SCOPE-RTM 0.4.7

PyTorch reimplementation of the SCOPE radiative transfer model.

SCOPE

PyTorch-first implementation of the SCOPE canopy radiative transfer model for reflectance, fluorescence, thermal radiance, and coupled energy-balance workflows.

What It Is

scope is designed for users who need:

• asset-backed SCOPE physics in Python
• batched ROI/time execution on xarray datasets
• differentiable model components in PyTorch
• reproducible MATLAB parity checks in CI and local development

The current implementation supports:

• leaf optics through FLUSPECT
• canopy reflectance through 4SAIL-based transport
• layered fluorescence and thermal radiative transfer
• leaf biochemistry and coupled energy balance
• directional and vertical-profile outputs on the homogeneous canopy path
• ROI/time workflows with xarray input and output assembly

Attribution

This package is a Python implementation of the original MATLAB SCOPE model:

• Soil Canopy Observation, Photochemistry and Energy fluxes (SCOPE)
• original repository: Christiaanvandertol/SCOPE
• upstream manual: scope-model.readthedocs.io

Please attribute the original SCOPE model and papers when using this package in research workflows:

• Van der Tol, C., Verhoef, W., Timmermans, J., Verhoef, A., and Su, Z. (2009), Biogeosciences 6, 3109-3129
• Yang, P., Prikaziuk, E., Verhoef, W., and Van der Tol, C. (2021), Geoscientific Model Development 14, 4697-4712

Install

Published package name:

python -m pip install SCOPE-RTM

Import name:

import scope

Top-level CLI:

scope --help
scope fetch-upstream --help
scope prepare --help
scope run --help
scope vars Cab
scope vars --workflow fluorescence
scope vars --related Rntot

1. Clone the repository

git clone https://github.com/MarcYin/SCOPE scope
cd scope

2. Fetch the pinned upstream SCOPE assets

python scripts/fetch_upstream_scope.py

If you installed the package in an environment already, the same helper is available as:

scope-fetch-upstream

3. Create an environment and install

python -m venv .venv
source .venv/bin/activate
python -m pip install --upgrade pip
python -m pip install -e ".[dev]"

4. Verify the install

PYTHONPATH=src python examples/basic_scene_reflectance.py
PYTHONPATH=src python -m pytest -q tests/test_scope_benchmark_parity.py tests/test_scope_timeseries_benchmark_parity.py

5-Minute Quickstart

Minimal scene reflectance run

PYTHONPATH=src python examples/basic_scene_reflectance.py

Expected output:

{
  "product": "reflectance",
  "dims": {"y": 1, "x": 1, "time": 1, "wavelength": 2001},
  "rsot_650nm": 0.047138178221010914,
  "rsot_865nm": 0.4100649627325952,
  "rsot_1600nm": 0.26994893328935227
}

High-level workflow run

PYTHONPATH=src python examples/scope_workflow_demo.py

Expected output:

{
  "product": "scope_workflow",
  "components": [
    "reflectance",
    "reflectance_directional",
    "reflectance_profile",
    "fluorescence",
    "fluorescence_directional",
    "fluorescence_profile"
  ],
  "rsot_650nm_t0": 0.04522854188089004,
  "LoF_peak_t0": 1.985767010834904e-05,
  "LoF_peak_wavelength_t0": 744.0
}

Prepared-dataset CLI run

scope prepare \
  --weather weather.nc \
  --observation observation.nc \
  --bio-npz post_bio.npz \
  --year 2020 \
  --output scope_inputs.nc

scope run \
  --input scope_inputs.nc \
  --output scope_outputs.nc \
  --scope-root ./upstream/SCOPE \
  --workflow reflectance

Repeat --output-var to write only selected output variables for large runs.

Main Entry Points

For most users, the preferred entry points are:

• ScopeGridRunner.run_scope_dataset(...) High-level reflectance/fluorescence/thermal workflow dispatch from prepared xarray inputs.
• ScopeInferenceModel Lightweight tensor-only inference surface for production workloads that only need selected outputs.
• prepare_scope_input_dataset(...) Build a runner-ready dataset from weather, observation, and Sentinel-2 bio inputs.
• validate_scope_dataset(...) Validate required variables, soil alternatives, and key dimensions before a workflow is executed.
• write_netcdf_dataset(...) Persist prepared or simulated outputs to NetCDF with safe backend selection and compression handling.

For direct lower-level use:

• FluspectModel
• CanopyReflectanceModel
• CanopyFluorescenceModel
• CanopyThermalRadianceModel
• CanopyEnergyBalanceModel

Documentation Map

• Installation Guide
• Quickstart
• Model Mechanics
• Input / Output Reference
• Variable Glossary
• Workflow Variable Guides
• Examples
• Production Notes
• Releasing
• Benchmark Policy

Build the docs locally with:

python -m pip install -e ".[docs]"
mkdocs build --strict

Production Notes

• Asset-backed constructors such as from_scope_assets(...) require an upstream SCOPE checkout. The recommended path is scope-fetch-upstream.
• The installed CLI now covers the common shell workflow: scope fetch-upstream, scope prepare, and scope run.
• Prepared inputs and assembled outputs now carry glossary-derived xarray metadata such as long_name, units, description, scope_category, and scope_relationship.
• NetCDF exports are now CF-enriched with dataset-level Conventions, title, source, references, history, and axis metadata on common coordinates.
• scope run validates workflow-specific inputs before execution, and the same validator is available directly as validate_scope_dataset(...).
• The default CI suite runs parity tests in live-or-pregenerated mode. On machines without MATLAB, the tests compare against checked-in MATLAB fixtures.
• The self-hosted GPU and live-MATLAB lanes remain optional operational lanes; see docs/benchmark-policy.md.
• Documentation can be built locally with mkdocs build --strict and is deployed by the dedicated GitHub Pages workflow.
• Distribution artifacts can be built locally with python -m build and validated with python -m twine check dist/*.
• Release notes are drafted automatically on main, and tagged releases publish PyPI artifacts plus GitHub artifact attestations.

Testing

Run the default suite with:

PYTHONPATH=src python -m pytest -q

The strongest automated checks currently include:

• kernel parity and execution-mode regression tests
• ROI/time runner consistency tests
• committed scene and time-series benchmark summary regression tests
• live-or-pregenerated MATLAB parity tests for the single-scene and time-series benchmark gates

Performance Benchmarking

Use the committed kernel benchmark harness to compare eager and compiled execution on your own hardware:

PYTHONPATH=src python scripts/benchmark_kernels.py \
  --device cpu \
  --dtype float64 \
  --batch 32 \
  --fixture scope-assets \
  --mode compare

Current reference behavior on CPU with torch 2.10.0:

• fluspect and reflectance show strong steady-state speedups under torch.compile, but still require repeated same-shape calls to amortize compile cost.
• thermal speeds up in steady state, but the compile break-even is much higher.
• layered fluorescence currently fails under torch.compile on this environment.
• leaf_biochemistry currently becomes slower under torch.compile because of scalar-control-flow graph breaks and recompilation churn.

Because of that mix, the package does not enable compiled execution by default.

Release Workflows

• .github/workflows/release.yml Verifies tag/version alignment, reruns the release-local CPU/docs gates, builds sdist and wheel artifacts for SCOPE-RTM, validates them with twine check, smoke-installs both artifact types through the documented scope CLI surface, and then publishes to PyPI on version tags. Manual dispatch still supports TestPyPI or PyPI.
• .github/workflows/docs.yml Builds the MkDocs site and deploys it to GitHub Pages.