Corresponding author: Michele Di Musciano ( michele.dimusciano@graduate.univaq.it ) Academic editor: Daniele Viciani
© 2020 Walter De Simone, Michele Di Musciano, Valter Di Cecco, Giorgia Ferella, Anna Rita Frattaroli.
This is an open access article distributed under the terms of the Creative Commons Attribution License (CC BY 4.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Citation:
De Simone W, Di Musciano M, Di Cecco V, Ferella G, Frattaroli AR (2020) The potentiality of Sentinel-2 to assess the effect of fire events on Mediterranean mountain vegetation. Plant Sociology 57(1): 11-22. https://doi.org/10.3897/pls2020571/02
|
Wildfires are currently one of the most important environmental problems, as they cause disturbance in ecosystems generating environmental, economic and social costs. The Sentinel-2 from Copernicus Program (Sentinel satellites) offers a great tool for post-fire monitoring. The main objective of this study is to evaluate the potential of Sentinel-2 in a peculiar mountainous landscape by measuring and identifying the burned areas and monitor the short-term response of the vegetation in different ‘burn severity’ classes. A Sentinel-2 dataset was created, and pre-processing operations were performed. Relativized Burn Ratio (RBR) was calculated to identify ‘burn scar’ and discriminate the ‘burn severity’ classes. A two-year monitoring was carried out with areas identified based on different severity classes, using Normalized Difference Vegetation Index (NDVI) to investigate the short-term vegetation dynamics of the burned habitats; habitats refer to Annex I of the European Directive 92/43/EEC. The study area is located in ‘Campo Imperatore’ within the Gran Sasso – Monti della Laga National Park (central Italy). The first important result was the identification and quantification of the area affected by fire. The RBR allowed us to identify even the less damaged habitats with high accuracy. The survey highlighted the importance of these Open-source tools for qualitative and quantitative evaluation of fires and the short-term assessment of vegetation recovery dynamics. The information gathered by this type of monitoring can be used by decision-makers both for emergency management and for possible environmental restoration of the burned areas.
burn severity classes, NDVI, post-fire, satellite remote sensing, short-term vegetation monitoring
Wildfires are currently one of the most important environmental problems as they cause disturbances in ecosystems generating environmental, economic and social costs (
Remote sensing has been used to monitor active fires and burned areas at the global and national scale (
Their daily revisit cycle is useful to capture active fire signatures or burn scars, however, the relatively coarse spatial resolution causes underestimations in case of burned area small extent. Comparisons between different burned area products generally show high variation in results (
Satellite remote sensing (SRS) is ideal for monitoring burned areas, especially for large and remote places. Among the available satellites, the Copernicus Program (Sentinel satellites) offers a great tool for post-fire monitoring, both for the great spatial and temporal resolution, and for the accessibility of the data completely open source (
For burned areas detecting, optical satellite data from Multi Spectral Instrument (MSI) on Sentinel-2 (S2), starting from 2015, have a potential five-day temporal resolution (when the two satellites are operational) and have a spatial resolution of 10 m. S2 carries a multispectral sensor with 13 bands, from 0.443 to 2.190 μm. The visible RGB and the NIR bands are available at a 10 m spatial resolution, highly suitable for application in vegetation canopies. Four red-edge bands at 20 m spatial resolution are also available and are particularly suited for chlorophyll content analysis and to parametrize ecophysiological largescale models (
In the European context, the year 2017 was a record season for fires, especially in Italy for the amounts of hectares burned (
Fire is a frequent disturbance and a dominant factor in the evolution and ecology of Mediterranean areas (
Nevertheless, these fires affected in some cases montane and subalpine regions where the plant communities are not adapted to this phenomenon, leading to net changes in the affected natural communities (
In this paper we use the S2 satellite imagery, whose data can be freely downloaded from the Open Access Hub of the European Space Agency (ESA) for burned area monitoring purposes, as well as for searching approaches to study the effects of fire on vegetation (
The aims of this study are i) to evaluate the potential of S2 to detect and quantify the burn scars and ii) to monitor the short-term vegetation dynamics in different ‘burn severity’ classes.
