Characterizing Atmospheric and Greenhouse Gas Regimes in Megacities of the Americas: A Non-Supervised Approach Combining PCA and K-Means Clustering

Megacities are highly vulnerable environments facing unprecedented challenges from climate change and rapid urbanization, resulting in localized environmental degradation and air pollution. Understanding the complex interactions between multi-sensor atmospheric variables and greenhouse gases (GHGs) is crucial for sustainable urban planning. This study employs an advanced, data-driven methodology that uses unsupervised machine learning techniques to identify recurrent climate regimes across nine megacities on the American continent. A high-resolution dataset spanning five years (2020–2024) of daily records was compiled, integrating eight core parameters from global reanalyses (ERA5 and CAMS-EAC4): 2-meter air temperature (TEMP2), total precipitation (TP), dew point temperature (DW), net surface solar radiation (SR), 10m u-component of wind (EW), 10m v-component of wind (NW), methane (CH₄), and tropospheric ozone (O₃). Principal Component Analysis (PCA) was applied to reduce dimensionality and capture the primary environmental processes, while the K-means algorithm was used for spatiotemporal grouping, validated using the Elbow and Silhouette metrics. The first three principal components explained between 60% and 70% of the total variance, isolating a thermohygrometric gradient (PC1) and an emission–precipitation interaction baseline (PC2). This architecture revealed three distinct urban climate regimes: humid, cold–dry, and warm. Spatial analysis highlighted convergent behaviors in cities such as Lima, Rio de Janeiro, and São Paulo under the warm regime, in contrast to the high temporal variability observed in Mexico City, New York, Los Angeles, and Buenos Aires. These findings offer actionable, spatiotemporal insights into urban environmental processes, providing a robust scientific framework for developing resilient and smart-city strategies.