Search

Classical statistical methods prove advantageous for small datasets, whereas machine learning algorithms can excel with larger datasets. Our paper challenges this conventional wisdom by addressing a highly significant problem: the identification of burned areas through satellite imagery, that is a clear example of imbalanced data. The methods are illustrated in the North-Central Portugal and the North-West of Spain in October 2017 within a multi-temporal setting of satellite imagery. Daily satellite images are taken from Moderate Resolution Imaging Spectroradiometer (MODIS) products. Our analysis shows that a classical Logistic regression (LR) model competes on par, if not surpasses, a widely employed machine learning algorithm called the extreme gradient boosting algorithm (XGBoost) within this particular domain. [ABSTRACT FROM AUTHOR]

Fire is one of the natural agents with the greatest impact on the terrestrial ecosystem and plays an important ecological role in a large part of the terrestrial surface. Remote sensing is an important technique applied in mapping and monitoring changes in forest landscapes affected by fires. This study presents a spectral separability analysis for the detection of burned areas using Landsat-8 OLI/TIRS images in the context of fires that occurred in different biomes of Brazil (dry ecosystem) and Portugal (temperate forest). The research is based on a fusion of spectral indices and automatic classification algorithms scientifically proven to be effective with as little human interaction as possible. The separability index (M) and the Reed–Xiaoli automatic anomaly detection classifier (RXD) allowed the evaluation of the spectral separability and the thematic accuracy of the burned areas for the different spectral indices tested (Burn Area Index (BAI), Normalized Burn Ratio (NBR), Mid-Infrared Burn Index (MIRBI), Normalized Burn Ratio 2 (NBR2), Normalized Burned Index (NBI), and Normalized Burn Ratio Thermal (NBRT)). The analysis parameters were based on spatial dispersion with validation data, commission error (CE), omission error (OE), and the Sørensen–Dice coefficient (DC). The results indicated that the indices based exclusively on the SWIR1 and SWIR2 bands showed a high degree of separability and were more suitable for detecting burned areas, although it was observed that the characteristics of the soil affected the performance of the indices. The classification method based on bitemporal anomalous changes using the RXD anomaly proved to be effective in increasing the burned area in terms of temporal alteration and performing unsupervised detection without relying on the ground truth. On the other hand, the main limitations of RXD were observed in non-abrupt changes, which is very common in fires with low spectral signal, especially in the context of using Landsat-8 images with a 16-day revisit period. The results obtained in this work were able to provide critical information for fire mapping algorithms and for an accurate post-fire spatial estimation in dry ecosystems and temperate forests. The study presents a new comparative approach to classify burned areas in dry ecosystems and temperate forests with the least possible human interference, thus helping investigations when there is little available data on fires in addition to favoring a reduction in fieldwork and gross errors in the classification of burned areas. [ABSTRACT FROM AUTHOR]

The study of areas affected by unplanned forest fires is crucial for identifying and developing prevention procedures, as well as for mitigating their effects. Forest fires that have occurred in Portugal in recent years have devastated numerous hectares of the territory, with serious consequences for the environment, resulting in changes in land cover and soil degradation. The Guarda district was one of the affected regions. Therefore, with the aim of analysing the burned areas, the methodology used to estimate the areas affected by the wildfires in August 2022, in the Guarda district, is presented. The purpose of this study is to obtain geospatial, qualitative and quantitative information about the burned areas and relate it to land cover and the severity with which the event marked the territory. To achieve this, Remote Sensing techniques were applied in conjunction with Geographic Information Systems, specifically supervised classification using the Support Vector Machine (SVM) algorithm. The main inputs consisted of spectral indices by normalized difference calculated from multispectral images from the Sentinel-2 satellite. Calculating the difference between spectral indices from different periods (pre-fire and post-fire) allowed for isolating the event within a specific timeframe. Thus, it was possible to record and analyse the areas affected by the fires during the period between August 2nd (pre-fire) and August 22nd (post-fire) of 2022. The results were subsequently intersected with the Conjectural Land Cover Map of 2021 (COSc) from Direção Geral do Território (DGT), which enabled the association of the three severity levels (Low, Medium, and High) of the burned areas with different land covers, resulting in statistical analysis of the affected areas and the respective thematic map. [ABSTRACT FROM AUTHOR]