This approach was tested in the southern sector of the Gran Sasso massif (Abruzzo, central Italy) where wildfire events recently occurred. Moreover, to assess the short-term dynamics of vegetation recovery in the different habitat types (with reference to the Annex I of the European Directive 92/43/EEC;
The study area is ‘Fonte della Vetica’ (Centroid coordinates: 42°24'56.44"N; 13°45'30.07"E), in the ‘Campo Imperatore’ upland plain falling within the Gran Sasso – Monti della Laga National Park in central Italy (Figure
The study area was chosen for the importance of its floristic (
An accidental fire triggered by anthropogenic causes, which occurred between 5 and 9 August 2017, has devastated several important plant communities and left huge burned areas which represent the object of the present study.
We analysed six images from S2 satellite (Spatial resolution: 10 m, Radiometric resolution: 12 bit) acquired by ESA's Open Access (https://scihub.copernicus.eu), with a time interval between July ÷ October 2017 and 2018, listed in the Appendix: Table
In order to build the dataset, the aforementioned Level 1C (TOA – Top of Atmosphere reflectance) images were chosen based on the low cloud cover percentage and pre-processed for atmospheric correction (
The detection of burned areas was carried out by Relativized Burn Ratio index (RBR) (
The RBR was chosen because of its reliability in situations where the pre-combustion biomass is low or very variable and heterogeneous (
The RBR is a relativized version of the Delta Normalized Burn Ratio (dNBR: NBRprefire – NBRpostfire) (
Relativized Burn Ratio (RBR) for the study area. The index has been discretized in three ‘burn severity’ classes (Moderate-Low, Moderate-High and High severity). The validation points (VIIRS) are marked in yellow, while the point where the 2017 fire was triggered is indicated with green star.
After the identification of the study area, rasters were manipulated in GIS (QGIS 3.4.5,
Annex I Habitats burned by the fire of 2017 in the study area. The numerical codes refer to the nomenclature used to identify the Annex I habitats (http://vnr.unipg.it/habitat/, see also Table
The NDVI was discretized into 10 classes with values ranging from 0 (0.0 to 0.1) to 10 (0.9 to 1.0), with 10 representing the highest photosynthetic activity (
Resulting NDVI raster maps were processed through zonal statistics for each type of Annex I habitats affected within each severity class of the RBR (
To make the analysis comparable with an undisturbed situation, we built a 500 m buffer around the burned area. The control area ("Unburned control area") was defined based on the presence of the same Annex I habitats identified in the burned area. Furthermore, the buffer was built based on the occurrence of similar topographical conditions. The approach used in this study is shown in the workflow in Figure
The area affected by the fire was identified by the Relativized Burn Ratio index. The RF model used to validate the RBR index showed a high degree of classification accuracy with an Overall accuracy of 83% and a Kappa of 77% (Appendix: Table
EU Annex I codes of the habitat types present in the study area, with relative description (with asterisk the priority habitats;
HABITAT CODE | DESCRIPTION |
3220 | Alpine rivers and the herbaceous vegetation along their banks |
4060 | Alpine and Boreal heaths |
6170 | Alpine and subalpine calcareous grasslands |
6210* | Semi-natural dry grasslands and scrubland facies on calcareous substrates (Festuco-Brometalia) (*important orchid sites) |
6230* | Species-rich Nardus grasslands, on siliceous substrates in mountain areas and submountain areas, in Continental Europe |
6430 | Hydrophilous tall herb fringe communities of plains and of the montane to alpine levels |
8210 | Calcareous rocky slopes with chasmophytic vegetation |
9210* | Apennine beech forests with Taxus and Ilex |
Zonal statistics for each Annex I habitat affected by fire within each severity class of the RBR (Relativized Burn Ratio). The NDVI values was discretized into 10 classes with values ranging from 0 (0.0 to 0.1) to 10 (0.9 to 1), with 10 representing the highest photosynthetic activity. The numerical codes of the habitats correspond to: Alpine rivers and the herbaceous vegetation along their banks (3220), Alpine and Boreal heaths (4060), Alpine and subalpine calcareous grasslands (6170), Semi-natural dry grasslands and scrubland facies on calcareous substrates (Festuco-Brometalia) (*important orchid sites) (6210*), Species-rich Nardus grasslands, on siliceous substrates in mountain areas and submountain areas, in Continental Europe (6230*), Hydrophilous tall herb fringe commu-nities of plains and of the montane to alpine levels (6430), Calcareous rocky slopes with chasmophytic vegetation (8210), Apennine beech forests with Taxus and Ilex (9210*).
