LEDT — The Smithsonian Living Earth Digital Twin

01 · The Gap

Planetary digital twins are biologically inert.

NVIDIA's cBottle, NASA's ESDT, and the EU's DestinE simulate atmospheric physics at kilometer scale, thousands of times faster than traditional numerical models. Yet tropical forests and coastal wetlands — Earth's most important carbon sinks — are treated as static boundary conditions, not as living, reactive systems with nonlinear feedback loops. Tipping-point behaviors cannot be captured until they are already underway.

⚠

Abrupt forest mortality

Drought-driven die-offs in the Amazon and boreal crowns are unresolved in current physics-only twins.

CH₄

Accelerating wetland methane

Blue carbon ecosystems flip between sink and source under warming — a nonlinearity absent from cBottle.

◐

Ecosystem tipping points

Collapse signatures are detectable in orbital data if you know where to look. We do.

02 · Our Idea

An ecological feedback layer, built from uniquely Smithsonian data.

The Living Earth Digital Twin (LEDT) is the first high-fidelity ecological feedback layer designed to plug into existing planetary twins from NVIDIA, NASA, and DestinE. Rather than duplicate climate models, LEDT occupies a complementary lane — one no single Smithsonian unit can build alone.

How does variability in solar radiation, including extreme events, propagate through the atmosphere and climate system to influence carbon cycling and ecosystem stability across latitudes?

The animating question behind this cross-Unit pursuit.

LEDT links what the atmosphere delivers — solar variability, pollutant loads, and greenhouse-gas concentrations from SAO's TEMPO and MethaneSAT — to how ecosystems respond: carbon flux shifts, species turnover, and tipping-point signatures from STRI's ForestGEO network (70+ permanent tropical forest plots), SERC's Global Change Research Wetland (host of the world's longest-running elevated-CO₂ experiment), and the Coastal Carbon Network (>15,000 blue-carbon observations).

These are world-unique assets under one institutional umbrella — and no other institution can fuse them. The work begins this September.

03 · Innovation Workshop

Sept 14–16, 2026 · SAO + SERC + STRI convene in Cambridge.

LEDT needs its people in one room. Smithsonian scientists from the Astrophysical Observatory, the Environmental Research Center, and the Tropical Research Institute will gather at SAO in Cambridge, MA for a three-day Innovation Workshop to ideate, scope, and begin prototyping the ecological feedback layer — together with partners from NVIDIA Earth-2, NASA ESDT, and DestinE.

⌖

Cambridge is lovely in September

We'll meet at the Smithsonian Astrophysical Observatory, 60 Garden Street, Cambridge, MA. Hotel block + lodging logistics forthcoming.

◎

Sept 14–16, 2026

Three days of working sessions, Monday through Wednesday. An informal welcome drinks reception will kick things off on Sunday evening, Sept 13, and we'll host a big team dinneron Tuesday evening, Sept 15.

✦

Travel fully funded

In-person participation is essential for the cross-unit collaboration we're building. All Smithsonian scientists can have their travel fully funded by the OUSSR seed award, at no cost to your home unit.

Booking deadline · July 25, 2026. Funding restrictions require all travel to be booked by July 25, 2026. If you're ready to book, please reach out to TremblayG@si.edu immediately so we can coordinate your itinerary against the seed-award budget.

04 · Innovation Workshop Registration

Register now. We'll cover your travel.

Sept 14–16, 2026 at SAO in Cambridge, MA, with a welcome reception the evening of Sunday Sept 13 and a great dinner on Tuesday Sept 15. Hotel blocks and lodging info are forthcoming. Smithsonian attendees: travel can be fully funded, but it must be booked by July 25, 2026 due to funding restrictions — email TremblayG@si.edu the moment you're ready.

Name

Email

Affiliation

Comments (optional)

Preferred travel details (optional, but strongly preferred)

Sketch your ideal itinerary — airports, dates, and any lodging needs. For example: "PTY → BOS, depart Sun Sept 13, return Wed Sept 17, govt-rate room needed." This helps us estimate your cost for budgeting, but does not book anything. To start the actual booking you must email TremblayG@si.edu — SAO's travel office can help. All travel must be booked (funding encumbered) by July 25, 2026.

I acknowledge 🫡 This will not be three days of scientists talking at each other. This will be a proper sprint — like a hackathon — and we'll really need you to do a little homework in the months before you depart for the meeting. We'll hold 2–3 virtual mini-prep meetings in July and August to get everyone prepped and ready to rock.

