UdeC researchers develop a new statistical methodology to predict natural and social phenomena

A group of researchers led by Dr. Guillermo Ferreira, director of the Department of Statistics at the University of Concepción, developed a new methodology that will make it possible to predict macro and micro economic variables measured at different times and locations; epidemiological variables, such as COVID-19 data recorded since its inception and in different geographical areas; environmental variables such as temperature, wind speed, humidity, among others.

“In this work, a new statistical methodology is proposed to predict phenomena that occur both in the temporal and spatial domains, considering multiple variables of interest… in this context, the application of our study considered the analysis of a set of bivariate data on the mean of daily temperatures and solar radiation from 21 meteorological stations located in three regions of central-southern Chile: Maule, Biobío and Araucanía”, explains Dr. Ferreira. These three locations concentrate more than 70% of the country’s agricultural production, which depends on variables such as temperature and solar radiation.

The study took five years to develop and represents a multivariate spatiotemporal process through the well-known Wold decomposition. “Such an approach allows an easy implementation of the Kalman filter to estimate linear temporal processes that exhibit short-range and long-range dependencies, together with a spatial correlation structure”, comments Ferreira, who developed the theoretical-practical foundations of this new methodology.

This implies carrying out a bibliographical review of the current methods and comparing them with the new proposal, in addition to programming the algorithm in C language code.

“The next step of the research is that our proposal considers a typical problem, shared in various practical applications, and that is that many multivariate spatiotemporal data sets are affected by missing data.

This is a promising topic for future research, which will make it possible to address missing values ​​in multivariate spatiotemporal data”, explains Dr. Ferreira, who also comments that they could not find any set of applications or software that would fit spatio-temporal models. temporary to use in this study, so the group of researchers plans to develop a complete multivariate package that takes full advantage of the results obtained in this study.

In addition, they hope to write a further paper with a specific comparison between the existing methods. “We take it as a challenge for future research”, concludes the researcher.

For more details of this research it is possible to visit: https://link.springer.com/article/10.1007/s00477-022-02266-3