BoreTHM: models of circulating temperature and borehole stability for drilling into superhot rock

BoreTHM is an open-source Python model for assessing drilling programmes in superhot geothermal rock. It couples circulating-fluid and downhole-tool temperatures with thermal stress at the borehole wall to calculate the mud-weight window: the range of fluid densities that meets the pressure constraints through drilling, connections and trips.

BoreTHM is intended for drilling engineers and researchers assessing a well design or reusing the thermal and stability calculations. Its reference cases support comparisons with field observations and published models.

• Documentation: guides, scientific explanations and the Python interface.
• Get started: set up Python and run an assessment.
• Read a result: understand the outputs and status labels.
• Explore the examples: site findings and their controlling constraints.
• Understand the models: methods and field comparisons.

Get started

Install BoreTHM with Python 3.11 or later:

python -m pip install borethm

For figures, install "borethm[plot]". Build the water-property cache and run a first assessment:

borethm-water-table
borethm-evaluate --site newberry --json newberry.json

The IAPWS-95 cache is built once, outside the installed package, in the platform's user cache directory. Set BORETHM_CACHE_DIR to choose another location. The evaluator also builds a missing cache on first use. The assessment command prints its result and writes JSON with the reported site inputs, evaluated checks and runtime provenance. Evaluations take several minutes; weighted-fluid cases require repeated coupled solves and can take longer.

The same assessment through the public Python interface:

from borethm import evaluate, export_result, find_site
site = find_site("newberry")
result = evaluate(site)
export_result(result, "newberry.json")
print(result["status"]["label"])

Reproduce the reference calculations

Use Python 3.14.4 and the locked dependencies in a source checkout:

git clone https://github.com/ldocallaghan/BoreTHM.git
cd BoreTHM
git checkout v1.5.0
python -m venv .venv

Use python3 if your system does not provide python. Activate the environment with .\.venv\Scripts\Activate.ps1 in PowerShell, or source .venv/bin/activate on Linux and macOS, then install:

python -m pip install -r requirements-lock.txt
python -m pip install --no-deps -e .
borethm-water-table

Commands for releases before v1.4.0 are in their release records and run from that release's tag.

Run your own assessment

Create a site file, edit its inputs and check them before running the assessment. These commands work from any directory after installation:

python -m pip install borethm
borethm-new-site mysite.toml
# edit mysite.toml
borethm-evaluate --site-file mysite.toml --check
borethm-evaluate --site-file mysite.toml --json mysite.json

The starter contains Soultz's inputs. Replace the temperature profile, target, stresses, rock properties and sources with the site's measurements or stated assumptions. The comments explain units and accepted ranges; the site-input guide describes the file format and provenance labels. --check reports input warnings and the evidence tier without running the model or building the water-property cache.

Use borethm-new-site --from newberry mysite.toml to start from another built-in site. The available keys are soultz, united-downs, larderello, pannonian and newberry. Existing files require --force to overwrite.

The same steps through the public Python interface:

from borethm import load_site, validate_site, evaluate, export_result

site = load_site("mysite.toml")
for warning in validate_site(site):
    print(warning)
result = evaluate(site)
export_result(result, "mysite.json", site_file="mysite.toml")

Keep the file unchanged between loading and exporting through this Python example. The CLI captures the input checksum before evaluation. Keep the site file and measured data with the result: JSON retains supplied provenance and input warnings, with the file path and SHA-256 when supplied, but does not contain the complete site file or programme definition.

To evaluate a built-in site directly, run borethm-evaluate --site soultz --json soultz.json. borethm-sites reports all five. Use borethm-evaluate --help for the options.

What you can change

A site file supplies geology and site inputs. The evaluator selects the well geometry, pipe, flow and drilling cycle through its reference rules.

Reading a result

Start with the reading at the top of the report. It describes numerical validity, the stress and strength cases at the target, the whole-hole pressure limits and the equipment checks. Its controlling constraint identifies the first status reason; follow that into components, the density bounds in windows, and the detailed report. The same text is exported under reading in JSON.

For example, the Newberry reference assessment gives:

Calculation: Valid: the reported numerical checks converged.

Scenarios at the target: Cases passing target stability, including stated fluid conditions: drilling 6/6; connection 6/6; connection, fresh rock 6/6; trip 6/6.

Whole programme: One drilling fluid and one trip fluid satisfy the pressure limits at every evaluated depth, state and case. Surface densities at 20 C: drilling fluid 0.998 to 1.237 SG; trip fluid 0.998 to 1.233 SG.

Equipment: The tool passes its 200 C temperature check. The TK-Drakon coating reaches 139.4 C against its stated 204 C rating (within rating).

Controlling constraint: The status lists no additional constraint.

The output guide explains the four entries, status precedence, pressure units and schedule identity. The full report and JSON retain the individual evaluated cases and supplied input provenance.

An overall NO-GO can coexist with an INDETERMINATE coating check. Exceeding a manufacturer's stated coating rating identifies a qualification question. It does not predict a particular failure mode.

A mud-weight window is the allowable range of fluid density: enough pressure to keep the borehole stable while staying below the modelled fracture-pressure limit. BoreTHM calculates the range that satisfies all evaluated depths, operating states and stress/strength cases, separately for drilling and trips.

