Most of the useful work in our satellite pipeline is done by code we did not write. The spectral index layer comes from eemont and ee_extra, created by David Montero Loaiza and the eemont contributors, built on the Awesome Spectral Indices catalogue. We maintain a small fork. The fork is not the contribution; it is an adapter.
What eemont Actually Solves
Computing a vegetation index in Earth Engine sounds trivial and is not, at scale.
NDVI is straightforward. But there are hundreds of published indices, each with its own formula and band requirements, and every satellite names its bands differently — Sentinel-2's near-infrared is `B8`, Landsat 8's is `SR_B5`, and a formula written for one does not run on the other.
| Without a library | With eemont / ee_extra |
|---|---|
| Hand-code each formula | Reference the index by name |
| Remap bands per sensor | Handled by the sensor abstraction |
| Track formula provenance yourself | Catalogue carries the reference |
| Errors are silent | Formula is canonical |
Concretely, the formula and its band requirements come from the catalogue rather than from memory:
from forestry_carbon_arr.utils.spectral_indices import SpectralIndicesUtils
idx = SpectralIndicesUtils() # loads the catalogue from ee_extra
print(len(idx.spectral_indices_list)) # hundreds of indices, not one
info = idx.get_index_info("SAVI")
print(info["formula"]) # canonical formula, with its referenceThe alternative is a hand-typed formula with a band index nobody rechecks.
That last row is the one that matters for carbon work. A hand-coded index with a transposed band is not obviously wrong — it produces a plausible-looking raster. Using a catalogue-backed formula removes a class of silent error from the pipeline.
What Our Fork Adds
One thing: a band-name adapter between the upstream format and the band naming used in our OSI pipeline.
That is a compatibility shim, worth perhaps a few dozen lines of real logic. It exists because our pipeline had a naming convention before we adopted eemont, and rewriting the pipeline would have been more disruptive than writing the adapter.
Calling that a contribution to spectral index science would be dishonest. The science, the catalogue and the implementation are upstream.
Why Say So Explicitly
Two reasons.
Open-source attribution is not optional courtesy; it is the mechanism by which maintained libraries continue to be maintained. Work that gets quietly absorbed into commercial pipelines without credit stops being worth doing, and the ecosystem everyone depends on thins out.
And a company that overstates what it built invites the question of what else it overstates. In a field where the entire product is a defensible number, that is not a reputation worth acquiring for the sake of a line in a capability deck.
The General Position
Our stack is mostly other people's work: Earth Engine, the Hansen Global Forest Change dataset, published allometric models, IPCC default factors, the Python geospatial ecosystem, and eemont among them.
What is ours is narrower and, we think, more valuable: field data collected to a documented protocol, local calibration and validation, and the integration that makes the pieces work together on Indonesian conditions.
If you are building something similar, use eemont directly rather than our fork. It is better maintained, more general, and it is the real project. Credit where it belongs: eemont by David Montero Loaiza.
Turn climate goals into a verified portfolio
TREEO connects the full carbon cycle — eligibility, simulation, real-time monitoring and registry-ready reporting — combining expert consulting with dMRV technology so the evidence exists before anyone asks for it.


