🗺️ Land-Cover Composite Map — Pamfila, Lesvos

Cartographic composite of a UAV land-cover classification, integrating elevation (nDSM), the 7-class semantic-segmentation result, and a thermal channel into a single shaded-relief poster, with an interactive web companion that documents the cartographic workflow.

🌐 Live page · 🗺️ Map design · 🛰️ Data--method · 👤 Author · 📚 Citation

---

Overview

This repository hosts an interactive web page (GitHub Pages) that explains how a cartographic land-cover composite was produced over Pamfila, Lesvos (Greece), together with the Python rendering pipeline that generates the final poster. The work is part of an MSc thesis at the University of the Aegean (Department of Geography, RSGIS Lab), supervised by Dr. Christos Vasilakos.

Source imagery comes from a MicaSense Altum-PT sensor. A DeepLabV3 + PointRend (ResNet-101) model classifies the scene into 7 land-cover classes — Tree, Building, Road, Vehicle, Grass, Bare Soil, Shadow-Noise — and the map integrates that classification with nDSM (height) and a thermal channel.

Language note: the web page is written in Greek, with standard English terminology kept (DeepLabV3, PointRend, ResNet-101, nDSM, hillshade).

🌐 Live page

https://nickkoro21.github.io/lesvos-altum-segmentation/

The page (docs/) includes:

Section	Content
Overview	Final poster with full-screen view, zoom and PNG / PDF download.
Input data	Sensor, classes and per-class coverage.
3+1 Views	nDSM, classification, thermal + RGB with synchronized zoom / pan (shared extent).
Diagonal composition	Adjustable per-zone footprint (nDSM top-left, thermal bottom-right, classification remainder; sum = 100%).
Tools · Ratios & colours · Path v1→v12	Implementation stack, layout geometry, palette, design history.

Every image supports full-screen, zoom in/out, keyboard-arrow panning, reset and exit.

Map design

• Composite of three diagonal zones with feathered transitions: nDSM (top-left), classification (centre), thermal (bottom-right).
• Shaded-relief synthesis: each layer is modulated by a hillshade (bidirectional multiply) to retain both colour and texture.
• nDSM ramp: neutral slate → tan (low elevation intentionally non-vegetation-green).
• Thermal: inferno colour map (~20-60 °C) with a light hillshade overlay.
• Output: A-ratio-agnostic custom layout (≈ 1.71), rendered at 300 dpi (7800 × 4560 px).

🛰️ Data & method

• Sensor: MicaSense Altum-PT (multispectral + thermal + panchromatic).
• Model: DeepLabV3 + PointRend, ResNet-101 backbone, 7-Band input, 7 land-cover classes.
• Reference system: EPSG:32635 (UTM 35N) · GSD 4.52 cm · area ≈ 500 × 350 m.

Repository structure

lesvos-altum-segmentation/
├── docs/                      ← GitHub Pages site
│   ├── index.html             ← interactive page
│   ├── assets_html/           ← co-registered layer images (nDSM, classified, thermal, RGB)
│   └── downloads/             ← hero_map.png (300 dpi) + hero_map.pdf
├── src/
│   ├── render_hero_panels.py  ← STEP 1: resample sources to a common extent, export panels
│   └── compose_hero.py        ← STEP 2: compose the final poster (matplotlib)
├── WORKFLOW.md                ← full design notes / change history
├── README.md
└── LICENSE

Reproducing the poster

The scripts require ArcGIS Pro Python (arcgispro-py3, with arcpy) and the original raster sources (not published). They contain local absolute paths and are provided as a methodological reference rather than a stand-alone executable.

# STEP 1 (only if the canvas extent changes)
"%ProgramFiles%\ArcGIS\Pro\bin\Python\envs\arcgispro-py3\python.exe" src\render_hero_panels.py
# STEP 2 — render the final poster (Hero_Layout.png / .pdf, 300 dpi)
"%ProgramFiles%\ArcGIS\Pro\bin\Python\envs\arcgispro-py3\python.exe" src\compose_hero.py

👤 Author & affiliations

Nikolaos Koroniadis — MSc Geography and Applied Geoinformatics, Department of Geography, University of the Aegean, Remote Sensing & GIS Research Group (RSGIS Lab).

Thesis supervisor: Dr. Christos Vasilakos.

🔗 Related projects

• 7-Band vs RGB Results Dashboard
• PostProcessing Toolbox
• JM Separability Toolbox
• Spectral Separability Explorer (Hugging Face Space)
• 3D Spectral Feature Space Explorer (Hugging Face Space)

📚 Citation

@software{koroniadis2026landcovermap,
  author    = {Koroniadis, Nikolaos},
  title     = {{Land-Cover Composite Map of Pamfila, Lesvos: Cartographic
               Synthesis of UAV Multispectral, nDSM and Thermal Data}},
  year      = {2026},
  publisher = {GitHub},
  url       = {https://github.com/Nickkoro21/lesvos-altum-segmentation},
  note      = {MSc Thesis deliverable, University of the Aegean, RSGIS Lab}
}

🙏 Acknowledgments

• Dr. Christos Vasilakos — thesis supervision.
• University of the Aegean, RSGIS Lab — academic environment and resources.
• Anthropic Claude — AI-assisted development of the rendering pipeline and web page.
• GitHub — free hosting via GitHub Pages.

📄 License

Released under the MIT License.

---

RSGIS Lab · Department of Geography · University of the Aegean · Mytilene, Greece · 2026