In the saga of The Foundation of Isaac Asimov appears a fictional science, psychohistory, which uses history and statistics to, based on the past, predict the future. Without going to these extremes, in the computers of the Ministries of Agriculture, Fisheries, and Food and Ecological Transition there is a similar tool that is capable of predicting where fires will occur and allow, in a way, to extinguish them even before the first flame appears the ARBARIA project.

This project is composed, in turn, of three complementary projects two of them have more to do with extinction and the third, with prevention. Thus, while the first two have to do with the mobilization of means to reduce fire in case of occurrence, the third seeks to know what happens in the territory, what characteristics are important and have consequences in the occurrence of fires, as explained by Antonio Lopez Santaella a worker from the Forest Fire Defense Area of ​​the Sub-Directorate General for Forest Policy and the Fight against Desertification. Together with Lopez, Faustino Sanchez Garcia works from the Data Analysis area of ​​the SG of Information Technologies and Communications of the Ministry of Agriculture. In a way, Sánchez is the one who creates the program and Lopez is in charge of contextualizing the data he provides.

A project of this type, Sanchez explains, has three pillars technological (the program itself), mathematical (the application of statistics), and knowledge (to understand what the data mean with fires). Thus, what the tool allows is, among other things, to search for a municipality and see what the risk of fire is in that area. But, above all, it allows knowing the socioeconomic parameters (age of the population, number of farms, how much livestock there are. that make this happen.

In the beginning, we took this type of data and we did not know if it made much sense, acknowledges Sanchez. That is why it is very important to be in constant contact with the specialist unit. The issue of fires in particular at the national level, not only in Spain, Lopez adds. Fire is a very complex phenomenon that affects many circumstances, particularly the climate, but human intervention is also fundamental. And behind the human intervention in the occurrence of fires, there are different patterns or population characteristics that determine the use, for example, of fire or accidents or negligence associated with different practices.

The basic work and the exploitation of these data with algorithms allow identifying which socio-economic characteristics of the territory most influence each municipality in the occurrence of fires. In this way, they can act in areas with similar occurrences and different characteristics and adapt the campaigns to each area. These campaigns can be communicative (in this case, the data would allow us to know what range of the population they want to reach) or direct prevention activities. If we know that livestock in some places is a determining factor in the occurrence, we would have to work with the livestock sector, summarizes Lopez. That is why it is so important to understand these patterns of causality associated with different factors in the territory.

The problem with putting out a fire before it happens is that there is no burned ground left to show that prevention has been successful. In any model of this type, it is essential to be able to contrast this admits Sanchez. For this reason, they have created a functionality that allows, in a way to simulate these predictions in the past and compare the result with what happened. In this historical test, at a general level, the tool is capable of achieving 69% accuracy. Of course just by statistics, due to the large number of fires that occur, it would be 47% correct.

Of the remaining 53%, therefore, 22 percentage points (that is, 41%) should be assigned to these socioeconomic factors. The Ministry has prevention teams that are distributed throughout the territory. It is personnel who work in extinction during the winter and in the summer, reconciling interest in areas where there are fire problems, explains Lopez. Therefore, the knowledge of which sectors or in which areas they should develop their work is essential. Also, the data can assist Seprona in its surveillance and investigative work.


The third part of the project is dedicated to fighting fires. In this case, artificial intelligence cannot be applied so much in real-time to know how the fire will behave – although data on fuel, territory, or meteorology are collected and analyzed to monitor it, these are different tools – but rather to know how resources must be allocated to literally put out the fires.

This functionality is trained with data from the summer campaign and for its predictions, it uses deep learning algorithms and takes into account both the weather (by default, with AEMET data) and what happened in previous weeks because there is usually a lot of autocorrelation in these events, contextualizes Sanchez. During the demonstration carried out by the technicians, the program faces week 31 of the year with enough ease and success (where it predicts 15, there were 22, for example). In any case, what is important is to be able to know the order of priority of risk; that is, the province with the highest risk, according to Sanchez.

The reason is explained by Lopez in matters of fire extinguishing, the competence is distributed to the Communities, while the Ministry coordinates and supports. It has many means that it can distribute throughout the territory (and that, in fact, it distributes) and many times they are decisive in the extinction. The resources are distributed before the summer campaign (and, in fact, statistics are used to do so) in the places with the greatest fire problems however, sometimes they have to be mobilized in the middle of the summer season. Knowing in advance in a week what is foreseeable that we can find in terms of the priority of places with greater risk allows us to anticipate in the mobilization of these means or, even, to carry out a more exhaustive surveillance and deterrence work.


The project has a budget of about 25,000 euros that is used to pay for the necessary cloud computing service provided by Amazon Web Services and to a company that helps with this technology. Everything else, which is related to development are hours of work for officials, summarizes Lopez. Of four, specifically, if we ignore the work that the Ministry of this one does.

Our sub directorate is a very small area: now we are four people, Sanchez acknowledges. Apart from him, there is a person who works in the descriptive part of the project and another two dedicated to the predictive one.

Together they have managed to implement a project that, in its own way, predicts the future. Will there come a point further away than the next few summers where we will look back and be surprised to see the numbers of fires we are currently facing or the mere fact of having to? Statistics show us that year after year there are fewer fires in the less affected area, and even fewer major fires, Sanchez shares. However, the world also changes, and although the fires are fewer, the risk increases. We must not forget that in the Mediterranean context especially and the Iberian, with the internalized use of fire in some regions, the fire will always be present, he laments. The good thing would be to have the ability to anticipate its occurrence or generate a culture of risk so that we are more prepared for the occasions when it occurs.

Faustino is a bit more optimistic and recalls that this type of project arises in contexts in which there is uncertainty. With fires, the degree of uncertainty is always very high our job is to reduce that. Asimov’s psychohistory was able to predict the fall of the Galactic Empire the more humble ARBARIA project is content with something as human as being prepared.