NDVI AVERAGE VALUES WITHIN HABITAT NATURA 2000 | JULY 2017 | ||||||||||
BURN SEVERITY CLASS (RBR) | 3220 | 4060 | 6170 | 6210 | 6230 | 6430 | 8210 | 9210 | Coniferous stands | Total | |
Moderate-low severity [1] | 5.20 | 5.95 | 6.98 | 5.32 | 7.82 | 5.06 | 4.97 | 8.27 | 7.16 | 6.42 | |
Moderate-high severity [2] | 5.00 | 6.54 | 6.90 | 5.16 | 7.44 | 5.29 | 6.09 | 7.89 | 7.42 | 6.57 | |
High severity [3] | 7.32 | 7.39 | 6.06 | 7.50 | 6.67 | 8.00 | 8.86 | 7.40 | |||
Total Burned | 5.19 | 6.35 | 7.01 | 5.26 | 7.67 | 5.12 | 5.39 | 8.16 | 7.33 | 6.55 | |
Total Unburned (buffer 500m) | 4.00 | 5.30 | 6.61 | 5.23 | 7.06 | 5.10 | 4.38 | 7.90 | 7.56 | 6.17 | |
NDVI AVERAGE VALUES WITHIN HABITAT NATURA 2000 | AUGUST 2017 | ||||||||||
BURN SEVERITY CLASS (RBR) | 3220 | 4060 | 6170 | 6210 | 6230 | 6430 | 8210 | 9210 | Coniferous stands | Total | |
Moderate-low severity [1] | 3.38 | 3.38 | 4.16 | 3.42 | 4.27 | 3.40 | 2.85 | 4.66 | 4.62 | 3.87 | |
Moderate-high severity [2] | 3.00 | 2.64 | 3.07 | 3.24 | 3.23 | 3.18 | 2.40 | 3.56 | 3.56 | 3.05 | |
High severity [3] | 1.75 | 2.09 | 1.69 | 2.00 | 1.50 | 1.50 | 3.21 | 2.00 | |||
Total Burned | 3.33 | 2.88 | 3.40 | 3.32 | 3.77 | 3.32 | 2.69 | 4.31 | 4.15 | 3.39 | |
Total Unburned (buffer 500m) | 3.17 | 4.97 | 6.09 | 4.34 | 6.52 | 3.20 | 4.36 | 7.63 | 7.02 | 6.16 | |
NDVI AVERAGE VALUES WITHIN HABITAT NATURA 2000 | OCTOBER 2017 | ||||||||||
BURN SEVERITY CLASS (RBR) | 3220 | 4060 | 6170 | 6210 | 6230 | 6430 | 8210 | 9210 | Coniferous stands | Total | |
Moderate-low severity [1] | 4.46 | 3.69 | 5.14 | 5.66 | 5.12 | 6.85 | 3.22 | 4.47 | 4.79 | 4.62 | |
Moderate-high severity [2] | 5.00 | 3.34 | 4.63 | 5.45 | 4.69 | 6.83 | 2.91 | 3.94 | 3.58 | 4.11 | |
High severity [3] | 3.53 | 4.19 | 4.00 | 5.00 | 3.50 | 4.75 | 3.43 | 3.84 | |||
Total Burned | 4.48 | 3.52 | 4.77 | 5.55 | 4.98 | 6.85 | 3.12 | 4.33 | 4.26 | 4.34 | |
Total Unburned (buffer 500m) | 4.09 | 4.90 | 6.03 | 5.27 | 5.74 | 5.06 | 4.57 | 6.59 | 7.12 | 5.85 | |
NDVI AVERAGE VALUES WITHIN HABITAT NATURA 2000 | JULY 2018 | ||||||||||
BURN SEVERITY CLASS (RBR) | 3220 | 4060 | 6170 | 6210 | 6230 | 6430 | 8210 | 9210 | Coniferous stands | Total | |
Moderate-low severity [1] | 6.17 | 5.05 | 7.15 | 7.38 | 7.69 | 7.60 | 4.04 | 6.16 | 5.21 | 5.98 | |
Moderate-high severity [2] | 6.00 | 5.10 | 6.20 | 7.34 | 7.54 | 7.50 | 3.84 | 5.50 | 4.55 | 5.67 | |
High severity [3] | 5.73 | 6.20 | 7.18 | 8.50 | 4.00 | 6.50 | 3.80 | 5.96 | |||
Total Burned | 6.15 | 5.16 | 6.50 | 7.35 | 7.65 | 7.57 | 3.97 | 5.99 | 4.86 | 5.84 | |
