RT Journal Article SR Electronic(1) A1 Zhang, Si-Hai A1 Xu, Yin A1 AbuAli, MahdiYR 2021 T1 Thin sand identification in complex depositional environment by supervised artificial neural networks JF First Break, VO 39 IS 11 SP 81 OP 88 DO https://doi.org/10.3997/1365-2397.fb2021087 PB European Association of Geoscientists & Engineers, SN 1365-2397, AB Abstract The thin sands within the study interval developed in a complex fluvial-to-shallow marine environment. This variety results in extremely heterogeneously distributed thin sands. The objectives are to identify heterogeneous thin sands via machine learning and evaluate the impact of tuning thickness on the recognition. Multi-attribute classification using supervised ANNs is employed to predict the distribution of these thin sands within six subintervals. Before the prediction, the investigation of the six subinterval shows that the major part of each subinterval is greater than the tuning thickness and can be resolved by the available seismic data. An integrated workflow of multi-attribute classification based on supervised ANNs was established. The supervised classification cannot only add significant details and enhanced lateral resolution, but also allows the interpreters to avoid manual labelling. The thin sands of the six intervals predicted by the supervised ANNs are verified qualitatively and quantitatively. The predicted sand thickness is validated by a log-based thickness map and shows similar distribution of thin sands. The cross plots of the sand thickness between the seismic predictions and the log measurement at well locations quantify the seismic prediction. Therefore, both verifications show seismic prediction characterizes the thin sands in the study area and supervised ANNs have great potential in solving the challenge of thin bed identification. The positive correlation of the coefficients between the predicted and log-based thickness at the well locations and the average thickness of the subintervals suggest that seismic prediction still depends on the tuning thickness though multi-attribute classification based on the supervised ANNs used., UL https://www.earthdoc.org/content/journals/10.3997/1365-2397.fb2021087