Identifying ultramafic rocks using artificial neural network method based on aeromagnetic data

Abstract

Copper‑nickel (CuNi) resources represent strategically critical mineral commodities facing global supply shortages, making their exploration particularly valuable. Magmatic CuNi sulfide deposits typically occur within ultramafic rock complexes that exhibit distinct magnetic signatures. High-resolution aeromagnetic surveys have proven effective in mapping the spatial distribution of these magnetic anomalies associated with ultramafic lithologies. Current interpretation methodologies predominantly rely on empirical expert analysis for lithological delineation, particularly when identifying concealed intrusions. Nevertheless, systematic approaches employing intelligent algorithms for automated detection of ultramafic rocks through aeromagnetic data analysis remain underdeveloped. This research introduces an Artificial Neural Network (ANN) method for ultramafic rock mapping based on aeromagnetic data. The processed magnetic data is normalized to a similar range and utilized as input feature vectors into machine learning models. After obtaining the final model parameters through training the known ultramafic rock data, the fully connected neural network model predicts the distribution of ultramafic rocks in the unknown region. Various theoretical models were designed to calculate magnetic datasets and test the regularity of data processing and effectiveness of model prediction. The results suggested that data normalization and the selection of feature vectors significantly influenced the prediction results. The prediction accuracy and stability of this method were tested under different spatial resolutions and noise levels. At last, the method was applied in the Northern Qilian area, China. The accuracy of predicted ultramafic rocks is up to 80 % compared with the expert interpretation results. Particularly, two predicted ultramafic rock masses were confirmed by field investigation, which proved the efficiency of this method. The prediction results presented in this paper can provide an objective basis for the delineation of ultramafic rocks, as well as further concentrate the target area for CuNi deposit prospecting.