Abstract – Since European settlement, Australia is going through massive Deforestation, endangering its wildlife and opening new doors for a climate crisis. Around 16% of Australia is covered by forests but that too is rapidly decreasing because of deforestation. From June 2019 to May 2020, massive bushfires destroyed a large area of Australia’s forests and other healthy natural vegetation. This bushfire did the most destruction from the end of 2019 to the beginning of 2020 and it became major news worldwide. Thus, analysis of the destruction done by this bushfire is important to calculate how much forest and healthy natural vegetation is lost and needs to be recovered. The increase of Landsat-class data opens up new opportunities for analysts to detect and calculate changes in biomass easily. The Landsat program has provided earth observation for over four decades continuously. With easily accessible high-resolution Landsat 8 data, it is becoming easier for observers to visually identify urban areas, water bodies, healthy and unhealthy vegetation, farmlands and even clouds. In this study, we used satellite images of a particular area of our choice in Australia, from 20 January 2019 and 23 January 2020, and calculated the change of healthy natural vegetation during this period. Images of 11 different spectral bands from Landsat 8 provide enough data in very resolution for unsupervised or supervised classification in ArcGIS, which is then used to classify vegetation and calculate the area associated with it. Using satellite data contributes significantly to reducing errors and detecting changes in land cover more accurately.
Keywords – bushfire, wildfire, earth observation, Landsat, forest monitoring.
I. Introduction
Australia is the seventh-largest continent and the sixth-largest country in the world, surrounded completely by water, being the only country to occupy the whole continental mass. Despite being one of the driest countries, Australia is one of the most developed countries in the world. It has seen a growth in the deforestation rate recently and wildfires are making the condition worse for Australia’s healthy vegetation. Australia’s worst natural disaster was probably the 7 February 2009 Black Saturday bushfires [1]. But the recent bushfires from June 2019 to May 2020 was probably the worst of them all. It cost about 34 direct and over 400 indirect deaths and more than 9300 homes got destroyed by the fire. In the global ecosystem, wildfires and bushfires play a very major role as they are a threat to lives and natural resources like vegetation. It is necessary to identify the fire events individually to get a better view of how fire events affect the Environment both locally and globally [3,4]. It’s also important to spot burned down forests and healthy vegetation and calculate the total affected area. This way we will be able to estimate the amount of destruction of natural resources by wildfire or bushfire. Our research work focuses on that problem and solves it with unsupervised classification and calculation of healthy vegetation before and after the fire event using satellite imagery. Calculations of the changes in vegetation cover rely on the comparison of data-driven attributes between two images from different dates [6]. In [7], the researchers calculated changes in multi-class land cover classification maps between 1967 and 2007 and analyzed the changes and variations in the vegetation cover in Mali, Africa. More experimental approaches can be found recently, for example, methods based on neural networks and decision trees.
An increment in the recognition of the important role biomass burning plays in the global carbon cycle has been seen because of the research works conducted in recent times. Estimates of the quantity of biomass consumed through combustion are typically based on a simple relationship noted below [10].
M = ABc …(1)
Here, the mass of vegetation combusted within a given time interval is denoted by M, during the same time interval, the area burned is denoted by A, B is the biomass density, and the factor describing the completeness of the combustion is denoted by c [10]. In our study though, we used other simpler techniques to get an overview of the scale of burned down vegetation in a quicker way.
Earth’s surface is being continuously observed for more than four decades by the Landsat satellite sensor series [8]. There are 4 different types of sensors in the current and past Landsat missions, to collect EO data and the characteristics of each sensor are provided in Table 1 [8].
To detect thermal anomalies, the usage of Landsat-class data has been used in a lot of studies. To study volcanic activity, Landsat-5 Thematic Mapper data have been used widely. To inform fire management systems, Satellite remote-sensing active fire data have been extensively used for over a decade [14]. The adoption of free data policies and the launch of new instruments helped increase the availability of Landsat-class data but the spatial and temporal coverage provided by the individual Landsat-class sensors remained relatively unchanged. In our study, we used Landsat data from https://usgs.gov [16]. We used Landsat imagery of a specific region of Australia’s southeast part from 20 January 2019 and 23 January 2020 to perform our study. Landsat-8 images were downloaded in 11 different spectral bands to perform our study, while we only used different combinations of the first 7 spectral band images for our study.