Total Unburned (buffer 500m) | 5.67 | 5.78 | 7.37 | 6.86 | 7.48 | 6.18 | 5.08 | 8.24 | 6.54 | 6.71 | |
NDVI AVERAGE VALUES WITHIN HABITAT NATURA 2000 | AUGUST 2018 | ||||||||||
BURN SEVERITY CLASS (RBR) | 3220 | 4060 | 6170 | 6210 | 6230 | 6430 | 8210 | 9210 | Coniferous stands | Total | |
Moderate-low severity [1] | 5.70 | 5.14 | 6.85 | 6.48 | 7.30 | 5.81 | 4.37 | 6.17 | 5.28 | 5.80 | |
Moderate-high severity [2] | 6.00 | 5.12 | 6.00 | 6.54 | 6.94 | 5.38 | 4.31 | 5.56 | 4.84 | 5.52 | |
High severity [3] | 5.94 | 6.26 | 6.86 | 8.00 | 5.00 | 6.75 | 4.47 | 6.05 | |||
Total Burned | 5.75 | 5.23 | 6.36 | 6.52 | 7.19 | 5.67 | 4.36 | 6.02 | 5.06 | 5.70 | |
Total Unburned (buffer 500m) | 5.56 | 5.60 | 6.77 | 6.07 | 7.00 | 5.32 | 5.14 | 6.96 | 6.46 | 6.06 | |
NDVI AVERAGE VALUES WITHIN HABITAT NATURA 2000 | OCTOBER 2018 | ||||||||||
BURN SEVERITY CLASS (RBR) | 3220 | 4060 | 6170 | 6210 | 6230 | 6430 | 8210 | 9210 | Coniferous stands | Total | |
Moderate-low severity [1] | 5.48 | 5.06 | 6.67 | 6.54 | 6.86 | 6.33 | 4.28 | 6.45 | 5.49 | 5.80 | |
Moderate-high severity [2] | 6.50 | 5.08 | 5.85 | 6.52 | 6.82 | 6.17 | 4.28 | 5.85 | 5.03 | 5.48 | |
High severity [3] | 5.86 | 6.13 | 6.67 | 8.00 | 4.33 | 6.25 | 4.76 | 5.95 | |||
Total Burned | 5.57 | 5.16 | 6.22 | 6.54 | 6.87 | 6.29 | 4.28 | 6.28 | 5.27 | 5.67 | |
Total Unburned (buffer 500m) | 5.22 | 5.53 | 6.71 | 6.15 | 7.13 | 5.75 | 4.94 | 8.61 | 6.85 | 6.55 |
The habitats more threatened by the wildfire resulted to be the grasslands at low altitude (⁓1,500 meters) covering 45.2% of the total area, the shrublands at medium altitude (⁓1,600 meters, 25.08%), the grasslands at high altitude (>1,700 meters, 18.78%), the coniferous stands (6.76%) and the beech forests (2.62%).
A strong trend resulted from the multi-temporal analysis of NDVI, above all in the grasslands vegetation.
The month of July 2017 (pre-fire) was taken as reference for a ‘no disturbed area’. The NDVI values resulted in an average of 6.65 for the grasslands (habitats 6170, 6210* and 6230*), 7.75 for the forests (habitat 9210* and Coniferous stands), and 6.35 for the shrublands (habitat 4060).
In the months following the fire event, the NDVI values were in clear decline, especially immediately after the event (August 2017), showing average values of 3.50 for the grasslands, 4.23 for the forests, and 2.88 for the shrublands.
Interestingly, in October 2017, only two months after the event, an increase in NDVI values was observed. In particular, the grasslands showed high average values (NDVI = 5.10), unlike the shrubland and forest habitats.
