ForeFlame: Wildfire Prediction Dashboard
ForeFlame is a dashboard designed for Fire Analysts to view the probabilities of wildfire occurrences, along with other relevant parameters like vegetation, terrain, temperature, wind speed, and soil moisture, to make informed tactical decisions on wildfire management and mitigation.
Live site: foreflame.com
About the Team
We are a team of University of Washington graduate students (Global Innovation Exchange) driven to understand the behavior of Wildfires. This project is being sponsored by Microsoft and supported by Department of Natural Resources.
Our Vision is to Provide a fire management tool to predict the probability of wildfire occurrence and provide a comprehensive data source for fire analysts to make informed tactical decisions.
Given the complexity of gathering and analyzing data for accreditation and the growing sophistication of technology systems available to fire departments, ForeFlame believes a standard visualization for fire analysts is an important next step in mitigating Wildfires.
My Role:
- Back-end development: Configuring and maintaining Azure Virtual Machines and disks, facilitating Machine Learning environments, postprocessing outputs from ML, creating back-end API endpoints through Flask application, and configuring and maintaining NGINX in Ubuntu for back-end APIs.
Architecture
ForeFlame uses various data aggregation techniques and machine learning methods to provide the possibility of occurrence of wildfires. There are three main components in its architecture:
Data Sources and Machine Learning
Data Sourcing and Machine Learning Architecture
The following are the different data sets currently being used to train and infer from the machine learning model.
- Weather data from NOAA.
- NDVI (Normalized difference vegetation index) from MODIS
- Wildfire fuel data from LANDFIRE.
From the datasets above, a DNN model is used to predict the probability of occurrence of wildfires, with another KNN model trained with NDVI datasets which acts as a mask to the existing predictions. The machine learning model itself is run in Azure VM using FarmVibes workflows.
Postprocessing and back-end APIs
Post processing ML outputs to GeoJSON
The machine learning outputs are converted to GeoJSON, and hosted through a Flask application, exposed through NGINX server (Azure VM).
Visualization
Front-end Architecture
The data is displayed through the Webapp built using NextJS and mapping tools from MapBox and Turf.js.
Features
The following are the functionalities currently implemented in ForeFlame.
Secure OAuth 2.0
Secure authentication using Google Firebase for storing user preferences and bookmark polygons.
Popup for each data points
View individual parameters of each pixels by clicking on each data point.
Create bookmarks
Create bookmarks using polygons, which are linked to individual accounts and can be accessed anywhere.
Navigate using bookmarks
Use bookmarks to keep tabs and navigate to places.
Tooltips
View the function and usage of each component through helpful tooltips.
Compared to the existing solutions, ForeFlame aims to integrate more comprehensive datasets to provide the high-accuracy predictions of fire occurrence probability. Designed by incorporating real-time feedback from fire-analysts from DNR, and with ML models run using FarmVibes, it also provides an intuitive interface with high responsiveness.