Estimating Ground-level Nitrogen Dioxide Concentration from Satellite Data

At present times monitoring of air quality is regarded to be of utmost importance. In that regard Nitrogen Dioxide concentration is an important factor to consider. Satellite measurements provide an excellent opportunity of global, round the clock measurements as opposed to in situ monitoring. However, given the fact that satellite measured Tropospheric Vertical Column Density does not offer much information about near-surface concentration, estimation methods become utterly necessary. In this paper we propose a novel Convolutional Neural Network based deep learning method to perform an estimation of near-surface NO2 concentration over Ireland using Tropospheric VCD as the sole input feature. The proposed method is validated against in situ data. The method shows a Root Mean Squared Error of 7.20 mu g/m(3). The work is novel in two aspects. Firstly, it treats the problem as an univariate estimation only dependent on satellite measurement. Therefore, independence from in situ data allows it to be used for estimation even in remote locations where no ground data is available. Secondly, in our knowledge Convolutional Neural Networks have not yet been explored as a possible method for estimation of near-surface NO2 concentration and our work is the first that does so with good results.