Spatial and Meteorological Behaviour of Daily Ozone Air Pollution in the Bay of Algeciras (2010–2015)

Resumen

The Bay of Algeciras (Spain) is one of the most industrialized areas in Spain. Furthermore, the Port of Algeciras moved about 100 Millions of Tons in 2018. Therefore, this region could be one of the most affected territories by air pollution in Spain. An exhaustive statistical analysis of the different monitoring stations has been carried out in order to find out spatial and temporal trends along the region. Also, the relationship of the different air pollutants and meteorological variables has been calculated. On the one hand, a descriptive statistical analysis has been conducted and, on the other hand, Principal Component Analysis (PCA) was performed in each monitoring station in order to discover the most relevant features in each different location. We focused our analysis on monitoring stations located in Algeciras and La Línea, the two principal cities in the study area.One of the objectives has been to develop an estimation approach for any hypothetical damaged station using the other monitoring stations and the meteorological variables. Consequently, we have studied the leverage between all the variables and Ozone pollutant using Principal Component Analysis (PCA), together with Multiple Linear Regression (MLR) models and Artificial Neural Networks (ANNs) models to estimate air Ozone pollution in each monitoring station.The results show general trends and particular differences depending on the location of the monitoring stations and depending on several meteorological variables such as wind speed and wind direction that are in most cases the most relevant features to explain each pollutant concentration values. Examining the results of the proposed approach we can obtain robust estimations of each pollutant in each location as a function of the previously computed PCA variables and also the original ones. The regression analysis showed promising results (in Algeciras R = 0.857 and in La Linea R = 0,894) in order to have at our disposal a computational estimation tool in each different location. This kind of approach could be useful in the design of a robust sensoring network.