Smithsonian scientists: travel can be fully funded by the OUSSR seed award. Reminder: travel must be booked by July 25, 2026 — email TremblayG@si.edu as soon as you're ready to book.

05 · Live Orbital Chemistry

TEMPO — hourly, North America, from geostationary orbit.

SAO's Tropospheric Emissions: Monitoring of Pollution instrument measures NO₂, O₃, HCHO, aerosols, and solar-induced chlorophyll fluorescence at unprecedented cadence. Below: a live tile layer served by NASA GIBS, overlaid with Smithsonian field network sites.

Species

Date (UTC) Hour of day · 18:00 UTC

Showing: latest TEMPO scan (default)

TEMPO scans E→W across N. America through daylight (≈ 12–23 UTC). A single timestamp only covers a longitude strip — drag the slider to sweep through the day.

Field Network

Column density

lowhigh

Tiles: NASA GIBS / TEMPO L3 · This visualization is illustrative; the LEDT prototype will ingest L2 swaths natively via harmony-py and NASA Earthdata.

06 · Integration Target

NVIDIA Earth-2 — the substrate we plug into.

Earth-2 is NVIDIA's planetary digital twin, powered by the cBottle ("Climate in a Bottle") generative AI foundation model and the earth2studio inference toolkit. It simulates kilometer-scale atmospheric state thousands of times faster than traditional numerical weather prediction. LEDT will ride on top of it, contributing the biological feedback layer Earth-2 currently lacks.

Drop-in diagnostic

LEDT's TEMPO ingest is packaged as an earth2studio-compatible DiagnosticModel — the same interface that CorrDiff and PrecipitationAFNO already use. Once wrapped, fusing orbital atmospheric chemistry into an Earth-2 forecast is a handful of lines:

from earth2studio.models.px import FCN3
from earth2studio.data      import GFS
from earth2studio.io        import ZarrBackend
from ledt.diagnostics       import TEMPONO2Diagnostic
import earth2studio.run as run

prognostic = FCN3.load_model(FCN3.load_default_package())
tempo_dx   = TEMPONO2Diagnostic(tempo_dataset)   # our layer
io         = ZarrBackend("outputs/fcn3_with_tempo.zarr")

run.diagnostic(
    ["2024-08-15T00:00:00"], nsteps=8,
    prognostic=prognostic, diagnostic=tempo_dx,
    data=GFS(), io=io,
)

cBottle

Generative foundation model for global km-scale atmospheric states.

CorrDiff

Diffusion-based regional downscaling. The natural hook point for a CorrDiff-Bio extension.

earth2studio

Open-source Python inference toolkit that glues models, data, and IO. LEDT targets this API.

NVIDIA earth2studio overview · via YouTube

Live · NVIDIA Omniverse GDN

Launch the Earth-2 Interactive Demo

Streamed from NVIDIA's Graphics Delivery Network. Opens in a new tab — NVIDIA's demo is a WebRTC stream and cannot be embedded cross-origin.

nvidia.com/earth-2/demo ↗

07 · Architecture

The LEDT sits between the atmosphere and the biosphere.

Rather than duplicate climate models, we occupy a complementary lane: a biological feedback layer that consumes state from existing planetary twins and returns ecological response — carbon flux shifts, species turnover, tipping-point probabilities — back into the loop.

Atmosphere

TEMPO

NO₂ · O₃ · HCHO · SIF · Aerosol · hourly

MethaneSAT

CH₄ plume flux · legacy maps

Earth-2 / cBottle

km-scale generative atmospheric state

LEDT

AI Feedback Model

Ecological response operators · trained on TEMPO × field time series

Tipping-Point Detector

Astrophysical signal-processing (AstroAI) ported to ecological regime shifts

CorrDiff-Bio

Biogeochemical extension to NVIDIA's CorrDiff downscaling

Biosphere

STRI ForestGEO

70+ tropical plots · tree census · longest tropical dynamic record

SERC GCReW

40 yrs elevated-CO₂ · eddy covariance · tidal wetland

Coastal Carbon Network

>15,000 observations · blue carbon globally

08 · The Smithsonian Advantage

The only institution with orbital + canopy + wetland under one roof.

SAO

Operates TEMPO, the world's first geostationary air-quality instrument, plus MethaneSAT legacy data. Home of the AstroAI and EarthAI centers.

Smithsonian Environmental Research Center

SERC

Global Change Research Wetland hosts the world's longest-running elevated-CO₂ experiment. The Coastal Carbon Network aggregates >15,000 blue-carbon observations.

STRI

ForestGEO network: 70+ permanent tropical forest plots, the richest longitudinal record of tropical forest dynamics on Earth.