OluAyo06/Satellite-Driven-PM2.5-Estimation-for-African-Cities

Air pollution is a major global health threat, with sub-Saharan Africa experiencing the most severe impacts due to limited monitoring systems and rapid urbanization. This project aims to close these data gaps by using satellite-derived information and machine learning to estimate PM2.5 pollution levels across eight key African cities --- # **Air Quality Prediction Using Satellite AOD and Machine Learning** This project aims to estimate **PM2.5 concentrations** across major African cities using **satellite-derived Aerosol Optical Depth (AOD)** and machine learning techniques. It supports ongoing efforts by **AirQo** and **Mozilla Foundation** to bridge air-quality data gaps in sub-Saharan Africa and empower communities with accurate environmental information. --- ## **Project Summary** Air pollution remains one of the world’s most serious environmental health risks, responsible for roughly **seven million premature deaths annually**. Sub-Saharan Africa is severely affected due to limited monitoring infrastructure and rising urban exposure. This project leverages: * **Satellite AOD measurements** * **Ground-based PM2.5 observations** * **Machine learning models** to estimate pollution levels in eight African cities: * **Lagos** * **Accra** * **Nairobi** * **Yaoundé** * **Bujumbura** * **Kisumu** * **Kampala** * **Gulu** The models developed here are intended to support deployment on the **AirQo digital platform**, enabling users to access real-time, hyperlocal air-quality information. --- ## **Project Structure** ``` 📂 Air-Quality-Prediction │ ├── 📓 Air Quality Prediction.ipynb ├── 📁 data/ │ ├── train.csv │ ├── test.csv │ └── additional datasets… │ ├── 📁 scripts/ │ └── preprocessing.py │ ├── README.md └── requirements.txt ``` --- ## **Notebook Workflow** ### **1. Cleaning of Dataset** * Handling missing values * Removing duplicates * Column formatting * Data validation ### **2. Exploration of Data** * Summary statistics * Variable distributions * Outlier detection * Time-series trends ### **3. Exploration of Test Dataset** * Checking structure and consistency * Matching columns with training dataset * Identifying missing or mismatched features ### **4. Encoding** * Converting categorical variables * Label and one-hot encoding * Preparing data for modeling ## …