shale and ore boundaries and can potentially replace the current manual detection.