Background

It is well known that most fires are human caused, this adds a significant difficulty when trying to predict where and when a fire will occur, and although the information is varied, in terms of environmental characteristics such as temperature, humidity, surrounding flora, etc., the task does not get any easier.

For this challenge, we have set as our main mission to find ways to predict a fire beyond what some institutions have already achieved, that is why we use historical data provided by MODIS available in the Fire Information for Resource Management System (FIRMS) to identify patterns to predict areas that could be at higher risk of fire occurrence. The result will be an easy-to-access website where users can select a calendar date in the future and visualize if they or their family members could be in an area at risk.

Phase I Understand the historical data provided by MODIS in the Fire Information for Resource Management System (FIRMS) and select a specific area to work in.

There are numerous datasets available to support our project's mission. However, we chose to rely on NASA's most renowned dataset. Our initial task was to decide between utilizing the Moderate Resolution Imaging Spectroradiometer (MODIS) or the Visible Infrared Imaging Radiometer Suite (VIIRS) dataset. We chose to concentrate on MODIS due to its wider data coverage, as its temporal coverage spans from November 2000, while VIIRS has fewer years of data. Additionally, MODIS has a lighter data load compared to VIIRS. Therefore, we decided to focus on MODIS for our analysis.

Our team's second task was to identify where to launch the project. We chose Mexico, our country of origin, due to the absence of dedicated fire prediction systems. We also chose Great Britain since it is the challenge's headquarters.

Phase II Data usage

The MODIS database was used to organize information into small pivot tables, enhancing data usability. A main table was then created to filter data and generate the primary web site map. The resulting map offers improved geographic visualization of historical fires since 2000.

Phase III A user-friendly Web site with a key fire classification designation

After finalizing the historical data on the site, we made it our mission to streamline the information and improve the site's user-friendliness. To achieve this, we utilized The National Fire Danger Rating System's (NFDRS) five color-coded risk levels. The "Brightness" variable's maximum value was subtracted from its minimum value, and the result was divided among the five classifications to determine the fire risk level (In collaboration with specialists, a more precise classification regarding how to categorize a forest fire will be established as future work.). In the present we utilized The National Fire Danger Rating System's (NFDRS) five color-coded risk levels

Fire Danger Level: Low (300.6 - 315.64) Blue Color

When the fire danger is "low" it means that fuels do not ignite easily from small embers, but a more intense heat source, such as lightning, may start fires in duff or dry rotten wood. Fires in open, dry grasslands may burn easily a few hours after a rain, but most wood fires will spread slowly, creeping or smoldering. Control of fires is generally easy.

Fire Danger Level: Moderate (315.64 - 330.68) Green Color

When the fire danger is "moderate" it means that fires can start from most accidental causes, but the number of fire starts is usually pretty low. If a fire does start in an open, dry grassland, it will burn and spread quickly on windy days. Most wood fires will spread slowly to moderately. Average fire intensity will be moderate except in heavy concentrations of fuel, which may burn hot. Fires are still not likely to become serious and are often easy to control.

Fire Danger Level: High (330.68 - 345.72) Yellow Color

When the fire danger is "high", fires can start easily from most causes and small fuels (such as grasses and needles) will ignite readily. Unattended campfires and brush fires are likely to escape. Fires will spread easily, with some areas of high-intensity burning on slopes or concentrated fuels. Fires can become serious and difficult to control unless they are put out while they are still small.

Fire Danger Level: Very High (345.72 - 360.76) Orange Color

When the fire danger is "very high", fires will start easily from most causes. The fires will spread rapidly and have a quick increase in intensity, right after ignition. Small fires can quickly become large fires and exhibit extreme fire intensity, such as long-distance spotting and fire whirls. These fires can be difficult to control and will often become much larger and longer-lasting fires.

Fire Danger Level: Extreme (360.76 - 375.8) Red Color

When the fire danger is "extreme", fires of all types start quickly and burn intensely. All fires are potentially serious and can spread very quickly with intense burning. Small fires become big fires much faster than at the "very high" level. Spot fires are probable, with long-distance spotting likely. These fires are very difficult to fight and may become very dangerous and often last for several days.

Phase IV Artificial intelligence algorithm for the prediction of fire risk areas on the basis of historical data.

For the development of the artificial intelligence algorithm, historical data provided by MODIS available in the Fire Information for Resource Management System (FIRMS) was downloaded. The objective was to analyze these data using 4 classifiers:

KNN - k-Nearest Neighbor

Decision Tree

Discriminant Analysis

Naive Bayes

However, due to the volume of data and the available time, it was not possible to validate results using these algorithms. Therefore, we opted to use a linear regression-based algorithm. Linear Regression was used as a model to predict the variable of interest.

Future Work?

As key points for our platform, we highlight the following aspects:

Validation of the artificial intelligence algorithm for fire prediction based on historical data.
Mapping of data from around the world (not only from the United Kingdom and Mexico).
Establishment of additional parameters for fire prediction, such as satellite images combined with readings provided by MODIS.

2023 NASA Space Apps Challenge

Fire Guardians