🗺️ Predicting CO₂ Emissions:

Accurately tracking carbon emissions is a vital foundation for addressing climate change.

This project aims to build machine learning models that use open-source CO₂ emission data collected by the Sentinel-5P satellite to forecast future carbon output.

📚 used Python libraries:

• pandas
  
• matplotlib
  
• scikit-learn
  
• scipy
  
• streamlit
  
• plotly
  
• PyTorch
  
• Keras
  
• TensorFlow
  
• keras_tuner
  
• optuna
  

✨ features of original dataset:

-basic information:

• ID_LAT_LON_YEAR_WEEK
  
• latitude
  
• longitude
  
• year
  
• week_no
  

-SulphurDioxide:

• SulphurDioxide_SO2_column_number_density
  
• SulphurDioxide_SO2_column_number_density_amf
  
• SulphurDioxide_SO2_slant_column_number_density
  
• SulphurDioxide_cloud_fraction
  
• SulphurDioxide_sensor_azimuth_angle
  
• SulphurDioxide_sensor_zenith_angle
  
• SulphurDioxide_solar_azimuth_angle
  
• SulphurDioxide_solar_zenith_angle
  
• SulphurDioxide_SO2_column_number_density_15km
  

-CarbonMonoxide:

• CarbonMonoxide_CO_column_number_density
  
• CarbonMonoxide_H2O_column_number_density
  
• CarbonMonoxide_cloud_height
  
• CarbonMonoxide_sensor_altitude
  
• CarbonMonoxide_sensor_azimuth_angle
  
• CarbonMonoxide_sensor_zenith_angle
  
• CarbonMonoxide_solar_azimuth_angle
  
• CarbonMonoxide_solar_zenith_angle
  

-NitrogenDioxide:

• NitrogenDioxide_NO2_column_number_density
  
• NitrogenDioxide_tropospheric_NO2_column_number_density
  
• NitrogenDioxide_stratospheric_NO2_column_number_density
  
• NitrogenDioxide_NO2_slant_column_number_density
  
• NitrogenDioxide_tropopause_pressure
  
• NitrogenDioxide_absorbing_aerosol_index
  
• NitrogenDioxide_cloud_fraction
  
• NitrogenDioxide_sensor_altitude
  
• NitrogenDioxide_sensor_azimuth_angle
  
• NitrogenDioxide_sensor_zenith_angle
  
• NitrogenDioxide_solar_azimuth_angle
  
• NitrogenDioxide_solar_zenith_angle
  

-Formaldehyde:

• Formaldehyde_tropospheric_HCHO_column_number_density
  
• Formaldehyde_tropospheric_HCHO_column_number_density_amf
  
• Formaldehyde_HCHO_slant_column_number_density
  
• Formaldehyde_cloud_fraction
  
• Formaldehyde_solar_zenith_angle
  
• Formaldehyde_solar_azimuth_angle
  
• Formaldehyde_sensor_zenith_angle
  
• Formaldehyde_sensor_azimuth_angle
  

-UvAerosolIndex:

• UvAerosolIndex_absorbing_aerosol_index
  
• UvAerosolIndex_sensor_altitude
  
• UvAerosolIndex_sensor_azimuth_angle
  
• UvAerosolIndex_sensor_zenith_angle
  
• UvAerosolIndex_solar_azimuth_angle
  
• UvAerosolIndex_solar_zenith_angle
  

-Ozone:

• Ozone_O3_column_number_density
  
• Ozone_O3_column_number_density_amf
  
• Ozone_O3_slant_column_number_density
  
• Ozone_O3_effective_temperature
  
• Ozone_cloud_fraction
  
• Ozone_sensor_azimuth_angle
  
• Ozone_sensor_zenith_angle
  
• Ozone_solar_azimuth_angle
  
• Ozone_solar_zenith_angle
  

-UvAerosolLayerHeight:

• UvAerosolLayerHeight_aerosol_height
  
• UvAerosolLayerHeight_aerosol_pressure
  
• UvAerosolLayerHeight_aerosol_optical_depth
  
• UvAerosolLayerHeight_sensor_zenith_angle
  
• UvAerosolLayerHeight_sensor_azimuth_angle
  
• UvAerosolLayerHeight_solar_azimuth_angle
  
• UvAerosolLayerHeight_solar_zenith_angle
  

-Cloud:

• Cloud_cloud_fraction
  
• Cloud_cloud_top_pressure
  
• Cloud_cloud_top_height
  
• Cloud_cloud_base_pressure
  
• Cloud_cloud_base_height
  
• Cloud_cloud_optical_depth
  
• Cloud_surface_albedo
  
• Cloud_sensor_azimuth_angle
  
• Cloud_sensor_zenith_angle
  
• Cloud_solar_azimuth_angle
  
• Cloud_solar_zenith_angle
  

-Target:

• emission

🌱 Understanding the dataset:

-Sulphur Dioxide(SO₂):
SO₂ mainly comes from fossil fuel combustion (coal, oil), volcanic eruptions, and industrial processes like metal smelting.
It contributes to acid rain and respiratory problems.

• Main sources:
  
  • Coal- and oil-fired power plants
    
  • Metal smelters and refineries
    
  • Volcanic emissions
    
  • Ship exhaust (bunker fuel)
    

-Nitrogen Dioxide (NO₂):
NO₂ is primarily produced from high-temperature combustion, such as in vehicle engines, power plants, and industrial boilers.
It contributes to urban smog and reacts with sunlight to form ground-level ozone.

• Main sources:
  
  • Vehicle exhaust
    
  • Power generation (especially fossil fuels)
    
  • Industrial boilers and furnaces
    
  • Aircraft engines
    

-Carbon Monoxide (CO):
CO is produced by incomplete combustion — when carbon-based fuels burn with insufficient oxygen.
It’s a colorless, odorless toxic gas mainly from traffic and fires.

• Main sources:
  
  • Vehicle exhaust (especially older engines)
    
  • Domestic heating and cooking (stoves, fireplaces)
    
  • Biomass burning and wildfires
    
  • Industrial combustion
    

-Formaldehyde (HCHO):
Formaldehyde (HCHO) is both directly emitted and formed through photochemical oxidation of volatile organic compounds (VOCs).
It’s a key intermediate in ozone and smog formation.

• Main sources:
  
  • Vehicle and industrial VOC emissions
    
  • Biomass burning and forest fires
    
  • Chemical manufacturing (resins, paints, glues)
    
  • Atmospheric oxidation of VOCs
    

-Ozone (O₃):
There are two types of ozone:
Stratospheric ozone (“good ozone”) protects from UV radiation.
Tropospheric ozone (“bad ozone”) is formed when NO₂ and VOCs react under sunlight, causing smog and respiratory problems.

• Main sources:
  
  • Secondary formation from NO₂ and VOCs
    
  • Sunlight-driven reactions in polluted cities
    

-Aerosols & UvAerosolIndex:
Aerosols are tiny particles suspended in the air, coming from both natural and anthropogenic sources.
They affect climate (by scattering/absorbing sunlight) and air quality.

• Main sources:
  
  • Dust storms and desert emissions
    
  • Volcanic ash
    
  • Biomass burning and wildfires
    
  • Industrial and vehicle emissions
    

-Cloud:
Cloud data are not pollutants but influence satellite observations through radiative transfer (reflection, absorption).

✏️ methods:

1. Data Collection
   

2. Data Exploration/Processing
   -> missing values : filled with interpolate(linear)
   -> outlier detection : IsolationForest
   -> data normalization
   -> handling skewed distribution
   

3. Data Visualization(streamlit/plotly)
   

• Boxplot
  
• Correlation
  
• Feature importance
  
• mutual information
  
• interactive
  

4. Feature Engineering
   

• Feature Creation
  
• Feature Elimination
  

5. Modeling
   

• neural network
  
• k fold cross validation
  
• optimizer
  
• early stopping
  
• hyperparameter optimization
  

in progress

🎉 run the programm:

python -m streamlit run app.py