🌊 Cottbuser Ostsee — Water Expansion & Biodiversity Monitoring (2019–2026)

Sentinel-2 · Planetary Computer · NDWI · NDVI · LSTM Forecasting · Anomaly Detection

---

📍 What is Cottbuser Ostsee?

Cottbuser Ostsee is one of Europe's largest post-mining lake restoration projects.
A former open-cast lignite coal mine in Brandenburg, Germany is being flooded to create a ~1,900 hectare recreational lake — one of the largest artificial lakes in Central Europe.

This project uses satellite remote sensing and deep learning to monitor and forecast the lake's growth from space — without ever visiting the site.

---

📊 Key Finding

2019:  ~550 hectares of water surface
2026:  ~1,800 hectares of water surface

Growth: 3.3× in 7 years
Trend:  +180 ha/year (satellite-derived)

---

🛰️ What This Notebook Does

Part 1 — Satellite Remote Sensing

Track water expansion, vegetation, and biodiversity using Sentinel-2 imagery.

Step	Method	Output
Data access	Microsoft Planetary Computer STAC	Monthly best-scene selection
Cloud masking	SCL Scene Classification Layer	Clean imagery
Water detection	NDWI = (Green − NIR) / (Green + NIR)	Water area in hectares
Vegetation	NDVI = (NIR − Red) / (NIR + Red)	Vegetation area in hectares
Wetland proxy	NDMI = (NIR − SWIR) / (NIR + SWIR)	Reed/wetland extent
Biodiversity	Shannon diversity index + k-means	Habitat heterogeneity map
Visualization	Animated GIFs	Water + NDVI time-lapse

Part 2 — Deep Learning (LSTM)

Forecast future water expansion and detect ecological anomalies.

Step	Method	Output
Forecasting	2-layer LSTM with seasonal encoding	24-month water area forecast
Anomaly detection	LSTM residual analysis (±1.5σ)	Flagged ecological events

---

🏗️ LSTM Architecture

Input: (batch, 6 months, 3 features)
       [water_area_norm, month_sin, month_cos]
              ↓
       LSTM Layer 1 (hidden=64)
       learns seasonal fluctuations
              ↓
       LSTM Layer 2 (hidden=64)
       learns long-term filling trend
              ↓
       Linear(64 → 32) → ReLU → Linear(32 → 1)
              ↓
Output: next month water area (hectares)

Why seasonal sin/cos encoding?
Water area fluctuates within each year (evaporation in summer, rainfall in autumn). Encoding month as sin/cos gives the LSTM a built-in calendar — it knows January follows December and that summer behaves differently from winter.

---

🤖 Anomaly Detection

The LSTM learns the expected trajectory of lake filling.
When a month deviates significantly from expectation, it is flagged.

Residual = Observed area − LSTM predicted area

Positive residual → more water than expected (unusual flooding / wet year)
Negative residual → less water than expected (drought / management change)
Flagged if |residual| > 1.5 standard deviations

Flagged months warrant investigation — they may reflect drought years, unusual rainfall, changes in mine water management, or ecosystem regime shifts as the lake matures and ecology develops.

---

📦 Data Source

Sentinel-2 L2A via Microsoft Planetary Computer

Bands used:
  B03 — Green  (10m) — NDWI water detection
  B04 — Red    (10m) — NDVI vegetation
  B08 — NIR    (10m) — NDWI, NDVI, NDMI
  B11 — SWIR   (20m) — NDMI wetland proxy
  SCL — Scene Classification (20m) — cloud masking

Resolution:    10m × 10m per pixel = 0.01 ha per pixel
Cloud filter:  < 30% cloud cover
Selection:     Best scene per month (lowest cloud cover)
Period:        January 2019 – present

No data download required — Planetary Computer provides free cloud access to the full Sentinel-2 archive directly from the notebook.

---

🚀 Quick Start

Run on Kaggle (recommended — free GPU)

Open the notebook on Kaggle and run all cells top to bottom.
No local setup required. All data loads from Planetary Computer automatically.

Run locally

pip install pystac-client planetary-computer stackstac rioxarray \
            imageio tqdm torch scikit-learn geopandas

Then run Cottbuser_Ostsee_AI_Monitoring.ipynb from top to bottom.

---

📁 Output Files

File	Description
water_area_timeseries.csv	Per-scene, monthly, yearly water areas (ha)
water_timeseries.png	Time series + linear trend plot
habitat_diversity.png	Shannon diversity map + vegetation proxy
water_expansion_animation.gif	Lake growth time-lapse (NDWI)
biodiversity_proxy_animation.gif	Vegetation dynamics time-lapse (NDVI)
lstm_training_history.png	LSTM loss curves
lstm_forecast.png	24-month water expansion forecast
anomaly_detection.png	Residual anomaly plot

---

🌍 Why This Matters

Ecological significance:
As the lake fills, entirely new aquatic and wetland ecosystems develop. Monitoring water extent, vegetation establishment, and habitat diversity provides early evidence of ecosystem formation — relevant for biodiversity assessment under the EU Habitats Directive.

Engineering relevance:
Cottbuser Ostsee is a complex hydro-engineering project managed by LMBV (Lausitzer und Mitteldeutsche Bergbau-Verwaltungsgesellschaft). Satellite-derived water area measurements provide independent verification of filling progress without requiring site access.

Climate relevance:
Large new water bodies affect local microclimate. Tracking the lake's growth from space contributes to understanding land-use change impacts in post-industrial landscapes — a growing challenge across Europe as coal mining regions transition to renewable energy economies.

---

📋 Validation (Thesis Recommendation)

For thesis-level rigor, satellite-derived areas should be validated against official LMBV project reports or Brandenburg Geoportal reference data.

Year	Satellite Area (ha)	Official Area (ha)	Error (%)
2020	—	—	—
2022	—	—	—
2024	—	—	—

---

📚 References

Sentinel-2 Mission:
ESA (2015) — Sentinel-2: ESA's Optical High-Resolution Mission
for GMES Operational Services. ESA Special Publication.

NDWI:
McFeeters, S.K. (1996) — The use of the Normalized Difference
Water Index (NDWI) in the delineation of open water features.
International Journal of Remote Sensing, 17(7), 1425–1432.

Planetary Computer:
Microsoft (2021) — Microsoft Planetary Computer.
https://planetarycomputer.microsoft.com

Cottbuser Ostsee project:
LMBV — Lausitzer und Mitteldeutsche Bergbau-Verwaltungsgesellschaft
https://www.lmbv.de

---

👤 Author

Zenith Patel
MSc Environmental and Resource Management
Brandenburg University of Technology (BTU) Cottbus-Senftenberg, Germany

This project monitors a restoration site located in the same region as BTU Cottbus — combining local environmental knowledge with satellite remote sensing and deep learning.

---

📄 License

MIT License — free to use, modify, and distribute with attribution.

---

Part of a portfolio in Environmental AI — applying deep learning to satellite remote sensing for biodiversity monitoring, ecosystem restoration, and climate impact assessment.