raoulbouri/Ground-level-NO2-Estimation

Zindi ML Contest: zindi.africa # Ground-level NO2 Estimation using Attention-based CNN-LSTM ## Project Overview This project addresses the **GeoAI Ground-level NO2 Estimation Challenge** from Zindi Africa, focusing on estimating ground-level nitrogen dioxide (NO2) concentrations using satellite data and machine learning techniques. NO2 is a critical air pollutant that affects human health and contributes to air quality degradation. **Competition Link:** Zindi GeoAI Ground-level NO2 Estimation Challenge ## Problem Statement Ground-level NO2 monitoring is essential for: - Air quality assessment and public health protection - Environmental policy making and regulation - Understanding urban air pollution patterns - Climate change impact assessment Traditional ground-based monitoring stations are expensive and sparse, making satellite-based estimation crucial for comprehensive coverage. ## Methodology ### Data Sources The project utilizes multiple satellite datasets: - **LST (Land Surface Temperature)**: Thermal infrared data for surface temperature - **AAI (Aerosol Absorption Index)**: Aerosol optical depth measurements - **Cloud Fraction**: Cloud coverage information - **Precipitation**: Rainfall data - **NO2 Stratospheric**: Stratospheric NO2 concentrations - **NO2 Total**: Total atmospheric NO2 - **NO2 Tropospheric**: Ground-level NO2 concentrations - **Tropopause Pressure**: Atmospheric pressure at tropopause level ### Data Preprocessing & Cleaning #### Null Value Handling - **KNN Imputation**: Missing values are filled using K-Nearest Neighbors algorithm - **Temporal Padding**: Incomplete time sequences are padded with synthetic data - **Feature Filtering**: Outliers and invalid measurements are removed - **Coordinate Validation**: Latitude/longitude coordinates are verified and standardized #### Temporal Sequence Processing - **15-day Lookback Window**: Each prediction uses the previous 15 days of data - **Location Grouping**: Data is organized by geographical coordinates (LAT, LON) …