Resty-l/uganda-air-quality-prediction

Predicting ground-level PM2.5 air quality in Uganda by fusing Sentinel-5P satellite data with ground sensor measurements using a hybrid CNN-LSTM deep learning architecture. # Air Quality Estimation in Uganda ### A CNN-LSTM Framework Integrating Satellite-Derived Aerosol Indices and Ground-Level PM2.5 Measurements ## Abstract This capstone project investigates the efficacy of using remote sensing data as a proxy for ground-level atmospheric monitoring in resource-constrained environments. By fusing Sentinel-5P satellite observations with localized sensor data from the AirQo network, this study develops a hybrid CNN-LSTM architecture to estimate PM2.5 concentrations across the Ugandan landscape, specifically addressing the infrastructure gap in traditional monitoring networks. ## Technical Deployment **Web-based Inference Dashboard:** uganda-air-quality-predicti… ## Problem Statement Air quality monitoring in Sub-Saharan Africa is hindered by the high capital expenditure required for stationary monitoring stations. In Uganda, urban centers like Kampala experience significant particulate matter fluctuations, yet rural and peri-urban areas remain unmonitored. This research proposes a scalable, cost-effective alternative by calibrating satellite-derived Aerosol Index (AI) data against localized ground truth measurements. ## Methodology & Architecture The predictive engine utilizes a dual-stage deep learning approach: * **Feature Extraction (CNN):** A 1D-Convolutional layer processes the multispectral satellite signals to identify high-frequency patterns in aerosol density. * **Sequence Modeling (LSTM):** Long Short-Term Memory units capture the temporal dependencies and diurnal cycles of pollution, utilizing a 24-hour look-back window. * **Normalization & Optimization:** Data was pre-processed using Min-Max scaling and optimized via a Huber Loss function to maintain robustness against sensor noise and outliers. ## Data Inventory & Fusion The research utilizes two primary data streams: * **In-Situ Measurements (Ground Truth):** Longitudinal PM2.5 data provided by Makerere University’s AirQo project. * …