A Genetic Programming Approach Applied To Airborne Salinity Prediction

Atmospheric corrosion is the deterioration of metals caused by atmospheric conditions, with airborne salinity being one of the most significant corrosive agents. Measuring airborne salinity requires months or years of field experimentation, which limits its practical application and motivates the development of predictive models. While empirical equations based on distance from the sea are commonly used, they present significant limitations. Machine Learning approaches, such as Random Forests, can leverage remote geospatial and atmospheric data to improve predictions; however, these models lack interpretability as explicit equations. To address this challenge, symbolic regression offers a promising alternative by generating interpretable mathematical expressions. This study proposes the use of symbolic regression, via the PySR genetic programming algorithm, to derive accurate and interpretable equations for predicting airborne salinity. The found equation obtained the highest test score (r2=0.76) compared to the literature models.