Designing a Visual Fire-Predicting Algorithm for U.S. Counties Using Weather Datasets

Abstract:

From the Woolsey and Camp Fire in California back in 2018 to the recent bushfires in Australia, this headlining natural disaster requires our immediate attention. A way to battle these wildfires could lie in the use of fire-predicting maps based on the weather data from a given area. The engineering goal of this project was to develop an algorithm to forecast fires in U.S. counties based on weather datasets of temperature, humidity, and wind speed for the public to have easy access to such warnings. A Python program was written that would process weather datasets through several conditions that checked to see if the dataset met certain fire-causing temperature, humidity, and wind speed thresholds. The thresholds were determined using weather datasets (from the online public database ClimaCell) taken from days counties had a fire and did not have a fire. If the dataset passed all conditions, it was determined as susceptible to a fire, and a marker was displayed on a Google map for that location. To test the program, data from four counties (Riverside, Plumas, Inyo, and Siskiyou) in California on a day a fire occurred in the areas was run through the program. To ensure that the program was not incorrectly predicting fires, data from Tampa Bay and Houston on days a fire did not occur was also run through the program. The project goal was accomplished because the algorithm accurately predicted fires and displayed data on a map. The algorithm was able to predict that a fire would happen during the day at an approximate, but not exact, hour. The program could be made more precise by adding in other vital fire-causing factors, such as thunderstorms, vegetation aridity, power line location, and drought history. Current and future weather data could be applied to this visual fire-predicting model to determine an area’s future fire susceptibility, drastically increasing awareness and decreasing the fire’s damage.