TorfSpace

💎 Idea

Europe has lost 80% of the wetlands it had a century ago. Of the protected wetland habitats that remain under EU law, 89% are in unfavourable condition — and 51% are actively deteriorating right now (EEA State of Nature Assessment, 2020). This is not a future projection. It is a measured, ongoing collapse of the continent's largest natural freshwater retention system.

Peatlands sit at the centre of this collapse. In their intact state, peatlands function as landscape-scale water sponges: they absorb rainfall, hold it in their peat matrix, and release it slowly into rivers and aquifers during dry periods. A single hectare of healthy raised bog stores water equivalent to a small reservoir and buffers entire river catchments against drought. When drained — for agriculture, peat extraction, or decades of neglect — that buffering disappears. The land stops holding water. Downstream communities experience more intense floods in wet seasons and deeper droughts in dry ones. Drained peatlands also emit approximately 220 Mt CO₂eq per year, around 5% of total EU greenhouse gas emissions.

Rewetting is the primary fix. The EU's Nature Restoration Regulation (2024/1991) sets legally binding targets: 30% of degraded habitats restored by 2030, 50% by 2040, 70% by 2050, with specific provisions for peatlands. Member States are currently writing their National Restoration Plans. Rewetting projects are being funded. Water is being put back into drained land across the continent.

The problem is evidence. Rewetting without verified, measurable proof that water tables have actually recovered is restoration on paper only. Member States must report biennially on progress. They have no scalable, affordable tool to produce that evidence across the hundreds of sites they will need to monitor. Field dipwell campaigns are accurate but expensive, slow, and cannot scale to EU-wide obligations.

We build the missing evidence layer.

Our system fuses 3 satellite data streams — Sentinel-1 SAR backscatter, Sentinel-2 vegetation and wetness indices, Copernicus DEM microtopography — with ERA5-Land precipitation reanalysis to predict water table depth (WTD) per site, per observation window. The ML model, trained on peer-reviewed piezometric ground truth (Toca et al. 2023, Remote Sensing 15, 1900 — Forsinard Flows, Scotland), converts satellite signals into a measurable hydrological indicator that correlates directly with restoration success.

The output is not a raw satellite product. It is a decision-ready compliance report.

A government official sees a plain-language Hydrological Recovery Score for each registered site — one number, one status, no technical background needed. The interface includes an AI-generated plain-language summary: what the score means, what changed since the last observation, and what uncertainty exists. A domain expert, one click deeper, sees the full explainability layer: which data sources drove the estimate, what weight each carried, what confidence interval applies, and which observations were flagged for cloud cover or sensor dropout. An auditor traces the complete data chain from raw satellite observation to reported compliance indicator.

This explainability architecture is the core competitive differentiator. Existing EO systems for peatland monitoring deliver raw products or consultancy reports — the interpretation burden lands on the user. We invert that. The complexity lives inside the pipeline; the output is transparent, traceable, and immediately actionable. The score is not a black box. It is a documented, weighted sum of verifiable satellite observations — and every layer of that sum is visible to whoever needs to see it.

Scalability makes the economics work. Once calibrated on a site with existing ground truth, the pipeline transfers to new sites in the same biogeographical region without a new field campaign. A Member State registering 50 rewetted sites under its National Restoration Plan does not need 50 monitoring contracts. The system learns from publicly available reference sites and applies to sites with no in-situ infrastructure. Each additional site costs computation, not data acquisition — because Copernicus data is free at point of use.

🛰️ EU Space Technologies

Sentinel-1 (Copernicus) is the backbone of the pipeline. C-band SAR backscatter in VV and VH polarisation penetrates cloud cover and responds to surface soil moisture and near-surface water table conditions in peatlands. Toca et al. (2023) demonstrated statistically significant correlation between Sentinel-1 multi-temporal backscatter composites and dipwell-measured WTD at Forsinard Flows — the only peer-reviewed demonstration of this link at field-relevant scale in temperate European peatlands. We replicate and extend this approach. Sentinel-1A is currently operational; Sentinel-1C, launched December 2024, is ramping up coverage. We access RTC-preprocessed Sentinel-1 data from Microsoft Planetary Computer, eliminating the preprocessing bottleneck for the hackathon build.