Two pressure descriptions appear in the outputs:

• Target SG-equivalent expresses pressure as an equivalent static column at the target. It is a local stability diagnostic, with the existing friction/backpressure treatment.
• Surface SG at 20 °C is fluid density divided by 1,000 kg/m³. The combined pressure constraints set the allowable range of that surface density.

A positive target interval does not establish a workable whole-hole fluid. If the required minimum density exceeds the permitted maximum, no mud density satisfies both pressure limits. The densities reported in that situation are calculation inputs and cannot be prescribed.

Example results

The reference assessments target rock at 400 °C. The table reports the base case with immediate breakout formation. Each status applies to the specified design and inputs.

These sites have different evidence quality. Soultz and United Downs use measured or calibrated stresses and strengths, extrapolated to the targets. Larderello and Pannonian retain unsourced strength assumptions. Newberry's well check does not reproduce the observed pattern in every rock type. The input-source records explain those differences.

Local stability and the complete programme

At Soultz and United Downs, some stress and strength cases have an open drilling window at the target depth. Neither reference programme has a fluid density that satisfies the pressure limits throughout the hole. The reference-programme comparison gives the case tables, whole-hole pressure bounds and equipment temperatures.

Breakout timing

The assumed time between rock exposure and breakout formation can change the whole-hole drilling-fluid window. In the saved scan, that window first opens between delays of 15 and 20 minutes at Soultz, and between 10 and 15 minutes at United Downs. These are brackets from calculations with assumed delays. Breakout formation time in hot granite has not been measured in the studies reviewed here. Both complete programmes remain NO-GO because other checks still fail. The breakout-timing study explains the assumptions and results.

Design comparisons

No evaluated design change or evaluated pair made either deep-site programme pass. The design comparisons describe the tested settings and remaining failures. They do not establish a complete operating envelope or show that every possible design would fail.

The study summary, also available as a two-page PDF, brings the findings and open questions together.

Models and evidence

The circulation model solves the downgoing and returning fluid temperatures with frictional heating and heat transfer through the pipe, casing and surrounding rock. Rock conduction carries each depth through its exposure and circulation pauses. Fluid properties and pressure losses feed back into the stability and tool calculations.

The circulation/FORGE comparisons, strength calibration and other checks have distinct scopes. They do not validate a complete 10-13 km drilling programme. Governing equations, implementation assumptions and diagnostic figures are in the physical-model guide; the bibliography, material properties and site inputs retain their sources.

Scope and limitations

A passing result depends on the modelled inputs and checks. A failed assessment does not establish that a different design or existing technology could not drill the site.

The deep targets lie below measured stress data. Strength calibration depends on the assumed stress profile, and the reference mud's viscosity and constant barite properties are approximations. Qualification of insulated pipe and managed-pressure equipment over the full hot drilling cycle remains unresolved.

The base case assumes that breakouts form immediately after rock exposure, before cooling can reduce the initial pressure requirement. Experiments with an assumed delay show how that requirement changes; they do not demonstrate that breakouts wait an hour to form.

The approximation comparisons test individual assumptions and numerical resolution. None changes either deep-site programme to a pass, but the study does not cover every combination of uncertainty. The limitations guide explains the input uncertainties and the approximations in coolant delivery, trip timing and fluid displacement.

Tests and reproducibility

Run the fast suite in the activated environment:

python -m pytest

The default excludes physical regressions marked slow. To include the complete suite after building the water-property cache:

python -m pytest -m "slow or not slow"

Full physical runs can take an hour or more. Single-threaded BLAS avoids oversubscription during repeated solves; set OPENBLAS_NUM_THREADS=1, OMP_NUM_THREADS=1 and MKL_NUM_THREADS=1 before starting Python for those runs.

Saved reference outputs and numerical regression tables support comparison with the published calculations. The changelog records changes. Historical diagnostic figures use their stated programmes and should be interpreted with their captions.

Find your way around the repository

How to cite and data credits

Cite BoreTHM using the software DOI. CITATION.cff provides the citation metadata. The scientific bibliography lists the scientific sources.

The benchmark dataset is published separately as version 1.0 on Zenodo.

The FORGE data come from the Geothermal Data Repository: well 16B drilling data and deep-well temperature surveys. The Newberry extracts come from well 55-29 stimulation data. These datasets are used under CC-BY 4.0. The UD-1 image interpretation contains NERC materials ©NERC 2024.

The FORGE and Newberry data guides describe the extracts and how to rebuild them. Other input sources are recorded alongside the corresponding data.

Questions and contributions

Use GitHub issues for questions and problem reports. See SUPPORT.md for what to include and CONTRIBUTING.md for development setup and proposing changes.

Licence

The code in src/, scripts/, tools/, tests/ and examples/ is released under the Apache License 2.0. The documentation, figures, results and summary PDF are released under CC BY 4.0. Data in src/borethm/data/ and benchmarks/data/ keep their source licences, as recorded in their data guides and in NOTICE. Releases up to v1.3.4 were published under the MIT licence.