AaronOnosala/AirQo-Africa-Air-Quality-Prediction-Challenge
Domain:
environment and energygeospatialRecord type:
project
Creator:
Aar
Host:
# AirQo Africa, Air Quality And Pollution Estimation Project
collaboration: AirQo Africa, Mozilla Foundation, Makerere University Artificial Intellegence Lab
## Background
Air pollution is the world's largest environmental health risk, leading to 7 million premature deaths globally each year, with the most severe impacts felt by poor communities in developing countries. Accurate air quality data is crucial for developing effective interventions to mitigate these risks. In Sub-Saharan Africa, low-cost IoT sensor networks have potential but are expensive to implement and maintain. Satellite-derived PM2.5 estimates based on Aerosol Optical Depth (AOD) and machine learning techniques offer a scalable solution to estimate pollution levels over large areas. This approach can support the development of vulnerability risk profiles, health preservation strategies, environmental protection measures, community empowerment, and climate change mitigation.
## Problem Statement
The project aimed to estimate PM2.5 levels from satellite observations based on AOD for eight cities in seven African countries: Lagos, Accra, Nairobi, Yaoundé, Bujumbura, Kisumu, Kampala, and Gulu. The challenge was to develop a machine learning model that could accurately estimate PM2.5 levels, which would then be validated using ground-based observations.
## Approach
To tackle this challenge, we utilized several machine learning models, ultimately finding that the TensorFlow Decision Forest model was the most suitable for this specific task. The models were trained using a dataset of satellite-derived AOD and corresponding PM2.5 ground observations from the selected cities.
## AI Methods Applied
## AI Framework Used In This Project
1. __Data Collection and Preprocessing:__
Collected satellite-derived AOD data and corresponding PM2.5 ground observations for the eight cities. Cleaned and preprocessed the data to ensure consistency and accuracy.
2. __Model Selection:__
Evaluated multiple mach …