The unprecedented availability of spaceborne hyperspectral data has great potential to provide fire severity
estimates that align with post-fire management needs, overcoming complex logistics and data acquisition costs of
airborne hyperspectral sensors, and the suboptimal sensitivity of broadband data to several post-fire ground
components. We analyzed the feasibility of the PRISMA mission -one of the first spaceborne spectrometers
operationally active- to assess fire severity by leveraging hyperspectral data dimensionality through the retrieval
of sub-pixel components directly related to fire severity in the field. Multispectral data provided by Sentinel-2,
commonly used in fire severity quantitative assessments, were used as a benchmark method. Multiple endmember
spectral mixture analysis (MESMA) was used to retrieve fractional cover of char, photosynthetic
vegetation (PV), as well as non-photosynthetic vegetation and soil (NPVS) from post-fire PRISMA Level 2D and
Sentinel-2 Level 2A scenes in one of the largest wildfires ever recorded in the western Mediterranean Basin.
Ground-truth data were obtained using the Composite Burn Index (CBI) to procure three field-measured severity
metrics: vegetation, soil and site. The relationship between the CBI data on a continuum scale and the cover of
char, PV and NPVS image fractions retrieved from PRISMA and Sentinel-2 was assessed through Random Forest
regression (RFR). Likewise, Ordinal Forests (OF) algorithm was used to classify categorized CBI data (low,
moderate and high fire severity). PRISMA-based RFR fire severity estimates at vegetation, soil and site levels (R2
= 0.64–0.79 and RMSE = 0.33–0.41) outperformed those of Sentinel-2 (R2 = 0.27–0.53 and RMSE = 0.54–0.60),
and were in line with previous studies using airborne hyperspectral sensors at higher spatial resolution. Fire
severity underestimation for high field CBI values was almost unnoticeable in the PRISMA estimates. Categorical
fire severity, not currently estimated using hyperspectral data but with high interest in post-fire management,
were accurately predicted by PRISMA-based OF classification, with consistent user’s and producer’s accuracy for
each fire severity category. The high confusion between moderate and low/high fire severity categories, typical
when unmixing broadband multispectral data, was overcome by the PRISMA-based classification scheme. Our
results suggest that new spaceborne spectrometer missions can support reliable fire severity assessments
equivalent to airborne spectrometers, but readily applicable to large-scale assessments of extreme wildfire
events.
ETS Ingenierías Agrarias Universidad de Valladolid - Avd. Madrid s/n
34004 - PALENCIA - Localización
www5.uva.es/etsiiaa/
INIA-CIFOR - Ctra. A Coruña km 7,5
28040 - MADRID - Localización
www.inia.es
Si deseas estar informado, apúntate a nuestro boletín electrónico.