A new methodology that uses Artificial Intelligence techniques and that has been trained and validated with data from the last 70 years makes it possible to accurately predict the average air temperature two meters from the surface one month in advance.
Knowing the temperature in advance has become a decisive factor when the summer months approach. When planning vacations, planning the energy supply or designing prevention campaigns, knowing in advance and with a certain degree of accuracy how much the mercury will mark can be very useful, especially in southern areas where during the summer the extreme heat and tropical nights are no longer a sporadic phenomenon.
Precisely, a work by the Córdoba University (UCO) has managed to develop a new methodology capable of accurately predicting the average temperature for the month of August in the South of the Iberian Peninsula. The methodology, which combines Artificial Intelligence and clustering techniques, has been trained and validated with reanalysis data from the last 70 years obtained by mathematical models that are fed by observations from various information sources, such as satellites and radiosondes. Once trained, it is able to predict the average air temperature two meters from the surface.
“This is a methodology that aims to create models of Artificial Neural Networks capable of obtaining better results than other current techniques and that, in addition, are interpretable”, point out Antonio Manuel Gómez and David Guijo, co-authors of the research, both members from the AYRNA group of the University of Córdoba and researchers at the UCO Higher Polytechnic School.
This is how the methodology works
A kind of sweep is carried out (every 0.25 degrees latitude/longitude) selecting only the inland geographic points of the southern part of the Peninsula. In each of these points, the predictions are made in the month of August using input variables corresponding to the month of July, such as temperature, wind components or mean pressure at sea level.
In total, 270 points distributed throughout the south of the peninsula are analyzed and are grouped into six sub-regions with a similar behavior in terms of air temperature. It is about what in Artificial Intelligence is called clustering algorithm or ‘clustering‘, through which groups formed by data that share similarity are obtained, and which is very useful to improve prediction.
‘Interpretable’ artificial intelligence
One of the main advantages of the developed methodology, underlines the researcher Pedro Antonio Gutiérrez, is that it falls within the field of Explainable Artificial Intelligence, known as XAI for its acronym in English.
Thanks to this type of tools, the human being is able to interpret how the system makes predictions, see how the different variables interact with each other and understand the cause-effect relationship between them, in contrast to other ‘black box’ methods in that not even the people who have designed them are capable of understanding the reason why the Artificial Intelligence model makes a certain prediction.
As explained by the Emeritus Professor and principal investigator of the AYRNA group, César Hervás, these models are increasingly in demand, since they allow us to interpret interactions and deduce, for example, the causes for which the temperature can fluctuate in a certain area.
The work, in which the Universities of Alcalá and East Anglia (United Kingdom) have also participated, is part of the ORCA-DEEP research project, which addresses the study of problems related to meteorology and the environment through new methods of Artificial Intelligence.
