In 2014, a series of forest fires [1] swept through Yakutia, destroying hundreds of thousands of hectares of forested areas. The fires occurred in various parts of the region and progressed dynamically, consuming significant forested areas daily. Satellite images were widely used for rapid assessment of fire impacts and the current situation; however, typically, the decryption process requires specialist involvement. To automate the process, Innoter's staff developed a model utilizing free satellite images from Landsat-8 and machine learning to promptly detect the presence of fires in the images and assess the area of burnt territories. The processing results and overall algorithm flowchart are presented in the image (Figure 1).
Figure 1. Result of automated processing of Landsat-8 satellite image over the fire area in Yakutia
The system operates as follows:
Upon receiving a new image over the area of interest, data is automatically uploaded to the server, where it undergoes preprocessing and quality assessment procedures in automatic mode (utilizing Landsat QA data [2]).
Next, automatic classification is performed using machine learning algorithms [3]. The image area is divided into 5 classes:
- Burnt areas (where present)
- Intact forest cover
- Water surfaces
- Cloud cover
- Other vegetation
Subsequently, upon detecting fires that pass the confidence threshold, their areas are assessed and provided as a geospatial report, including date, coordinates, and areas of identified fires.
Thus, this system allows for rapid (within minutes) localization of fire spots and calculation of their area across territories of 30,000 sq. km. and more.
References:
[1] "Fires in Yakutia in 2014," [Online]. Available: h ttp://yakutiamedia.ru/news/374184/.
[2] "Landsat 8 Pre-Collection Quality Assessment Manual," [Online]. Available: https://landsat.usgs.gov/qualityband.
[3] "Analysis of Maximum Likelihood Classification on Multispectral Data," [Online]. Available: http://www.m-hikari.com/ams/ams-2012/ams-129-132-2012/ahmadAMS129-132-20....