Estimating high temporal resolution wildfire behaviour using Sentinel-2
(2026) In Student thesis series INES NGEM21 20261Department of Earth and Environmental Sciences (MGeo)
- Abstract
- Wildfire observations are increasingly important as changing wildfire regimes globally affect societies and ecosystems. Wildfire behaviour, comprising the way that wildfires spread through the landscape, connects wildfire to its physical controls and has been shown to predict other wildfire characteristics. However, observations of wildfire behaviour are limited due to the spatial and temporal resolution of available observation methods. This thesis aims to investigate wildfire behaviour from the observation perspective, testing the ability to predict wildfire behaviour free from physical variables through spatiotemporal interpolation applied on low temporal resolution Sentinel-2 observations, determining whether simplistic assumptions are... (More)
- Wildfire observations are increasingly important as changing wildfire regimes globally affect societies and ecosystems. Wildfire behaviour, comprising the way that wildfires spread through the landscape, connects wildfire to its physical controls and has been shown to predict other wildfire characteristics. However, observations of wildfire behaviour are limited due to the spatial and temporal resolution of available observation methods. This thesis aims to investigate wildfire behaviour from the observation perspective, testing the ability to predict wildfire behaviour free from physical variables through spatiotemporal interpolation applied on low temporal resolution Sentinel-2 observations, determining whether simplistic assumptions are sufficient to estimate wildfire spread characteristics, and finding how often those assumptions apply to general wildfires. I tested two interpolation methods that assume a constant fire rate of spread between two observed fire perimeters, classified from Sentinel-2 imagery. The two interpolation methods differ by the path along which fire travels: the shortest lines to the two perimeters (“nearest path”) or the shortest path avoiding unburned areas (“visibility graph”). Neither method closely matched observed wildfire behaviour, though otherwise both performed similarly well. Estimated wildfire behaviour strongly tended to underestimate rates of growth, with average estimated fire arrival times being later than observations (+22 hours) and average burn areas during interpolation being underestimated (10% of total end burn area). The main reason for underestimated rates of growth is the assumption of constant rate of spread, which universally underestimated rate of growth at the start of an interpolation interval and overestimated rate of spread at the end of an interpolation interval. This result strongly suggests that fire rate of spread gradually decreases over a fire’s lifetime, indicating that measures of fire rate of spread over longer time intervals, as are common among existing products, are scale dependent, and poorly account for fire behavioural characteristics at different temporal scales. (Less)
- Popular Abstract
- Wildfires are an important feature of the natural world’s dynamics, but the modern scientific understanding of how wildfires spread is limited because it is difficult to map. Satellite images are the best way to map fires, but, depending on what satellite is used, the images either have too little detail (grainy) or take images too infrequently to actually track the spread of wildfires. My study uses a satellite mission, Sentinel-2, which takes images rather infrequently (every 5 days) but whose images are very detailed and can be used to detect active fires confidently. With images from Sentinel-2, I detect actively burning areas and areas that have already burned. Then, I try to figure out what happened in between two successive images,... (More)
- Wildfires are an important feature of the natural world’s dynamics, but the modern scientific understanding of how wildfires spread is limited because it is difficult to map. Satellite images are the best way to map fires, but, depending on what satellite is used, the images either have too little detail (grainy) or take images too infrequently to actually track the spread of wildfires. My study uses a satellite mission, Sentinel-2, which takes images rather infrequently (every 5 days) but whose images are very detailed and can be used to detect active fires confidently. With images from Sentinel-2, I detect actively burning areas and areas that have already burned. Then, I try to figure out what happened in between two successive images, separated by about 5 days, to get from one burned area to the next. I do this with two different methods, which both work by estimating “paths” along which the fire might have spread. Then I compare what I estimated to what has been captured from other satellite missions that take images more frequently. What I found is that my estimates consistently underestimated how fast a wildfire expands in between two images. While in reality fires tend to expand faster at first and slow down, my estimation assumed that the fire would expand at a fixed speed. What is surprising about this result is that it is very one-directional: the estimate is consistently too slow, and almost never too fast. I think this means that the speed of fire spread slows down as they grow bigger. However, other researchers have not acknowledged this, while estimating the speed of a fire as an average speed over a time interval. If the speed of a fire decreases over time, consistently, then the average speed over a time interval will actually underestimate how fast the fire has really expanded. Furthermore, there will be different amounts of underestimation when measured over different time intervals, meaning that the measure of a fire’s speed depends on how frequently the fire is measured. Moving forward, researchers should look into how estimating fire speed from satellite imagery depends on how often images are captured, and how much this might bias measurements of speed. (Less)
Please use this url to cite or link to this publication:
https://lup.lub.lu.se/student-papers/record/9238841
- author
- Tucker, Kadin LU
- supervisor
- organization
- course
- NGEM21 20261
- year
- 2026
- type
- H2 - Master's Degree (Two Years)
- subject
- keywords
- Geographical Information Science, Remote Sensing, Satellite Remote Sensing, Sentinel-2, Wildfire, Wildfire Behaviour, Fire Behaviour, Wildfire Behavior, Fire Behavior, Interpolation, Spatiotemporal Interpolation, Geospatial Interpolation, Visibility Graph, Spatial Analysis
- publication/series
- Student thesis series INES
- report number
- 769
- language
- English
- id
- 9238841
- date added to LUP
- 2026-06-16 11:10:26
- date last changed
- 2026-06-16 11:10:26
@misc{9238841,
abstract = {{Wildfire observations are increasingly important as changing wildfire regimes globally affect societies and ecosystems. Wildfire behaviour, comprising the way that wildfires spread through the landscape, connects wildfire to its physical controls and has been shown to predict other wildfire characteristics. However, observations of wildfire behaviour are limited due to the spatial and temporal resolution of available observation methods. This thesis aims to investigate wildfire behaviour from the observation perspective, testing the ability to predict wildfire behaviour free from physical variables through spatiotemporal interpolation applied on low temporal resolution Sentinel-2 observations, determining whether simplistic assumptions are sufficient to estimate wildfire spread characteristics, and finding how often those assumptions apply to general wildfires. I tested two interpolation methods that assume a constant fire rate of spread between two observed fire perimeters, classified from Sentinel-2 imagery. The two interpolation methods differ by the path along which fire travels: the shortest lines to the two perimeters (“nearest path”) or the shortest path avoiding unburned areas (“visibility graph”). Neither method closely matched observed wildfire behaviour, though otherwise both performed similarly well. Estimated wildfire behaviour strongly tended to underestimate rates of growth, with average estimated fire arrival times being later than observations (+22 hours) and average burn areas during interpolation being underestimated (10% of total end burn area). The main reason for underestimated rates of growth is the assumption of constant rate of spread, which universally underestimated rate of growth at the start of an interpolation interval and overestimated rate of spread at the end of an interpolation interval. This result strongly suggests that fire rate of spread gradually decreases over a fire’s lifetime, indicating that measures of fire rate of spread over longer time intervals, as are common among existing products, are scale dependent, and poorly account for fire behavioural characteristics at different temporal scales.}},
author = {{Tucker, Kadin}},
language = {{eng}},
note = {{Student Paper}},
series = {{Student thesis series INES}},
title = {{Estimating high temporal resolution wildfire behaviour using Sentinel-2}},
year = {{2026}},
}