Contribution of free satellite time-series images to mapping plant communities in the Mediterranean Natura 2000 site: the example of Biguglia Pond in Corse (France)

Abstract

Mapping plant communities, which is essential to assess the conservation status of natural habitats, is currently based mainly on time-consuming field surveys without the use of satellite data. However, free image time-series with high spatial and temporal resolution have been available since 2015. This study assessed the contribution of Sentinel-2 time-series images to mapping the spatial distribution of 18 plant communities within a Natura 2000 site (1978 ha) located on the Mediterranean biogeographical region (Corsica, France). The method was based on random forest modeling of six Sentinel-2 images acquired from 26 February to 24 October 2017, which were calibrated and validated using a field vegetation map. The results showed that the 18 plant communities were modeled correctly, with 72% overall accuracy. The uncertainty map associated with the model indicated areas that required additional field observations.

Functional Analysis for Habitat Mapping in a Special Area of Conservation Using Sentinel-2 Time-Series Data

“…Therefore, these activities are useful for achieving the goals set out in European environmental policies (e.g., Directive 92/43/CEE) [7][8][9][10]. Due to the high heterogeneity, fragmentation, and diversity of vegetation, the mapping and monitoring of Mediterranean plant associations and habitats are particularly challenging, when relying only on in situ observations [10][11][12].…”

“…This challenge can be supported by remote-sensing data, which, combined with field plots, enables the accurate mapping of plant communities and habitats [13][14][15][16]. The processing of remotely sensed data provides a better understanding of the conservation status, which is crucial in helping decision makers to plan and evaluate the impacts of management practices [9,[17][18][19][20].…”

“…Habitat (92/43/EEC) and plant association maps (using the Braun-Blanquet approach) rarely reach an overall accuracy greater than 80% [17]. In most cases, the number of classes is up to five [11,14,37,[77][78][79][80], while an increase in the number of classes typically has a negative impact on the overall performance, reaching values close to 75%, as in [9,16,[80][81][82][83][84].…”

“…In the study area, the high temporal resolution of the Sentinel-2 mission makes mapping (through the functional approach) of the vegetation and habitats up-to-date and repeatable, according to the timing required by the directive (six years), as well as with higher frequencies (e.g., every 2-3 years) [9]. Cyclical mapping (at least every six years, in accordance with the Habitats Directive) of the plant associations-and, therefore, of the habitats-with high accuracy (in terms of OA) could provide a valid tool for monitoring the transformation of habitats over time [10,90].…”

Combined use of environmental and spectral variables with vegetation archives for large-scale modeling of grassland habitats

“…Second, spatial grid size (250 × 250 m 2 ) and comprehensiveness have been improved compared with kilometer-scale presence/absence maps derived from spatially incomplete vegetation archives (Carli et al, 2020; Chytrý et al, 2020). In addition, spatialization of model uncertainty (Figure 5) provides a critical interpretation of the map, because areas of low certainty may correspond to ecological gradients (Raab et al, 2018) or rare or non-grassland habitats that were excluded from the classification (Rapinel et al, 2020). However, this national map of EUNIS level 2 habitats has two disadvantages.…”