The proposed system has two parts: the management dashboard and the mobile app. The management dashboard must show active fires and high-risk areas in a map divided in several cells that form a grid. The high-risk areas are cells that have borders to cells with active fires and are classified as such by a machine learning model. Such a model takes into account the relative humidity in the cell on fire, wind speed, wind direction, atmospheric temperature and fuel to predict whether the fire will reach a given cell and when. The target of the training database for a given sample (cell) is how long the fire from a neighbor burning cell took to arrive, since the fire started in the given neighbor cell. Also, using land cover data, cells with human constructions are detected and when these areas are at risk, a stronger alert is emitted. With such a feature, a commander can take strategical decisions in order to better distribute firefighters and brigadists in an attempt to minimize human, wildlife and economic losses, even though firefighters and brigadists are scarse. Also, with this data being quickly available, important actions can be taken as early as possible.

However, there are periods and places that remain in blind spot areas for some time, which can be critical to avoid major damages. Therefore, an alternative mean for early notification of wildfires is through crowd-sourcing. Anyone using the mobile app can quickly send a geolocalized notification to the system, which will be shown in the strategical dashboard. This way, blind spots in the satellite-provided data can be diminished by the users of the mobile app. Our hypothesis is that we can reach nearly real-time notifications of active fires with a higher coverage than the sattelite-only based solutions. Also, notifying a fire station through a telephone call can be very inefficient, since a fire notification usually takes some time to reach a commander and be transformed into an effective action. Thus, the fire notifications sent through the system can quickly reach whoever is in charge to take strategical decisions and reduce the time to response. In case the user does not have internet connection, what usually happens in poor countries when the user is far from urban areas, an SMS with its location as text content is sent to the central system, which must receive the text, parse it and store the sent notification.

Still regarding the dashboard visualization, the commander can also quickly notify registered firefighters and brigadists to show up as soon as possible in the fire station in order to give the briefing of the mission and combat a given notified wildfire. Besides, the people that are located in high risk areas and are using the mobile app can receive a fire alert whenever the commander finds it necessary. Such a decision can be taken by observing the areas classified as high risk by the prediction model. The mobile app will then show the alert to the user and it will be aways showing a map with the high risk area highlighted. Our hypothesis is that such a solution can reduce the number of deaths or material losses if people leave early with their most importan belongings from high risk areas.