Combination of GEOS-CF model with Machine Learning as a tool for forecasting regional pollution in Bogotá

Through joint work between the Bogotá Environment Secretariat, NASA-GMAO and WRI, Bogotá became a participant in the CityAQ project, which seeks to combine locally monitored information with the GEOS-CF model through Machine Learning for various pilot cities, including Bogotá. Likewise, this work evaluated the performance of GEOSCF + Machine Learning in a period of influence of foreign pollution in the city (February 2021). 2018-2019 monitoring data and GEOS-CF results from the same period were used to implement the XGboost Machine Learning algorithm to correct the bias between GEOS-CF results and the observations in Bogotá as a function of meteorological variables and the behavior of PM2.5, NO2 and O3 in the atmosphere. With the results obtained for Bogotá, the performance of GEOS-CF+ML was evaluated, regarding results of GEOS-CF without bias correction and other forecasting air quality models used in Bogotá. Machine Learning improve performance of GEOS-CF in Bogotá, especially in perimeter stations and of high spatial representativeness stations, and PM2.5 was de pollutant with the best performance indicators in a period with high influence of regional pollution.