This paper investigates the reliability of free and open-source algorithms used in the geographical object-based image classification (GEOBIA) of very high resolution (VHR) imagery surveyed by unmanned aerial vehicles (UAVs). UAV surveys were carried out in a cork oak woodland located in central Portugal at two different periods of the year (spring and summer). Segmentation and classification algorithms were implemented in the Orfeo ToolBox (OTB) configured in the QGIS environment for the GEOBIA process. Image segmentation was carried out using the Large-Scale Mean-Shift (LSMS) algorithm, while classification was performed by the means of two supervised classifiers, random forest (RF) and support vector machines (SVM), both of which are based on a machine learning approach. The original, informative content of the surveyed imagery, consisting of three radiometric bands (red, green, and NIR), was combined to obtain the normalized difference vegetation index (NDVI) and the digital surface model (DSM). The adopted methodology resulted in a classification with higher accuracy that is suitable for a structurally complex Mediterranean forest ecosystem such as cork oak woodlands, which are characterized by the presence of shrubs and herbs in the understory as well as tree shadows. To improve segmentation, which significantly affects the subsequent classification phase, several tests were performed using different values of the range radius and minimum region size parameters. Moreover, the consistent selection of training polygons proved to be critical to improving the results of both the RF and SVM classifiers. For both spring and summer imagery, the validation of the obtained results shows a very high accuracy level for both the SVM and RF classifiers, with kappa coefficient values ranging from 0.928 to 0.973 for RF and from 0.847 to 0.935 for SVM. Furthermore, the land cover class with the highest accuracy for both classifiers and for both flights was cork oak, which occupies the largest part of the study area. This study shows the reliability of fixed-wing UAV imagery for forest monitoring. The study also evidences the importance of planning UAV flights at solar noon to significantly reduce the shadows of trees in the obtained imagery, which is critical for classifying open forest ecosystems such as cork oak woodlands.

This paper aims to develop a supervised classification integrating synthetic aperture radar (SAR) Sentinel-1 (S1) and optical Sentinel-2 (S2) data for land use/land cover (LULC) mapping in a heterogeneous Mediterranean forest area. The time-series of each SAR and optical bands, three optical indices (normalized difference vegetation index, NDVI; normalized burn ratio, NBR; normalized difference red-edge index, NDRE), and two SAR indices (radar vegetation index, RVI; radar forest degradation index, RFDI), constituted the dataset. The coherence information from SAR interferometry (InSAR) analysis and three optical biophysical variables (leaf area index, LAI; fraction of green vegetation cover, fCOVER; fraction of absorbed photosynthetically active radiation, fAPAR) of the single final month of the time-series were added to exploit their correlation with the canopy structure and improve the classification. The random forests (RF) algorithm was used to train and classify the final dataset, and an exhaustive grid search analysis was applied to set the optimal hyperparameters. The overall accuracy reached an F-score_M of 90.33% and the integration of SAR improved it by 2.53% compared to that obtained using only optical data. The whole process was performed using freely available data and open-source software and libraries (SNAP, Google Earth Engine, Scikit-Learn) executed in Python-script language.

This paper investigates the capability of the free synthetic aperture radar (SAR) Sentinel-1 (S-1) C-band data for burned area mapping through unsupervised machine learning open-source processing solutions in the Mediterranean forest ecosystems. The study was carried out in two Mediterranean sites located in Portugal (PO) and Italy (IT). The entire processing workflow was developed in Python-based scripts. We analyzed two time-series covering about one month before and after the fire events and using both VH and VV polarizations for each study site. The speckle noise effects were reduced by performing a multitemporal filter and the backscatter time averages of pre- and post-fire datasets. The spectral contrast between changed and unchanged areas was enhanced by calculating two single-polarization radar indices: the radar burn difference (RBD) and the logarithmic radar burn ratio (LogRBR); and two temporal differences of dual-polarimetric indices: the delta modified radar vegetation index (ΔRVI) and the delta dual-polarization SAR vegetation index (ΔDPSVI), all exhibiting greater sensitivity to the backscatter changes. The scene’s contrast was enhanced by extracting the Gray Level Co-occurrence Matrix (GLCM) textures (dissimilarity, entropy, correlation, mean, and variance). A principal component analysis (PCA) was applied for reducing the number of the GLCM image layers. The burned area was delineated through unsupervised classification using the k-means clustering algorithm. A suitable number of clusters (k value) were set using a silhouette score analysis. To assess the accuracy of the resulting detected burned areas, an official burned area map based on multispectral Sentinel-2 (S-2) was used for PO, while for IT, a reference map was produced from S-2 data, based on the normalized burned ratio difference (ΔNBR) index. Recall (r), precision (p) and the F-score accuracy metrics were calculated. Our approach reached the values of 0.805 (p), 0.801 (r) and 0.803 (F-score) for PO, and 0.851 (p), 0.856 (r) and 0.853 (F-score) for IT. These results confirm the suitability of our approach, based on SAR S-1 data, for burned area mapping in heterogeneous Mediterranean ecosystems. Moreover, the implemented workflow, completely based on free and open-source software and data, offers high adaptation flexibility, repeatability, and custom improvement.

