Gaussian mixture analysis of basic meteorological parameters: Temperature and relative humidity

Abstract

Gaussian mixture modelling was applied to describe the annual distribution of two important meteorological variables, temperature and relative humidity, inside the Costa Rican Central Valley from 2010 to 2017. A fixed number of components of Gaussian mixtures were used to fit data to a general mixture curve that represented data behavior throughout the year, this was performed through specific functions of Scikit-learn and SciPy libraries of Python language. Low values of approximation error were obtained when modelling temperature data and the relationship between its distribution and hourly variability was observed, finding high values around noon. For relative humidity, the Gaussian mixture model presented issues when fitting values greater than 90 %, as a result of this variable saturation limit at 100 %. The relationship with time was not clearly determined due to the many mixture components used to model, but a tendency of low values between the late morning and early afternoon was visualized. Iterative minimization of the error was considered as a future approach to achieve a better fit with Gaussian mixtures of these and other meteorological variables.