Machine Learning Based Subsurface Temperature Forecasting to Reduce Drilling Uncertainty in Geothermal Systems

Subsurface temperature uncertainty represents a major risk in geothermal drilling, particularly in volcanic systems where permeability, lithology, and fluid circulation create highly heterogeneous thermal regimes. This study presents a hybrid physics-informed machine learning framework for forecasting subsurface temperature using geothermal drilling and geophysical log data from Icelandic geothermal fields. A publicly available dataset from the GEOTHERMICA/RESULT project comprising 16 deep geothermal wells from the Elliðaár geothermal field was used for model development and validation. Ensemble and neural network models were optimized using Bayesian hyperparameter tuning and evaluated against conventional geothermal gradient methods. The present study represents a proof-of-concept demonstration of the proposed framework. Ongoing work is focused on expanding the dataset to 52 geothermal wells and enabling real-time deployment for geothermal drilling operations.