The monitoring for the year 2018 showed other interesting values. In particular, the grasslands at low and high elevation (habitats 6170*, 6210*, 6230*, 6430) revealed an average value of NDVI = 7.27 for the month of July, and ~ 6.50 for the months of August and October.
In the same year, the forest habitat and the conifer stands also showed increasing values. All these trends are shown in Figure
NDVI trends in the burned area and cumulative areas shared by habitats threatened by fire. (a) Monthly comparison in two years of monitoring (July 2017 and July 2018). (b) Comparison of NDVI discrete maps for the monitored months (August 2017, October 2017, August 2018 and October 2018). (c) NDVI average trends in all the monitored months, within the ‘burn severity’ classes (M-L= Moderate-Low severity; M-H = Moderate-High severity; H = High severity). The different colors of the cumulative columns represent the Annex I habitats threatened by fire in the study area.
The NDVI values (Table
NDVI trends of the total burned and unburned area, and within ‘burn severity’ classes, in the threatened habitats. ‘Burn severity’ classes: M-L = Moderate-Low Severity (orange solid line); M-H = Moderate-High Severity (red dash line); H = High Severity (violet dotted line). (a) NDVI average trends of unburned areas in a 500 meters buffer around the burned scar (green solid line), and ‘burn severity’ classes (M-L, M-H, H) for the total threatened area. The other graphs represent the NDVI average trends within the ‘burn severity’ classes for different Annex I habitats plus coniferous stands: (b) 6210* (Semi-natural dry grasslands and scrubland facies on calcareous substrates), (c) 4060 (Alpine and boreal heaths), (d) 6170 (Alpine and subalpine calcareous grasslands), (e) coniferous stands, (f) 9210* (Apennine beech forests with Taxus and Ilex). The habitats 6230* [Species-rich Nardus grasslands, on siliceous substrates in mountain areas (and sub-mountain areas, in Continental Europe)], 6430 (Hydrophilous tall herb fringe communities of plains and of the montane to alpine levels) and 8210 (Calcareous rocky slopes with chasmophytic vegetation) are not included because of the limited extent of their patches.
Fire is an important disturbance process in many ecosystems, e.g. Mediterranean ecosystems (
The study area was well suited for testing the potentiality of the S2 data to obtain the specific spectral indices used for vegetation monitoring (RBR and NDVI). Moreover, it allowed the investigation of the short-term dynamics of vegetation recovery, even verifying in the field the satellite-collected data (pers. obs.).
This study identified and characterized the burned areas at medium-high resolution, taking advantage of the spectral and temporal characteristics of the MSI S2 data. The large temporal resolution of these data (5 days) allowed a very precise use of the RBR index, as described above, based on pre- and post-fire imagery. This index, calculated at the spatial resolutions we used, allowed the identification of burn scars with great rigor. It also enabled us to discretize the burned area into severity classes and allowed to carry out both an exploratory analysis of the most affected areas and a more thorough investigation using the NDVI.