II. Related studies
In 2010, Roger A. Pielke Jr. and Katharine Haynes did a study on the history of building damages their effects on the climate [1]. The study also covered the effects of bushfires in Australia from 1925 to 2009, focusing on the 2009 Black Saturday fires in Victoria. A study published in 2013 reports on the application of fire severity studies through spectral analysis [15]. The study is focused on the fire severity analysis in north Australian tropical savannas, describing the immediate post-fire spectral responses of fire-affected vegetation. For fire severity classification, the paper suggested a candidate set of models incorporating Moderate Resolution Imaging Spectroradiometer (MODIS) channels 2, 5, 6 and 7. A similar study was performed by L. Giglio et al. in 2005 which presented a global estimation of the burned area using MODIS [10]. A study published in 2016 focuses on how satellite data can be used to determine ignition points of bushfires and their probable dates [2]. A paper focusing on the implications of Landsat-8 data in active fire detection was published in 2015 [14]. One of the techniques their algorithm focuses on is detecting bring surfaces from satellite imagery and detecting potential fire and smoke. With a wide variety of training and validation scenes, the paper proposes an algorithm with very little error. A study by M. Schmidt et al., published in 2015 presents the usage of multi-resolution time-series imagery in the field of forest disturbance and growth monitoring in Australia [11]. Grahame Douglas et al. proposed the usage of extreme value analysis in bushfire probability detection [12]. The Generalised Extreme Value method was extended in their study to the determination of the Forest Fire Danger Index (FFDI) for design bushfire detection.
III. Landsat Data and Classification Methods
The characteristics of the 4 different sensors for the collection of EO data used in Landsat missions is provided in Table 1. From 1972 to 1992, the multispectral scanner (MSS) acquired data in Landsat 1–5. With the spatial resolution of approximately 80 meters coverage with radiometric coverage in 4 spectral bands, the MSS collected data. The 4 spectral bands were from visible green to near-infrared (NIR) [8]. Thematic Mapper was used in Landsat 4–5 with spatial resolution in the range from 30 meters to 120 meters. From 1984 to 2012 Thematic Mapper (TM) was used. 7 spectral bands were used to collect data. Enhanced Thematic Mapper Plus (ETM+) is being used in Landsat 7 with a spatial resolution of approximately 30 meters in 8 different spectral bands. Operational Land Imager (OLI) and Thermal Infrared Sensor (TIRS) with resolution in the range of 15 meters to 100 meters in 11 different spectral bands. It is operational since 2013. The wavelengths and usage of all the spectral bands used in Landsat 1–8 are noted in Table 2.
Dataset and Licensing:
The Landsat Data Access page states that, "Landsat data products held in the USGS archives can be searched and downloaded at no charge from a variety of data portals" [17].
Image Selection
We chose a zone in one of the worst affected areas by the Australian bushfire event from 2019 to 2020. Fig. 1 shows the selected area of our study. Our study will be confined to only this area. The area we selected is 33,300 square kilometres (approximately). The selected area is shown by a dark shade in Fig. 1. We downloaded Landsat 8 imagery of this section in 11 different spectral bands of the two dates mentioned previously. We selected these two specific dates keeping in mind that images with the lowest Land Cloud Cover (LCC) will give the best and most accurate results. On 20 January 2019, LCC was 0.15% and on 23 January 2020, LCC was 1.79%.
We used ArcGIS, which is a geographic information system (GIS), to use different band compositions and classify different areas in the Landsat imagery, such as forests, urban areas, and water bodies etc., based on different colours.
∗Following the failure of the TM sensor on Landsat 5, the MSS sensor was reactivated for a short period of time prior to satellite decommissioning [8].
Classification
Different band combinations provide different results for visual inspection and supervised classification. It also helps the unsupervised classification. Different combinations of bands and their usage is presented in Table 3.
We used the 9th band combination from Table 3 to visually inspect the area of our research. Fig. 2 and Fig. 3 shows the area of our research in the ‘Vegetation Analysis’ band combination from 20 January 2019 and 23 January 2020 respectively. From visual inspection, the difference is very clear and we will perform some more steps to get the exact area deforested by the bushfire event.
We used unsupervised classification in ArcGIS to classify different regions in the imagery with different colour schemes. We colour-coded the area classified as healthy vegetation light-green. After properly classifying the healthy vegetation from both the 2019 data and 2020 data, the deforested area becomes easily visible. The classified area containing healthy vegetation from the imagery of 20 January 2019 and 23 January 2020 are shown below.
In the unsupervised classification process, the accuracy of the classification depends on the number of classes we classify different regions of the image. Too many or too few numbers of classes will result in the inaccuracy of the classification. In our study, using 7 classes gave us the best results upon visual inspection of the classified image and the original image. After the classification, we calculated the area that is classified as healthy vegetation.
IV. Results and Discussion
The total area on which our study is conducted is 33,300 square kilometres (approximately). After calculating the area containing healthy vegetation from both the dates we get the results presented in Table 4.
According to this calculated data, in the area we performed our study on, 3522.27 square kilometres of forests or healthy vegetation got deforested by bushfires or other reasons. The main reason for such heavy deforestation here is the bushfire outbreak that started in June 2019. Almost 22.46% of the existing forests got burned during this one year, from 20 January 2019 to 23 January 2020.
Detecting and calculating the damage after bushfire events is very important for land-use planning and construction practice. With the increased volume of easily available satellite imagery, it has become more efficient for researchers to use these data and estimate the aftermath of bushfire events. In our study, we used a very commonly used classification technique in ArcGIS, the unsupervised classification technique, and used it to calculate the area containing healthy vegetation on two different dates. Unsupervised classification is often used to classify water bodies like rivers and lakes but here we used it for a different purpose. This methodology is applicable for monitoring purposes at a continental scale, as well as on small scale too. It opens up new approaches in calculating the aftermath of bushfire events using Landsat data, which will be helpful for governments, corporates and environmental researchers in creating ecologically sustainable infrastructures in the future.
