La Inferno| Spot That Fire V3.0
Spot That Fire V3.0
Predict, Analyze and Visualize : Wild Fire Occurrence
Summary
As climate change has become a key factor in increasing the risk and extent of wildfires across the globe, this prototype addresses wildfire damage prediction and visualization through a Web App for prevention-planning programs aiming to reduce human and material loss.Highlights•Predict Wild Fire damaged area using SVM classification•Analysing and visualizing Fire Weather parameters for fire intensity•Rescaling the burnt area using Logarithm function to improve symmetry•Correlating the Fire Weather parameters and extent of damage by Wild Fire•Develop a web App-Spotify Wild Fire to view Wildfire damage details
How We Addressed This Challenge
Why is it important?
Post Wild Fire disaster prediction of damage and visualization of damage area to help Disaster Management authorities view details from anywhere using Web App- Spotify Wild Fire.
What was developed & its purpose:
This team proposes Wildfire prediction damage through Machine Learning - SVM sigmoid classification, visualization of the Fire Weather Data and the extent of damage that can be viewed in a Web App developed called Spotify Wild Fire from anywhere by the concerned authorities.
The prototype comprises of the following modules:
· Predict: Support Vector Machine (SVM) is a popular Machine Learning algorithm for classification. Binary sigmoid function was found to be more appropriate to classify: High probability fire damaged area and low probability fire damaged areas.
· Analyze: The burnt area has been rescaled with logarithm function to reduce skewness & improve symmetryfor better inference.
· Visualize: The single variable dataset was visualized using Histogram & Kernal Density. To depict fire intensity with model features & natural features KDE plots were used. Heat Maps for temperature range in different areas within the park and fire intensity of burned areas in the Montesinho park are depicted. Both the heat maps were used to depict the summer conditions in the considered area.
· Web App: Was designed to visualize the fire weather data, dataset considered and the output of prediction and visualization.
What we hope to achieve:
Visualizing current and past wildfire events by identifying location subjected to fire damage serves as a vital tool for addressing wildfire prevention-planning programs aiming to reduce human and material loss.
Full Demo Link of Prototype:
How We Developed This Project
Challenge chosen?
Spotify Wild Fire V3.0 submission track is chosen as forest fires are a major environmental issue leading to economic, ecological, infrastructure destruction alongside endangerment to animal and human lives.
Approach & tools:
The prototype addresses this issue using Machine Learning - SVM to determine high probability fire damaged area and low probability fire damaged areas. Then the visualization of Fire Weather data and damaged area during summer using graphs is proposed for the concerned authorities for post disaster management. The approaches & tools used are as follows:
Machine Learning
SVM Classification-Python 3
• Sklearn
• Pandas
• Numpy
• matplotlib
Analyze & Visual
Python3
• Regression model - scipy.
• matplotlib.pyplot
• seaborn
• mpl_toolkits.mplot3d
Web Application
· Firebase – Cloud service for SAAS application
· HTML – For Web page creation
· ReactJS - platform for creating interactive UIs
Problems faced:
· We were unable to extend the services of Google Cloud hence had to shift to Firebase
· APIs for connecting & executing Python code from backend could not be completed
How We Used Space Agency Data in This Project
The standard Machine Learning dataset for forest fire was used for Montesinho natural park, a region in the northeast region of Portugal has been used. This region has around 53,300 fire pixels detected in 2000-2016 in Portugal and this region of interest has been encompassed within a 9×9 grid (x and y axis) for prediction, analysis & visualization.
Reference:
1.forestfires.csv dataset accessed from UCI Machine Learning Repository
https://archive.ics.uci.edu/ml/datasets/Forest+Fires
This dataset comprises of:
· Data Set Characteristics – Multivariate
· Number of Attributes: 13
· Attribute Characteristics: Real
· Number of Instances: 517
2.Similarly for the Fire weather data inference, the Canadian Forest Fire Weather Index (FWI) System has been referenced
https://cwfis.cfs.nrcan.gc.ca/background/summary/fwi
Project Demo
https://youtu.be/c1roTHKmWkE
Project Code
Data & Resources
1. UCI Machine Learning Repository
https://archive.ics.uci.edu/ml/datasets/Forest+Fires
2. Canadian Forest Fire Weather Index (FWI) System
https://cwfis.cfs.nrcan.gc.ca/background/summary/fwi
3.Paulo Cortez and Anibal Morais , “A Data Mining Approach to Predict Forest Fires using Meteorological Data”, New Trends in Artificial Intelligence, pp. 512-523, 2007
4.Vikram Devi Eswaramoorthy, “Increasing the Statistical Significance for MODIS Active Fire Hotspots in Portugal Using One-Class Support Vector Machines”, Master’s Thesis, Technische Universitat Munchen, 2017
5.National Wildfire Coordinating Group,
https://www.nwcg.gov/publications/pms437/cffdrs/fire-weather-index-system, 2019
6.Sophie Tan, “Predicting Forest Fires with Machine Learning”, Maplesoft.com 2020
7.Forest Fire Area Estimation using Support Vector Machine as an Approximator. Available from: https://www.researchgate.net/publication/327912193_Forest_Fire_Area_Estimation_using_Support_Vector_Machine_as_an_Approximator [accessed Sep 30 2020].