Climate Change Impact Simulation (Melbourne, Australia)

This project analyzes real-time and historical weather data to simulate and visualize the impact of climate change in Melbourne.
Using live data from the OpenWeatherMap API, it explores temperature variations, rainfall trends, and urban heat zones — applying machine learning (DBSCAN & KMeans) and forecasting (Prophet) to predict climate patterns and identify risk-prone regions.

The goal is to help understand how +1.5°C and +2.0°C global warming scenarios could affect city-level climate dynamics, guiding sustainability and resilience planning. For the original notebook and report PDF, see docs:

PDF - https://drive.google.com/file/d/1rocu_2cXfBJ0hv2FvzIJS3kDmnJoTQT4/view?usp=sharing

PYTHON NOTEBOOK - https://drive.google.com/file/d/1s3Y-wWnGVfmtHFROr_9_GmzHpjcXQMrL/view?usp=sharing

---

Project Overview

• Collected and cleaned live weather data from OpenWeatherMap API (JSON-based).
• Built a dataset combining temperature, humidity, rainfall, wind, and pressure metrics.
• Applied DBSCAN clustering to detect heatwave-prone and cool zones across Melbourne.
• Forecasted temperature and rainfall using Facebook Prophet (time-series forecasting).
• Simulated warming scenarios (+1.5°C and +2.0°C) aligned with IPCC climate targets.
• Created interactive visualizations and heatmaps for spatial and temporal insights.

---

Key Findings

• Urban Heat Zones: Detected clusters of elevated temperature, primarily in dense suburbs.
• Tree Coverage Impact: Areas with higher vegetation (urban forests) showed lower heat intensity.
• Rainfall Forecast: The Prophet model predicts variable rainfall trends with potential decline post-2030.
• Temperature Trend: Notable +1.8°C increase simulated across Melbourne by 2050 under high-emission scenarios.
• Heat Index: "Feels-like" temperatures rise significantly beyond measured values, highlighting humidity’s role.

This project showed how data, weather APIs, and analytics can reveal crucial insights about our planet’s future.

---

Tools & Libraries

• Python: pandas, numpy, matplotlib, seaborn
• Geospatial: geopandas, folium, shapely
• Clustering: scikit-learn (DBSCAN, KMeans)
• Forecasting: Prophet
• Data Source: OpenWeatherMap API
• Environment: Google Colab / Jupyter Notebook

---

How to Run

1. Clone this repository:

   git clone https://github.com/rohang-7/Climate-Change-Impact-Simulation.git
   cd Climate-Change-Impact-Simulation
   
   
   

Why this repo

Urban heat and rainfall volatility are rising issues for councils and planners. This project ingests live weather from OpenWeatherMap, builds time-series forecasts (Prophet), detects urban heat clusters (DBSCAN/K‑Means), and explores warming scenarios (+1.5 °C / +2 °C) to highlight risk hot-spots and potential adaptation levers (trees, cool roofs, heat-health planning).

---

Project structure

.
├── README.md
├── requirements.txt
├── .env.example
├── quickstart.py                 # runs on sample data end-to-end (no API key needed)
├── src/
│   ├── fetch_weather.py          # live API ingestion (OpenWeatherMap)
│   ├── clean.py                  # cleaning & feature engineering
│   ├── forecasting.py            # Prophet utilities
│   └── cluster.py                # DBSCAN + K-Means helpers
├── data/
│   └── sample_weather.csv        # tiny cached sample for reproducibility
├── figures/                      # output charts (saved here by quickstart.py)
│   └── .gitkeep
└── notebooks/
    └── (optional notebooks)

---

Quick start (no API key needed)

1. Install dependencies

   pip install -r requirements.txt

2. Run the end-to-end sample

   python quickstart.py

3. See outputs

   • Figures saved to ./figures/:
     • fig_prophet_temp.png (if Prophet available)
     • fig_kmeans_clusters.png
     • fig_dbscan_clusters.png

The sample uses a tiny CSV (data/sample_weather.csv) so anyone can reproduce plots without credentials.

---

Live data (optional)

To pull fresh observations:

1. Get a free API key from OpenWeatherMap.
2. Create a .env file using the template below and add your key:

   cp .env.example .env

3. In your code, call src/fetch_weather.py:fetch_openweather() to retrieve data for a city or coordinates.

.env.example

OWM_API_KEY=YOUR_KEY_HERE
OWM_BASE_URL=https://api.openweathermap.org/data/2.5/weather
DEFAULT_CITY=Melbourne,AU

---

Methods in 60 seconds

• Forecasting (Prophet): daily/Hourly temperature y ~ trend + seasonality; 95% CI; configurable horizon.
• K‑Means: k chosen via elbow/silhouette (default k=3); features include temp, humidity, “feels_like”, and (optionally) lat/lon.
• DBSCAN: density-based clusters on standardized spatial features (default eps=0.12, min_samples=10 after scaling).
• Scenarios: baseline vs +1.5 °C and +2 °C (applied additively to forecast mean) to examine relative risk shifts.

CRS: inputs are assumed in WGS84 (EPSG:4326) when mapping geolocated points.

---

Reproduce the figures

# Prophet forecast (skips gracefully if Prophet not installed)
python quickstart.py --forecast

# Spatial clustering (K-Means + DBSCAN)
python quickstart.py --cluster

# All steps
python quickstart.py --forecast --cluster

---

Results (example highlights — replace with your latest):

• Urban heat zones coincide with lower tree/vegetation coverage and denser built form.
• “Feels-like” temperature escalates faster than dry-bulb temperature on humid days.
• 2050 warming scenario (high-emission sim) shows ~+1.8 °C mean temperature increase; rainfall variance widens.

---

References

• IPCC Assessment Reports (AR6), WG1 Climate Change 2021 (for warming context)
• OpenWeatherMap API docs (data dictionary & endpoints)
• Ester et al. (1996) “A Density-Based Algorithm for Discovering Clusters in Large Spatial Databases with Noise” (DBSCAN)

---

License

MIT

---

Streamlit app

Run an interactive demo locally:

streamlit run streamlit_app.py

Controls (left sidebar):

• K-Means: choose k
• DBSCAN: tune eps and min_samples
• Forecast: toggle Prophet forecast (skips gracefully if Prophet not installed)

---