Restored peatlands take decades to recover, but management decisions can't wait for annual surveys. MAZE produces 96-hour net ecosystem exchange forecasts for any UK peatland — calibrated to the European flux tower network, validated against ground truth, delivered without on-site instrumentation. Catch heatwave-driven respiration before it happens. See your site breathing, hour by hour.
A typical restoration site is monitored through annual or biennial chamber surveys, with results returned months later. By then, the management window for responding to a dry summer or a hydrological intervention has closed.
Eddy-covariance towers produce the gold standard for site-level carbon flux, but cost £50,000–100,000 to install and require specialist maintenance. Fewer than 20 operate across the UK — none on most active restoration projects.
Carbon flux is most volatile and most consequential during summer growing seasons. Without continuous forecasting, anomalous flux events — heatwave-driven respiration, hydrological tipping points — are only discovered after the fact, in retrospective analyses.
The UK has world-leading peatland flux research at Auchencorth Moss, Forsinard, and a handful of others. The 99% of restoration projects without research infrastructure rely on emission-factor estimates and visual inspection.
Draw a site polygon, classify the peatland (blanket bog, raised bog, fen), and we pull historical flux dynamics for the site's ecosystem class. No on-site instrumentation required to get started.
No on-site instrumentation requiredA fine-tuned 80M-parameter time-series foundation model produces a 96-hour hourly NEE forecast. Pre-trained on diverse temporal signals across multiple domains, adapted on European peatland flux tower records, and validated against held-out FLUXNET2015 sites.
Validated R² = 0.68–0.74 on held-out sitesEnsemble disagreement provides per-hour 95% confidence intervals. High-uncertainty regimes — typically summer high-VPD days — are flagged explicitly. Track day-to-day changes, catch anomalies, support management decisions with evidence.
Calibrated uncertainty boundsThese are live 96-hour forecasts from MAZE's fine-tuned TEMPO-80M models, benchmark-validated against held-out FLUXNET2015 eddy-covariance sites. The preset markers return instantly from pre-computed forecasts; draw your own site on the map and you're watching real inference run — up to ~90 seconds on our CPU instance.
MAZE's predictor is a fine-tuned TEMPO-80M time-series foundation model, evaluated for cross-ecosystem carbon flux forecasting across seven European FLUXNET sites including held-out wetland and forest reference sites.
MAZE's carbon flux models are validated against held-out FLUXNET2015 eddy-covariance sites, reproducing measured net ecosystem exchange to R² = 0.60 (same-ecosystem wetland, UK-AMo) and R² = 0.73 (cross-ecosystem forest, SE-Htm) at the 96-hour horizon. Forest-TEMPO lifts forest performance to R² = 0.77.
The current product delivers 96-hour operational forecasts validated against the published benchmark. Extending the methodology to a defensible annual carbon balance — the Peatland Code use case — is on the roadmap, planned as a separate product line.