Characterization and forecast of airborne natural radioactivity in Mallorca

Abstract

Radioactivity is easily found in the environment due to the existence of radionuclides of cosmogenic, naturally occurring, and anthropogenic origin. Quite relevant given its potential mobility is the radioactivity associated to aerosols. Radionuclide content in the atmosphere is systematically measured as key part of most surveillance programmes. The acquired information is key in emergency situations, but it can also be used to model aerosol transport in the atmosphere. This thesis is devoted to the study of the main properties of 7Be and Gross Beta (GB) in near surface air. The work explained in this thesis has been done at the Environmental Radioactivity Laboratory of the University of the Balearic Islands (LaboRA-UIB). It makes use of a dataset composed by 7Be and GB activity concentrations in air, sampled at the University of the Balearic Islands Campus between July 2004 and December 2014. The dataset is complemented by a set of meteorological parameters (temperature, precipitation, relative humidity, and wind speed) gathered at a station close to the sampling point. One of the main objectives of this thesis is to describe the existing relations between the variables that conform the dataset, to better understand the processes that influence the dynamics of radioactivity in near air surface air. Additionally, several forecasting models are developed to predict 7Be and GB activity concentration values. Three different dimension reduction techniques are applied: Principal Component Analysis (PCA), t-distributed Stochastic Neighbor Embedding (t-SNE), and Uniform Manifold Approximation and Projection (UMAP). PCA brings a general visualization of the dataset although doesn’t allow to grasp the details of the relation between the different variables. UMAP on the other hand moves a step forward allowing to identify several clusters of states with identifying and unique properties. To the best of our knowledge this is the first use of the UMAP algorithm in an environmental radioactivity study, and the obtained results point out that it can be a useful research tool in this field. The relations between the radiometric and meteorological dataset variables allow to exploit a neural network algorithm to predict 7Be and GB activity concentrations. The obtained results applying this non-linear model are satisfactory and outperform those obtained by a Multiple Linear Regression model, although the improvement is not so significant as to discard the alternative linear model that is simpler to implement and interpret. Finally, a set of forecasting models based on the previous values of the activity concentrations are developed. Although Seasonal Autoregressive Integrated Moving Average (SARIMA) models are traditionally used in these kinds of studies, it is shown that Exponential Smoothing (ETS) models provide similar results and that they can even outperform SARIMA results. ETS are simpler and faster models, and might then be preferable compared to SARIMA models. The results presented in this thesis provide a proper characterization and understanding of the atmospheric 7Be and Gross Beta activity concentration in Mallorca, and provide evidences that support the idea that there exist alternative techniques to those traditionally used in similar studies, that are easier to apply and optimize providing similar or even better results. In particular, both UMAP and ETS might be extremely useful in the near future in the field of environmental radioactivity.