The analysis conducted within the ‘severity classes’ using the NDVI showed a clear pattern of similarity of the less severe ones, which showed relatively regular trends in values. On the other hand, a faster increase was found in the higher class, confirming a greater resilience at highly burned sites (
Results show that the most affected habitats are the grasslands (6170, 6210, 6230, 6430) and shrublands (4060). As far as the former are concerned, the analysis lead to interesting results, showing how the herbaceous vegetation, even in montane and sub-alpine environments, can rapidly recover, similarly to the Mediterranean communities. This might also (partly) be due to a rapid vegetation recovery in the first two years after the fire, becoming more gradual in the following years (
However, extreme events, such as fires, can have very important effects on the structure and composition of grasslands (
Shrublands were damaged for a quarter of the total area. These communities are mainly constituted by common juniper (Juniperus communis L.). Our results are in line with
The coniferous forests in the study area are characterized by mixed reforestations with European silver fir (Abies alba Mill.), European spruce [Picea abies (L.) H. Karst.], European larch (Larix decidua Mill.) and Austrian pine (Pinus nigra J.F.Arnold). They represent less than 10% of the entire study area. The scarcity of active maintenance over the years made these woods very susceptible to fire, given the quantity of combustible material in the undergrowth (
The last habitat affected by the fire of 2017 was the beech forest (9210), representing less than 3% of the burn scar. Beech forests are not very resilient to the passage of the fire, because of the low seed vitality after such events. Interestingly, a positive trend was observed in the NDVI index values, accounting for a good regenerative capacity, due to the development of the herbaceous undergrowth species present in the seed bank. In fact, the fertile soil of beech woods can facilitate the development of herbaceous vegetation (and therefore an increase in its photosynthesis index) in the period following the fire (
Satellite remotely sensed data and analyses have been widely applied in both conservation science and practice, but there are limitations to the information they can provide. In this study, the spatial resolution of S2 sensors allowed the identification of short-term trends in the photosynthetic activity of the different plant communities after the fire event, and a very accurate study of the vegetational dynamics in progress. On the other hand, in order to obtain more accurate information regarding individual populations or single species, there is a need for satellite data with a very high spatial resolution (VHR, usually commercial) that have a relatively high cost, depending on the purpose and the request (
The field sampling effort for monitoring an area of this size requires considerable monetary and human resources. The use of SRS not only reduces the field work but also facilitates possible actions to restore the area interested (
The use of satellite remote sensing for vegetation monitoring has allowed to observe the situation from a different perspective, compared to the classic field surveys. Through this methodology, it was possible to carry out a high resolution multitemporal analysis.
This study provided a rapid and effective monitoring of the conditions of various plant communities, including a quantification of damage and tendency to short-term recover. Furthermore, with the applied methodology it was possible to detect and quantify different burn severity classes, allowing an even more extensive study of the vegetation recovery dynamics. This approach can also make field work more efficient by focusing sampling efforts on certain areas, in order to implement specific environmental recovery measures for damaged habitats at a relatively low cost.
The proposed approach provides useful management information for fire prevention in protected areas, planning monitoring activities and implementing functional rehabilitation actions for habitats affected by these disturbing events.
This study was supported by the Department of Life Health and Environmental Sciences of University of L’Aquila (Abruzzo – Italy). The authors thank the European Space Agency for making data (and software) publicly available, and the NASA Fire Information for Resource Management System for provisioning the georeferenced data used in this study (https://firms.modaps.eosdis.nasa.gov/) for validating the burned scar.
Satellite images used for the analysis.
Satellite Instrument | Acquisition date | Product types | Pre/Post Fire |
Sentinel-2 MSI | 20.07.2017 | Level-1C | PRE |
Sentinel-2 MSI | 29.08.2017 | Level-1C | POST |
Sentinel-2 MSI | 28.10.2017 | Level-1C | POST |
Sentinel-2 MSI | 10.07.2018 | Level-1C | POST |
Sentinel-2 MSI | 29.08.2018 | Level-1C | POST |
Sentinel-2 MSI | 13.10.2018 | Level-1C | POST |
Planet Scope MSI | 10.08.2017 | Level 1B | POST |
Accuracy statistics. Classification accuracy (%) for the index classification when applied to the validation dataset. Training data consisted of a random sample (80%) of the full dataset and validation was conducted on the remaining data.
Overall Statistics | ||||
Accuracy: 0.8333 | Kappa: 0.7778 | |||
95% CI: (0.3588, 0.9958) | P-Value [Acc > NIR]: 0.01783 | |||
No Information Rate: 0.3333 | ||||
Statistics by Class: | ||||
Class: 1 | Class: 2 | Class: 3 | Class: 4 | |
Sensitivity | 0.5000 | 10.000 | 10.000 | 10.000 |
Specificity | 10.000 | 10.000 | 10.000 | 0.8000 |
Pos Pred Value | 10.000 | 10.000 | 10.000 | 0.5000 |
Neg Pred Value | 0.8000 | 10.000 | 10.000 | 10.000 |
Prevalence | 0.3333 | 0.3333 | 0.1667 | 0.1667 |
Detection Rate | 0.1667 | 0.3333 | 0.1667 | 0.1667 |
Detection Prevalence | 0.1667 | 0.3333 | 0.1667 | 0.3333 |
Balanced Accuracy | 0.7500 | 10.000 | 10.000 | 0.9000 |