Urban monitoring application in new administrative capital in Egypt using machine learning techniques and Sentinel 2 data

Abstract

Recently, Egypt has been undergoing rapid development in the urban and transportation sectors with a focus on the New Administrative Capital. This requires temporal monitoring and assessment for such projects with low cost, high accuracy, and simple application. Therefore, sentinel-2 data was utilized to assess the effectiveness of several methods for urban classification and monitoring. The study discusses three main approaches: histogram threshold, spectral indices, and machine learning techniques. In machine learning, we focus on K-nearest neighbor (KNN), Linear discrimination analysis (LDA), and Random Forest (RF) due to their robustness and simplicity. It was found that the RF technique presents the highest accuracy for our study region with 98.8%. Moreover, this study monitored the effect of such development on air pollution in the New Administrative Capital using sentinel-5P data which showed an increase in the CO and NO2 levels during the last five years by nearly 11% and 75% respectively, with a noticeable drop during the year 2020 due to the Covid19 lockdown. The classification model was trained and verified using ground truth data obtained on 10th of Ramadan City. Then, it was applied to monitoring urban development and the associated environmental impact in the New Administrative Capital in Egypt.

Keywords

• random forest
• k-nearest neighbor
• linear discrimination analysis
• sentinel-5p