Application of aggregation operators for forecasting [formula omitted] fluctuations: From available Caribbean data sites to unequipped ones

Abstract

Air pollution is a substantial issue for public health. Predicting the levels of airborne particles, especially those originating from natural phenomena like sand mists from the African dust belt, enables more effective warnings for affected populations. Currently, there are only three islands that continuously measure PM10 concentrations in the Caribbean area. Thus, there is a main issue for the forecast of these particulate pollutants for unequipped sites. In this article, a new approach based on the use of aggregation operators in a non-fuzzy framework allows to predict the quantity of PM10 for certain sites in this geographical area. It is the characteristics of non-linearity and downward reinforcement of the fuzzy operators used (triple Pi and t-norms) which make it possible to model the evolution of a phenomenon as complex as the sand mists from African dust belt. The time series having been normalized, the tests were carried out on two of the three sites (Martinique and Guadeloupe) to predict the values of the third site (Puerto Rico) and assess the quality of the prediction. The mean absolute error (MAE), the mean bias error (MBE), the root mean square error (RMSE) and the index of agreement (IOA) values show that this approach provides significant results. The African dust seasonality strongly impacts these performances metrics. These initial promising findings emphasize that the utilization of aggregation operators is a robust method for forecasting PM10 fluctuations, eliminating the need for complex parameterization or learning phase that would otherwise entail significant computational costs.