Physics-constrained, surrogate-assisted inversion for thermophysical calibration and basin-scale present-day temperature-field reconstruction in the Sichuan Basin

Reliable reconstruction of the basin-scale present-day temperature field is essential for geothermal resource assessment and deep subsurface thermal analysis, but it remains challenging because well-temperature observations are unevenly distributed and layerwise thermophysical properties are insufficiently constrained. To address this issue, a well-temperature-constrained, physics-based surrogate-assisted inverse calibration framework was developed to reconstruct the present-day temperature field of the Sichuan Basin. The workflow integrates terrestrial heat-flow data, stratigraphic-interface data, prior thermophysical properties, and multi-source well-temperature observations, including drill stem test temperatures (DSTs), static state temperature (SSTs), and bottom-hole temperatures (BHTs). A one-dimensional layered steady-state conductive heat-transfer model was used as the physical forward model. Random Forest (RF), multilayer perceptron (MLP), and XGBoost were used as alternative surrogate regressors to accelerate parameter-space exploration rather than as direct empirical temperature predictors. Each candidate parameter set was ultimately evaluated using the physics-based forward model, and the calibrated thermal conductivity and radiogenic heat production were then propagated basin-wide for temperature-field reconstruction. Compared with the prior physics-based model, the surrogate-assisted inversion schemes produced moderate but consistent improvements in prediction performance. Among the tested schemes, XGBoost showed the most balanced overall behavior, reducing the test-set RMSE from 10.34 °C to 9.76 °C and increasing the proportion of predictions within ±5 °C from 32.69% to 39.40%. The calibrated thermophysical column preserves clear stratigraphic heterogeneity, with calibrated thermal conductivity ranging from 1.53 to 5.04 W/(m·K) and radiogenic heat production ranging from 0.46 to 5.52 μW/m 3 . These results indicate that the prior thermophysical structure is broadly reasonable but requires formation-dependent correction to better reconcile modeled temperatures with well observations. The reconstructed temperature field reveals a stable regional thermal zonation across the Sichuan Basin, characterized by persistently higher temperatures in the southwestern basin, relatively cooler conditions in the eastern and southeastern structural domains, and an intermediate transition zone in the central basin. The southwestern Sichuan Basin is therefore identified as a priority sector for deep geothermal screening, while the reconstructed present-day temperature field may also provide thermal boundary conditions for future integrated subsurface and petroleum-system studies.