Machine learning based bottom hole temperature prediction using well log data in the Taranaki Basin New Zealand

Accurate estimation of bottom hole temperature (BHT) is essential for optimizing drilling operations, reservoir characterization, and petrophysical assessments. This study focuses on the development of various machine learning (ML) models for predicting BHT. The well log data from the Taranaki Basin, New Zealand, is used for the model development. After data acquisition and preprocessing, a feature selection process using correlation analysis identified the most relevant input feature logs. Six ML models—Multiple Linear Regression, Random Forest, XGBoost, AdaBoost, Gradient Boosting, and a Voting Regressor—were developed and evaluated using blind-test data. This study compares the machine learning models to predict the BHT using well log data from the Taranaki Basin, providing one of the first basin-specific assessments of model performance and prediction refinement through exponential smoothing. Model evaluation was conducted on three datasets: (1) 20% random test split, (2) unseen data from the Goldie-1 well (blind-test well), and (3) the same well (Goldie-1) after exponential smoothing of the predicted values. The Random Forest consistently delivered significant accuracy, an R 2 score of 0.990 on blind testing, a mean absolute error (MAE) of 0.38 °C, a root mean squared error (RMSE) of 0.77 °C, a mean absolute percentage error (MAPE) of 0.01%, and a maximum error of 8.94 °C. After applying exponential smoothing on predicted data, the error further decreased with an MAE of 0.33 °C, an RMSE of 0.45 °C, a MAPE of 0.01%, and a maximum error of 1.31 °C. The ensemble methods, particularly Voting Regressor, demonstrated comparable performance, achieving an R 2 score of 0.974 on blind test data, an MAE of 1.09 °C, an RMSE of 1.27 °C, an MAPE of 0.02%, and a maximum error of 3.23 °C. The findings of this study indicate the effectiveness of machine learning algorithms and the workflow for predicting subsurface temperature (BHT) with significant accuracy.

Authors

Institutions

Publication Details

Journal
Scientific Reports
Published
2026-11-17
DOI
https://doi.org/10.1038/s41598-026-63064-3
Primary Topic
Drilling and Well Engineering
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Machine learning based bottom hole temperature prediction using well log data in the Taranaki Basin New Zealand

Sarthak Singh, Shubham Anand
2 links
Scientific Reports
Drilling and Well Engineering
article

Machine learning based bottom hole temperature prediction using well log data in the Taranaki Basin New Zealand

Sarthak Singh, Shubham Anand
article en

Abstract

Accurate estimation of bottom hole temperature (BHT) is essential for optimizing drilling operations, reservoir characterization, and petrophysical assessments. This study focuses on the development of various machine learning (ML) models for predicting BHT. The well log data from the Taranaki Basin, New Zealand, is used for the model development. After data acquisition and preprocessing, a feature selection process using correlation analysis identified the most relevant input feature logs. Six ML models—Multiple Linear Regression, Random Forest, XGBoost, AdaBoost, Gradient Boosting, and a Voting Regressor—were developed and evaluated using blind-test data. This study compares the machine learning models to predict the BHT using well log data from the Taranaki Basin, providing one of the first basin-specific assessments of model performance and prediction refinement through exponential smoothing. Model evaluation was conducted on three datasets: (1) 20% random test split, (2) unseen data from the Goldie-1 well (blind-test well), and (3) the same well (Goldie-1) after exponential smoothing of the predicted values. The Random Forest consistently delivered significant accuracy, an R 2 score of 0.990 on blind testing, a mean absolute error (MAE) of 0.38 °C, a root mean squared error (RMSE) of 0.77 °C, a mean absolute percentage error (MAPE) of 0.01%, and a maximum error of 8.94 °C. After applying exponential smoothing on predicted data, the error further decreased with an MAE of 0.33 °C, an RMSE of 0.45 °C, a MAPE of 0.01%, and a maximum error of 1.31 °C. The ensemble methods, particularly Voting Regressor, demonstrated comparable performance, achieving an R 2 score of 0.974 on blind test data, an MAE of 1.09 °C, an RMSE of 1.27 °C, an MAPE of 0.02%, and a maximum error of 3.23 °C. The findings of this study indicate the effectiveness of machine learning algorithms and the workflow for predicting subsurface temperature (BHT) with significant accuracy.

Scientific Reports
Indian Institute of Technology Dhanbad (IN), University of Petroleum and Energy Studies (IN)
Drilling and Well Engineering
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon