Machine Learning

Multivariate Analysis

The data observations were less when Cloud Fraction (CF) value exceeded 0.6. This shows the prediction can be affected by the cloud a lot.

The NO2 were observed to be higher below certain temperatures.

The NO2 were observed between specific range of 10m wind especially at low velocity but no correlations were observed.

Limitations

Most of the NO2 values for CF > 0.85 weren’t present so to account that data, the NO2 concentrations were assumed to be 0.

Only the limited amount of parameters were taken into account to observe the changes in NO2 concentrations.

Based on these observed parameters a simple neural network was built to determine the complexity which restricted the model performance.

Neural Network Approach

A simple neural network was trained on the parameters based on the observations from multivariate analysis.

Input Parameters: Ship Density, UV-b, wind velocity components, CF

Output Parameters: NO2 column densities

The data accumulation was done for the duration of Dec 2019 to Jun 2020 away from the coastal region to avoid the NO2 pollution from land.

Limitations

Model performance was not good due to less consideration of input, missing data values of NO2 and less consideration of optimization.