This study aimed to compare and assess different Geographic Object-Based Image Analysis (GEOBIA) and machine learning algorithms using unmanned aerial vehicles (UAVs) multispectral imagery. Two study sites were provided, a bergamot and an onion crop located in Calabria (Italy). The Large-Scale Mean-Shift (LSMS), integrated into the Orfeo ToolBox (OTB) suite, the Shepherd algorithm implemented in the Python Remote Sensing and Geographical Information Systems software Library (RSGISLib), and the Multi-Resolution Segmentation (MRS) algorithm implemented in eCognition, were tested. Four classification algorithms were assessed: K-Nearest Neighbour (KNN), Support Vector Machines (SVM), Random Forests (RF), and Normal Bayes (NB). The obtained segmentations were compared using geometric and non-geometric indices, while the classification results were compared in terms of overall, user’s and producer’s accuracy, and multi-class F-scoreM. The statistical significance of the classification accuracy outputs was assessed using McNemar’s test. The SVM and RF resulted as the most stable classifiers and less influenced by the software used and the scene’s characteristics, with OA values never lower than 81.0% and 91.20%. The NB algorithm obtained the highest OA in the Orchard-study site, using OTB and eCognition. NB performed in Scikit-learn results in the lower (73.80%). RF and SVM obtained an OA>90% in the Crop-study site.

Wildland fires are among the main factors affecting the surrounding territory in terms of ecological and socioeconomic changes at different temporal and spatial scales. In the Mediterranean environment, although fire can positively influence some bio-physical dynamics of habitats, it acts as a pressing disturbance on ecosystems when the severity, spatial scale, and/or frequency are high, thereby determining their degradation. Therefore, knowing and mapping the accurate quantitative spatial distribution of all areas affected by fire during an entire high-frequency fire season and on a relatively large scale (regional/national scale) is an essential step to initialize the numerous subsequent effect monitoring analyses that can be carried out. This work proposes a reliable and open-access workflow to map burned areas on regional and national scales during the entire fire season. To achieve this, we integrated optical (Sentinel-2, S2) and Synthetic Aperture Radar (SAR; Sentinel-1, S1) free high spatial and temporal resolution data into a multitemporal composite criterion. Open-source software and Python-based libraries were used to develop the workflow. In particular, the second-lowest near infra-red (NIR) image composite (secMinNIR) criterion, based on the retrieval of the second minimum values that the NIR values reached in each pixel during the entire time frame considered, was applied to cloud-free S2 imagery to optimize the separability between burned and unburned areas. Subsequently, a second temporal composite criterion was developed and applied to the S1 time series, relying on the SAR capacity to detect vegetation fire-induced structural and humidity changes. It was based on retrieving the S1 pixel value of the first next (or the same) date to the corresponding date of the pixel value previously found by secMinNIR. The burned area map was created using an object-based geographic analysis (GEOBIA) process, using two optical and SAR composite images as input layers. The large-scale mean-shift (LSMS) algorithm was employed to segment the image, while the random forest (RF) classifier was the machine-learning model used to perform supervised classification. GEOBIA-based burned area classification was also performed using only the optical composite. The resulting accuracy values were compared using the precision (p), recall (r), and F-score accuracy metrics. The classification achieved high accuracy levels (F-score value greater than 0.9) in both cases (S1+ S2, 0.956; S2, 0.914), highlighting the increased effectiveness of this approach in detecting burned areas, heterogeneous in terms of amplitude, and affected site-specific characteristics that occurred during the fire season. Although the use of only optical data is sufficient to map the fire-affected areas early, some commission errors, represented by small regions scattered over the entire study area, remain, proving that the integration of SAR data improves the quality of the obtained results.

In the present study, the temporal and spatial dynamics of the post-fire recovery of different Mediterranean vegetation types during the three years after the fire event were analyzed, according to different fire severity categories, integrating the use of Synthetic Aperture Satellite Radar (SAR) (Sentinel-1) and optical (Sentinel-2) image time series. The results showed that Mediterranean forest species and shrub/herbaceous species are adapted to fire, with high efficiency in restoring the vegetation cover. Differently, the ecological vulnerability of non-native eucalyptus plantations was found in a lower recovery trend during the observation period. The use of optical short-wave infrared (SWIR) and SAR C-band-based data revealed that some ecological characteristics, such as the woody biomass and structure, recovered at slower rates, comparing to those suggested by using near-infrared (NIR) and red-edge data. An optimized burn recovery ratio (BRR) was proposed to estimate and map the spatial distribution of the degree of vegetation recovery.