• We modelled vegetation and soil moisture content with Sentinel-1 and 2 satellite data. • The hypothesis of a universal landscape model for VMC and SMC was not supported. • Vegetation type models showed promise for VMC in shrubland conditions. • The combination of Sentinel optical and radar data improved model performance. • The relationships between satellite predictors and VMC and SMC varied with time. The satellite monitoring of vegetation moisture content (VMC) and soil moisture content (SMC) in Southern European Atlantic mountains remains poorly understood but is a fundamental tool to better manage landscape moisture dynamics under climate change. In the Atlantic humid mountains of Portugal, we investigated an empirical model incorporating satellite (Sentinel-1 radar, S1; Sentinel-2 optical, S2) and ancillary predictors (topography and vegetation cover type) to monitor VMC (%) and SMC (%). Predictors derived from the S1 (VV, HH and VV/HH) and S2 (NDVI and NDMI) are compared to field measurements of VMC (n = 48) and SMC (n = 48) obtained during the early, mid and end of summer. Linear regression modelling was applied to uncover the feasibility of a landscape model for VMC and SMC, the role of vegetation type models (i.e. native forest, grasslands and shrubland) to enhance predictive capacity and the seasonal variation in the relationships between satellite predictors and VMC and SMC. Results revealed a significant but weak relationship between VMC and predictors at landscape level (R2 = 0.30, RMSEcv = 69.9 %) with S2_NDMI and vegetation cover type being the only significant predictors. The relationship improves in vegetation type models for grasslands (R2 = 0.35, RMSEcv = 95.0 % with S2_NDVI) and shrublands conditions (R2 = 0.52, RMSEcv = 45.3 %). A model incorporating S2_NDVI and S1_VV explained 52 % of the variation in VMC in shrublands. The relationship between SMC and satellite predictors at the landscape level was also weak, with only the S2_NDMI and vegetation cover type exhibiting a significant relationship (R2 = 0.28, RMSEcv = 18.9 %). Vegetation type models found significant associations with SMC only in shrublands (R2 = 0.31, RMSEcv = 9.03 %) based on the S2_NDMI and S1_VV/VH ratio. The seasonal analysis revealed however that predictors associated to VMC and SMC may vary over the summer. The relationships with VMC were stronger in the early summer (R2 = 0.31, RMSEcv = 90.1 %; based on S2_NDMI) and mid (R2 = 0.37, RMSEcv = 70.8 %; based on S2_NDVI), butnon-significant in the end of summer. Similar pattern was found for SMC, where the link with predictors decreases from the early summer (R2 = 0.33, RMSEcv = 16.0 %; based on S1_VH) and mid summer (R2 = 0.30, RMSEcv = 17.8 %; based on S2_NDMI) to the end of summer (non-significant). Overall, the hypothesis of a universal landscape model for VMC and SMC was not fully supported. Vegetation type models showed promise, particularly for VMC in shrubland conditions. Sentinel optical and radar data were the most significant predictors in all models, despite the inclusion of ancillary predictors. S2_NDVI, S2_NDMI, S1_VV and S1_VV/VH ratio were the most relevant predictors for VMC and, to a lesser extent, SMC. Future research should quantify misregistration effects using plot vs. moving window values for the satellite predictors, consider meteorological control factors, and enhance sampling to overcome a main limitation of our study, small sample size. [ABSTRACT FROM AUTHOR]

Highlights • Several indicators of wildfire disturbance from MODIS time series were compared. • No single indicator was best for detection in time and space simultaneously. • Combining LST and TCT features ranked best for yearly fire mapping of burned area. • Combining TCT features ranked best for estimating date of occurrence. • A multi-indicator approach allows to select best indicators to improve available databases. Abstract Wildfires constitute an important threat to human lives and livelihoods worldwide, as well as a major ecological disturbance. However, available wildfire databases often provide incomplete or inaccurate information, namely regarding the timing and extension of fire events. In this study, we described a generic framework to compare, rank and combine multiple remotely-sensed indicators of wildfire disturbances, in order to not only select the best indicators for each specific case, as well as to provide multi-indicator consensus approaches that can be used to detect wildfire disturbances in space and time. For this end, we compared the performance of different remotely-sensed variables to discriminate burned areas, by applying a simple change-point analysis procedure on time-series of MODIS imagery for the northern half of Portugal, without external information (e.g. active fire maps). Overall, our results highlight the importance of adopting a multi-indicator consensus approach for mapping and detecting wildfire disturbances at a regional scale, that allows to profit from spectral indices capturing different aspects of the Earth's surface, and derived from distinct regions of the electromagnetic spectrum. Finally, we argue that the framework here described can be used: (i) in a wide variety of geographical and environmental contexts; (ii) to support the identification of the best possible remotely-sensed functional indicators of wildfire disturbance; and (iii) for improving and complementing incomplete wildfire databases. [ABSTRACT FROM AUTHOR]