FiRadar - Greece


A tool that predicts the risk level of wildfires in Greek regions

Scope

FiRadar is a fire prevention application that continuously monitors areas in Greece in near real-time, providing up-to-date information regarding their risk level.
FiRadar empowers decision-makers with valuable insights to take proactive measures in fire prevention and resource allocation. By leveraging advanced analytics, machine learning, artificial intelligence algorithms and visualization tools, the application assists in developing effective fire management strategies and enhancing emergency response.

FAQ

How it works ?

  • Retrieve weather data for each selected area.
  • Input the data into the developed Artificial Neural Network model.
  • The model generates the fire risk level for each area.
  • Update the map based on the results and categorize the cities into four risk classes.
  • The entire process is automated and repeated every four hours.

How was it implemented ?

  1. Collect past data for fire incidents in Greece as well as temperature, wind speed and dew point.
  2. Develop an Artificial Neural Network model using the collected data.
  3. Scrape the current temperature, wind speed, and dew point for each area to input into the model and classify the areas based on their risk level.
  4. Display the information on a dynamic map.

Which tools does it use ?

Implementation code was written in Python 3.8 using the following libraries:

  • Numpy and Pandas: For data processing.
  • Sklearn, Tensorflow, Keras: For implementing various machine learning algorithms.
  • Folium: To visualize the results on a dynamic map.
  • BeautifulSoup and Requests: For the web scraping process.

Details

Historical data on wildfires that occurred in Greece, along with weather data, was collected. Various data mining techniques were employed to convert the data into valuable information. Multiple Machine Learning models were developed and carefully evaluated, leading to the selection of the Artificial Neural Network model, which showed the most promising results. To maintain up-to-date information, current weather data for each city is retrieved through the Web Scraping process and fed into the model. The model's outcomes are classified into four risk classes, and the dynamic map is updated every four hours.