Digital Twin — Land Use Forecasting

A digital twin for forecasting and visualising land use change around Norwich, UK. The pipeline turns satellite and GIS rasters (RGB, NDVI, temperature, elevation, population, water distance, and land cover labels) into decadal land use forecasts up until 2124, shown in an interactive React web app.

Author: Michael Shehata Supervisor: Dr. Mohsin Raza

---

Prerequisites

• Python 3.11+
• Node.js 20+ (for the frontend)

---

Setup

From the project root:

python -m venv .venv

# Windows
.venv\Scripts\activate

pip install -r requirements.txt

data/ 
  elevation/elevation.tif
  rgb_landsat/rgb_{1985,1995,2005,2015,2024}.tif
  ndvi/ndvi_{1985,1995,2005,2015,2024}.tif
  temperature/temperature_{1985,1995,2005,2015,2024}.tif
  population/population_{2000,2010,2020}.tif
  water/distance_to_water.tif
  lcm/lcm1990_25m_gb.tif, lcm2007_25m_gb.tif, lcm2015_25m_gb.tif, lcm2023_25m_gb.tif

---

Pipeline (preprocessing → prediction)

Run all commands from the project root with the virtual environment active.

1. Inspect and preprocess feature rasters

Reprojects, cleans, and normalises all .tif files under data/ to a common grid in processed_data/:

python -m scripts.inspect_raw_data
python -m scripts.preprocess_data
python -m scripts.inspect_preprocessed_data
 

2. Preprocess land cover labels

Builds aligned label rasters (labels_*.tif) in processed_data/ (requires processed_data/elevation.tif from step 1):

python -m scripts.preprocess_lcm_labels.py

3. Train models

Production model (Random Forest saved to best_model/ — used by prediction):

python -m model.best_model.py

Optional — compare Random Forest vs XGBoost (metrics in model_results/):

python -m model.train_models.py

4. Generate forecasts

Projects features forward and writes GeoTIFFs and PNGs to forecasts/:

python -m model.predict_le.py

Requires best_model/rf3_final_model.pkl and best_model/rf3_final_scaler.pkl from step 3.

---

Run the web app

Copy forecast images into the frontend public folder so the Visualization page can load them:

# Windows (PowerShell)
Copy-Item -Path forecasts\forecast_*.png -Destination frontend\public\forecasts\ -Force

Then start the dev server:

```bash
cd frontend
npm install
npm run dev

Open [http://localhost:5173]

Outputs

Step	Output
Preprocessing	processed_data/*.tif
Training	best_model/, model_results/
Prediction	forecasts/forecast_<year>.tif, .png, timeline and change maps