Published October 6, 2026 | Version v1
Dataset Open

Data, Code, and Model Weights Release: In-Context Earth

Description

This release accompanies the arXiv preprint In-context learning enables continental-scale subsurface temperature prediction from sparse local observations (https://arxiv.org/abs/2605.16665). It provides bottom-hole temperature datasets, train/test splits, model configuration information, and trained PyTorch model weights for the In-Context Earth model, a lightweight transformer-based approach for predicting subsurface temperature from sparse local borehole observations. The dataset contains approximately 650,000 reorganized public borehole temperature measurements from the contiguous United States, Alberta, Australia, and the United Kingdom, along with random, spatially blocked, and no-borehole-contamination train/test splits. The release is intended to support reproduction of the associated experiments, benchmarking of interpolation and machine-learning methods for geothermal temperature prediction, evaluation of spatial generalization, and development of uncertainty-aware subsurface temperature models.

Files

tlr-snapshot.zip

Files (229.1 MB)

Name Size
md5:06c040b18bb1c4c9c0653df6541de435
3.4 MB Download
md5:da7183163ab566e6d30426ebd8712d03
2.7 MB Download
md5:bb45d95cdf97b3630c638aaf99edd3b9
28.9 MB Download
md5:e714cfcfa4972e649b3c50fe36c642f7
3.1 MB Download
md5:2a0cd1ad7ccef34433916e6c7392cc6f
28.5 MB Download
md5:7bf65c8212d34e12ba00904d17b780b0
3.2 MB Download
md5:3ed4cda7603445050fdd6f144441d2d4
28.5 MB Download
md5:a3e141bf722a194d5e105d3f43981b2a
1.0 MB Download
md5:15db0fb885ff5f3cfe69f6861799d629
13.8 MB Download
md5:d8fe410eb68bcfcbe60b5618a5172e9a
1.7 MB Download
md5:795ebac619084e2af1cdfe5ba96a061e
15.0 MB Download
md5:4417df017fb8bf47f5ac35ed4f76b3ad
104.9 kB Download
md5:ab7e1a94aed1833875288d9d298c09d5
3.5 kB Download
md5:88c459f9e7d591071d212c52b667481c
99.0 MB Preview Download