Vegetation fuel characterization using remote sensing data over Southern Portugal

Abstract

Portugal will be warmer and drier under future scenario projections linked to climate change, favouring more extreme wildfire events. Fire has a worldwide scale with a critical role in water and carbon cycles. For this reason, it is essential to know better and understand the vegetation dynamic and its role in the Earth system. Remote sensing can be helpful for better comprehension, once it is able to cover large areas with good temporal consistency. In such a context, the study aims to improve the representation of fuel load and moisture content from satellite data for use in fire propagation models. In this study, three above-ground biomass (AGB) datasets are used: first, samples collected by “Instituto da Conservação da Natureza e das Florestas” (ICNF) in 2015 for the Portuguese National Forest Inventory; second, AGB derived from ~3.000 trees in-situ dendrometric variables measurements (total height, tree diameter at 1.30 m above the ground) collected in the Herdade da Mitra at the University of Evora for 2020 and 2021; and third, AGB derived from eucalyptus trees on a field site in Serra de Ossa between 2016 and 2021 provided by the Navigator company. Otherwise, for live fuel moisture content (LFMC), biweekly samples over two field sites (Herdade da Mitra and Serra de Ossa) were collected during the period between April and October 2022, counting almost 250 samples. These samples combined with satellite data information derived from Sentinel-2 (spectral bands and spectral indexes) were used to build a model using a machine learning approach, such as Random Forest (RF) classifier, considering more than 30 variables to predict the AGB and LFMC. Results showed reasonable agreement between predicted and observed values, with r2 and RSME values of 0.56 (0.74) and 17 ton ha-1 (7%) for AGB (LFMC). Finally, the RF model was used to generate wall-to-wall AGB and LFMC maps. The study suggests that remote sensing data combined with a machine learning approach may produce information to characterize vegetation conditions for wildfire risk assessment in Portugal.