Sentinel-2 (Copernicus) provides vegetation and wetness indices: NDWI, NDVI, and canopy water content proxies that capture vegetation response to hydrological state. Sphagnum and sedge communities show distinct spectral signatures at different WTD ranges. Cloud masking is applied before any index extraction. On high cloud-cover scenes, the model runs on Sentinel-1 only with a confidence reduction flag surfaced in the explainability layer — the system tells the user why the estimate is less certain, rather than silently producing a degraded output.

Copernicus DEM provides microtopographic context. Slope and topographic wetness index derived from the DEM are static model features — strong predictors of spatial WTD variation within a site that do not require repeated acquisition.

ERA5-Land (Copernicus Climate Change Service) supplies lagged precipitation and air temperature: 7-day and 30-day precipitation accumulation are key dynamic drivers of WTD fluctuation between observation windows. Frozen-ground periods (T_air < 0°C) are automatically flagged and excluded from inference, as SAR backscatter loses hydrological interpretability on frozen soils — a data quality decision that is surfaced transparently in the output.

The Copernicus programme delivers all primary data at zero marginal cost per scene. This is not incidental — it is the economic foundation of the scalability argument. Monitoring 100 sites costs no more satellite data than monitoring 10.

🌊 EU Space for Water

Challenge #1 — Water Access and Retention.

Europe's freshwater is disappearing faster than most people realise — not only from rivers and aquifers, but from the landscape itself. Peatlands are the continent's largest natural freshwater retention infrastructure. Intact peatland holds water in its peat matrix and releases it slowly, buffering river catchments across weeks and months of dry weather. This is not a metaphor. It is a measurable hydrological function — and it is the function that disappears when peatlands are drained.

Restoring that function is what rewetting achieves. A rewetted peatland that has recovered its water table stops amplifying drought and starts absorbing and holding rainfall again. At landscape scale, across the millions of hectares of degraded peatland that EU Member States are now committing to restore under NRR 2024/1991, this represents a real and quantifiable increase in freshwater retention capacity for the European continent.

Our system makes that increase measurable. Water table depth — the single most important indicator of peatland hydrological function — is predicted per site from satellite data. Each site's Hydrological Recovery Score is a direct indicator of restored water retention capacity: not a proxy, not an index, but a depth estimate in centimetres, translated into a plain-language compliance status.

The secondary application follows from the same hydrological logic: low water tables are the primary driver of peat fire ignition and spread (Kettridge et al. 2015, Scientific Reports 5, 8063). Sites where water table recovery is lagging are also the sites at elevated fire risk. We surface this as a secondary dashboard indicator, drawing on Copernicus CEMS / EFFIS fire monitoring data, so that responsible authorities can prioritise both restoration intervention and fire prevention at the same sites, from the same platform.

TEAM

• Karolina Ossuch — Project management. Student of systems engineering. Project support with data analysis at Nokia, part of the PMO team at Kaizen Institute Poland. Certified in Project Management Fundamentals (PMI) and Digital Skills (Google). Business, marketing, and finance experience through student associations and workshops.

• Aleksander Biskup — Software and data engineer working at the intersection of satellite data and machine learning. A problem-solver who genuinely enjoys finding the leverage point in a complex system, and an enthusiast for emerging technologies — especially when they let us see things we couldn't before. At TorfSpace, leads the satellite data analysis pipeline.

• Filip Antoniak — Developer. Systems Engineer & Developer. Commercial experience as a Django developer in an international environment. He combines a strong aesthetic sensitivity and UI/UX background with a current focus on system architecture and DevOps. Passionate about nature and leveraging technology to build scalable, environmentally focused solutions.

• Maksymilian Mazur — Water resources engineering. Systems engineering student with a strong mathematical foundation and broad, cross-disciplinary knowledge. Focused on translating rigorous analytical thinking into modern, practical solutions — with a particular interest in water and hydraulic systems.

• Kamil Fedio — Software engineering. Systems engineering student. Commercial experience as software engineer in delivering end-to-end solutions, focused on combining AI-driven solutions into real world applications. Participant of Amazon and Google courses about Cloud Computing.

• Grzegorz Wiącek — Data science. Systems engineering student with commercial experience as a data scientist, specializing in time series analysis and forecasting. Strong academic foundation and delivery of data-driven solutions across machine learning, data analysis and big data technologies.