• The nNBR vary considerably between Landsat 8 and Sentinel 2 within the study areas.• The agreement or disagreement in the classification is not related to any particular burn
severity class.
• The difference in biogeographical regions and the vegetation types might be a major factor ofdisagreement.
Material and methods
Comparison of forest fire severity mappping using Landsat 8 and Sentinel 2 around the Mediterranean basin
Mustapha Lateb1,2, Chariton Kalaitzidis1 and Ioannis Gitas3
1Mediterranean Agronomic Institute of Chania (MAICh), Greece2Mouloud Mammeri University based Tizi-Ouzou (UMMTO), Algeria
3Aristotle University of Thessaloniki, Greece
UMMTO
Sumforest Conference “Bridging research, policy and practice for sustainable forest management” (17-18 October 2017, Barcelona, Spain)
IntroductionMapping wild fire effect on the forest ecosystem using Landsat image is an advanced topic. Manyauthors have used different combination of bands to measure the fire severity and assess theforest resilience in broad scale. The revolution comes with the Normalized Burn Ratio (NBR). Itwas used first time the US land management agencies to derive severity maps. Having the sameformula with the normalized difference vegetation index (NDVI), the NBR uses mid-infrared (MIR)band in place of the red band. It integrates the two Landsat TM/ETM+ bands (4 and 7) thatrespond most, but in opposite ways to burning. Many improvements were made to the NBR bydifferent authors; Key and Benson (2006) subtracted a post-fire NBR from a pre-fire NBR andcreated the differenced NBR (dNBR). It allows mapping the absolute vegetation change. The
different sensitivity of the near-infrared (NIR) and MIR Landsat bands to pre and post-firevegetation state showed high correlation with field data. The launch of Landsat 8 (L8) andSentinel 2 (S2) satellites offer the opportunity for the continuity of dNBR use. Both satellitesshow spectral and radiometric differences that might affect the NBR values.
AimThe aim of this study is to test the performance of NBR for forest fire severity mapping using theLandsat 8 and Sentinel 2 sensors, when assessing forest fire severity in different biogeographicalregions and vegetation types.
Mediterranean sea
Class 1 Class 2 Class 3 Class 4 Class 5 Class 6 Class 7 Sum
Class 1 0 0 3 4 5 2 0 14
Class 2 0 0 1 0 0 0 0 1
Class 3 27 1104 23015 1214 73 14 1 25448
Class 4 4 68 3733 12830 3379 190 9 20213
Class 5 1 3 163 3242 12754 2678 19 18860
Class 6 0 0 5 67 1458 6726 690 8946
Class 7 0 0 2 4 9 156 679 850
Sum 32 1175 26922 17361 17678 9766 1398 74332
dNBR Sentinel 2
dN
BR
Lan
dsat8
Confusion matrix, Greece
Conclusion
Results
Figure2: Sentinel 2, ESA
Figure3: Landsat8, NASA
𝑵𝑩𝑹𝑺𝒆𝒏𝒕𝒊𝒏𝒆𝒍 𝟐 =
𝑩𝟖 − 𝑩𝟏𝟐
𝑩𝟖 + 𝑩𝟏𝟐
𝑵𝑩𝑹𝑳𝒂𝒏𝒅𝒔𝒂𝒕 𝟖 =
𝑩𝟓 − 𝑩𝟕
𝑩𝟓 + 𝑩𝟕
References and Image credits• Key CH, Benson NC (2006) ‘Landscape assessment: sampling and analysis methods.’ USDA Forest Service, Rocky Mountain Research Station General Technical Report RMRS-GTR-164-CD. (Ogden, UT)• Figure 1: https://landsat.gsfc.nasa.gov/wp-content/uploads/2015/06/Landsat.v.Sentinel-2.png• Figure 2: https://cnes.fr/fr/le-monde-agricole-face-la-revolution-sentinel-2• Figure 3 : https://www.nasa.gov/
Study sites
For each study site, pre-fire and post-fire images are selected, NBR is computed for on each of them, so preNBR andpostNBR are obtained. Based on the latter, dNBR is computed. dNBR of both Sentinel2 and Landsat8 are compared.
𝒅𝑵𝑩𝑹 =
preNBR - posNBR
Seven (7) forest fires from seven different Mediterranean countries occurred in summer 2016 have been selected from European Forest Fire Information System (EFFIS) using “Size Class = Major” ascriteria since the NBR is designed for big forest fire events.
Figure 1
Figure 4 Figure 5