Appraisal of the Sentinel-1 & 2 use in a large-scale wildfire assessment: A case study from Portugal's fires of 2017

• Main Authors: Brown, A.R (Author), Ferentinos, K.P (Author), Petropoulos, G.P (Author)
• Format: Article
• Language: English
• Published: Elsevier Ltd 2018
• Subjects: agricultural land; Agriculture; artificial neural network; assessment method; Burn severity; Burnt area mapping; data set; Earth observation; Forestry; forestry modeling; GIS; mapping method; Maximum likelihood; Portugal; Revised Universal Soil Loss Equation; RUSLE; Sentinel; Sentinel-1; Sentinel-2; Soil erodibility; Support vector machines; wildfire
• Online Access: View Fulltext in Publisher

 The recent launch of Sentinel missions offers a unique opportunity to assess the impacts of wildfires at higher spatial and spectral resolution provided by those Earth Observing (EO) systems. Herein, an assessment of the Sentinel-1 & 2 to map burnt areas has been conducted. Initially the use of Sentinel-2 solely was explored, and then in combination with Sentinel-1 and derived radiometric indices. As a case study, the large wildfire occurred in Pedrógão Grande, Portugal in 2017 was used. Burnt area estimates from the European Forest Fires Information System (EFFIS) were used as reference. Burnt area was delineated using the Maximum Likelihood (ML) and Support Vector Machines (SVMs) classifiers, and two multi-index methods. Following this, burn severity was assessed using SVMs and Artificial Neural Networks (ANNs), again for both standalone Sentinel-2 data and in combination with Sentinel-1 and spectral indices. Soil erosion predictions were evaluated using the Revised Universal Soil Loss Equation (RUSLE) model. The effect of the land cover derived from CORINE operational product was also evaluated across the burnt area and severity maps. SVMs produced the most accurate burnt area map, resulting a 94.8% overall accuracy and a Kappa coefficient of 0.90. SVMs also achieved the highest accuracy in burn severity levels estimation, with an overall accuracy of 77.9% and a kappa of 0.710. From an algorithmic perspective, implementation of the techniques applied herein, is based on EO imagery analysis provided nowadays globally at no cost. It is also robust and adaptable, being potentially integrated with other high EO data available. All in all, our study contributes to the understanding of Mediterranean landscape dynamics and corroborates the usefulness of Sentinels data in wildfire studies. © 2018 Elsevier Ltd