V. Conclusion
The Landsat series of satellite sensors are one of the longest-running Earth observation programs and it is still monitoring the Earth. The Landsat data is one of the most used data for forest management and analysis after the free and open distribution policy of Landsat data in 2008, and it is very well suited for this as Landsat data is offered in spatial, spectral and temporal resolution.
This paper describes the methodology used to calculate the spread of healthy vegetation from Landsat 8 data but this methodology can be used flexibly to calculate the spread of other features like urban areas, water bodies, farmlands, etc.
Data from satellites can contribute significantly to fire detection with accurate information and with our methodology, it becomes easier to monitor forests and natural resources after a fire event. This will contribute to automated burnt area mapping systems and fire intensity analysis systems.
I hope you enjoyed reading this article. If you did, please become a member of Medium. Just for 5$ per month, you can read any article on Medium (not just my articles, any article). Click the link below.
https://samratduttaofficial.medium.com/membership
This article is written by me, Samrat Dutta, and my friend Jinia Parvin. Both of our information is below.
Samrat Dutta:
Github: https://github.com/SamratDuttaOfficial
Linkedin: https://www.linkedin.com/in/SamratDuttaOfficial
Buy me a coffee: https://www.buymeacoffee.com/SamratDutta
Jinia Parvin:
LinkedIn: https://www.linkedin.com/in/jinia-parvin-9a2b06211/
Github: https://github.com/jiniaparvin
References
[1] RP Crompton, KJ McAneney, K Chen, RA Pielke Jr, K Haynes. Influence of location, population, and climate on building damage and fatalities due to Australian bushfire: 1925–2009. Weather, Climate, and Society. 2010; 2, 300-10.
[2] A Benali, A Russo, AC Sá, R Pinto, O Price, N Koutsias, J Pereira. Determining fire dates and locating ignition points with satellite data. Remote Sensing. 2016; 8, 326.
[3] LB Lentile, ZA Holden, AM Smith, MJ Falkowski, AT Hudak, P Morgan, SA Lewis, PE Gessler, NC Benson. Remote sensing techniques to assess active fire characteristics and post-fire effects. International Journal of Wildland Fire. 2006; 15, 319-45.
[4] L Giglio, JD Kendall, CO Justice. Evaluation of global fire detection algorithms using simulated AVHRR infrared data. International Journal of Remote Sensing. 1999; 20, 1947-85.
[5] EA Lehmann, JF Wallace, PA Caccetta, SL Furby, K Zdunic. Forest cover trends from time series Landsat data for the Australian continent. International Journal of Applied Earth Observation and Geoinformation. 2013; 21, 453-62.
[6] P Coppin, I Jonckheere, K Nackaerts, B Muys, E Lambin. Review ArticleDigital change detection methods in ecosystem monitoring: a review. International journal of remote sensing. 2004; 25, 1565-96.
[7] D Ruelland, F Levavasseur, A Tribotté. Patterns and dynamics of land-cover changes since the 1960s over three experimental areas in Mali. International Journal of Applied Earth Observation and Geoinformation. 2010; 12, S11-7.
[8] A Banskota, N Kayastha, MJ Falkowski, MA Wulder, RE Froese, JC White. Forest monitoring using Landsat time series data: A review. Canadian Journal of Remote Sensing. 2014; 40, 362-84.
[10] L Giglio, GR Van der Werf, JT Randerson, GJ Collatz, P Kasibhatla. Global estimation of burned area using MODIS active fire observations. Atmospheric Chemistry and Physics, 2006; 6, 957-974.
[11] M Schmidt, R Lucas, P Bunting, J Verbesselt, J Armston. Multi-resolution time series imagery for forest disturbance and regrowth monitoring in Queensland, Australia. Remote Sensing of Environment. 2015;158, 156-68.
[12] G Douglas, Y He, Y Xiang, EC Morris. Use of the extreme value analysis in determining annual probability of exceedance for bushfire protection design. Fire Safety Science. 2014; 11, 1379-92.
[13] Landsat Missions, Available at: http://landsat.usgs.gov, accessed November 2020.
[14] W Schroeder, P Oliva, L Giglio, B Quayle, E Lorenz, F Morelli. Active fire detection using Landsat-8/OLI data. Remote sensing of environment. 2016; 185, 210-20.
[15] AC Edwards, SW Maier, LB Hutley, RJ Williams, J Russell-Smith. Spectral analysis of fire severity in north Australian tropical savannas. Remote Sensing of Environment. 2013; 136, 56-65.
[16] Landsat Missions, Available at: http://landsat.usgs.gov, accessed November 2020.
[17] Landsat Missions, Available at: https://www.usgs.gov/landsat-missions/landsat-data-access, accessed November 2020.
