Advancing Radiometric Mapping in Australia: Bridging Gaps and Enhancing Mineral Exploration
Published by Geoscience Australia Data
About this dataset
This study advances radiometric mapping in Australia by applying machine learning and deep learning to enhance potassium concentration predictions from airborne gamma-ray spectrometric surveys. Addressing gaps in national radiometric data, which are caused by logistical and regulatory challenges, impacts effective environmental management and mineral exploration. Traditional interpolation methods like Inverse Distance Weighting and Kriging are compared with advanced machine learning algorithms, including Random Forest and XGBoost, and deep learning models such as Simple Net and Transformer. The research uses a dataset integrated with 25 geospatial layers to evaluate the models across various data gap scenarios. An important finding is the challenge in predicting extreme values, where both traditional and some machine learning methods often underperform due to premature convergence at local minima within non-convex optimization landscapes. Future improvements might include utilizing sophisticated initialization schemes and adaptive learning rates to enhance the capability of deep learning models in handling these extremes, potentially significantly improving predictive accuracy over traditional methods. Presented at the IGARSS 2025 - IEEE International Geoscience and Remote Sensing Symposium.