MAPPING OF MORPHOLOGICAL EVOLUTION OF TERRESTRIAL COASTAL ECOSYSTEMS (INCLUDING THE BACKSHORE ZONE)

Paolo Villaa, Federica Bragab

a Institute for Electromagnetic Sensing of the Environment, National Research Council of Italy (CNR-IREA), Milan (Italy)

b Institute of Marine Sciences, National Research Council of Italy (CNR-ISMAR), Venice (Italy)

Research theme aims

The target of this research activity is to explore and evaluate the potential of Sentinel-2 data for assessing the status and evolution of coastal vegetation as the primary indicator of ecosystem conditions. In addition, a preliminary assessment of hyperspectral data collected by PRISMA was carried out, focusing on spectral reflectance quality and capabilities in discriminating different vegetation communities. For deriving satellite-based maps of coastal and wetland vegetation communities we adopted a rule-based classification approach modelled on the one implemented by Villa et al. (2015) [1], which exploits spectral features derived from annual time series of Sentinel-2 data.

Materials and Methods

Study area

Venice Lagoon is the largest lagoon in Italy, covering an area of around 550 km2, with average depth around 1 m. It is characterized by a semidiurnal tidal regime with an average value of ± 0.7 m. The lagoon consists of a complex mosaic of different vegetation, depending mainly on water salinity and freshwater input. The salt marshes are dominated by different halophytic species (e.g. Spartina maritima), while marginal freshwater sectors are dominated by Phragmites australis, with Juncus maritimus as dominant species at intermediate conditions. Coastal dunes and areas along the shoreline of the Adriatic Sea are mainly populated by herbaceous species, with some small patches dominated by Pinus pinea.

Spatial distribution information included into habitat maps of Venice lagoon (salt marshes and coastal vegetation) following Natura 2000 nomenclature (covering a temporal range from 2007 to 2010) was used to compile a reference dataset of vegetation community types. Habitat types were first grouped into 11 classes (Level 2), which were then aggregated into 6 higher level classes (Level 1). For training and validating the satellite based maps, we first randomly sampled 1000 points (10×10 m pixels, consistent with Sentinel-2 resolution) for each Level 2 class, and then we checked the 1000 points against vegetation conditions in 2016 and 2017, corresponding to the timeframe of Sentinel-2 time series, by excluding points not covered by natural vegetation. Finally, we aggregated classes at Level 1 divided the whole reference set into subsets to be used for different classification tests: test A) composed by training set (2/3 of points of 2016 set, for each L1 class), validation set (2/3 of points of 2016 set, for each L1 class), and transferability test set (all points of 2017 set); and test B) composed by training set (2/3 of points of 2016 and 2017 merged sets, for each L1 class), and validation set (1/3 of points of 2016 and 2017 merged sets, for each L1 class).

Vegetation community maps from Sentinel-2 and PRISMA

A time series of Sentinel-2 (S2A) data for the years 2016 and 2017, with cloud cover under 50% on Venice lagoon area has been gathered and processed (22 dates for 2016 and 21 dates for 2017). Sentinel-2 dataset was converted to surface reflectance after atmospheric effect correction with SEN2COR [2]. Starting from Sentinel-2 surface reflectance dataset, two spectral indices sensitive to vegetation features have been derived: the Water Adjusted Vegetation Index (WAVI), developed specifically to maximize the sensitivity to the density and biomass of aquatic vegetation [3], and the Normalized Difference Flood Index (NDFI), providing information about soil moisture and flooding conditions of vegetated areas [4]. The annual time series of WAVI and NDFI were divided into three different periods, following an approach developed by Villa et al. (2015) [1] for freshwater macrophytes classification, namely: i) early spring period, centred on April (DOY 85-125); ii) full summer period, ranging from mid-July to late August (DOY 190-250); iii late autumn period, ranging from mid-October to mid-November (DOY 280-325). Synoptic seasonal features over the three periods and the whole year (DOY 1-366) were computed from combining Sentinel-2 data falling within the temporal range covered by each period specified (min, max, mean and std). As a result, four synoptic features for each period considered were calculated, bringing to a total of 16 features for WAVI and 16 features for NDFI. In addition to those 32 features, surface reflectance response at vegetation peak of season conditions was added to input feature set (derived from the Sentinel-2 scene acquired on 27/08/2016 and 02/08/2017). The vegetation community classification, following the approach developed in Villa et al. (2015) [1], is based on a hierarchical set of cascade rules structured in a binary tree. The algorithm is run using the J48 routine, modelled on Quinlan (1996) [5] and included into WEKA data mining suite [6].

In absence of scenes acquired during the commissioning phase (up to October 2019) over Venice lagoon, the preliminary assessment of PRISMA data capabilities for mapping coastal vegetation was performed over a different study area hosting similar vegetation communities, located along the western Sardinia coast, near the town of Arborea, south of Oristano city. Near-simultaneous scenes of the area were acquired by S2A and by PRISMA on 25/08/2019. S2A was gathered as level 2A, surface reflectance data, and PRISMA scene was expressed as TOA radiance and converted to surface reflectance after atmospheric effect correction performed with ATCOR code [7]. PRISMA spectral bands were resampled at Sentinel-2 MSI spectral resolution, and surface reflectance spectra of the two scenes were compared to provide an overview of the spectral quality and reliability of PRISMA data. The capabilities provided by PRISMA in discriminating different vegetation communities were assessed by calculating the separability of each pair of selected coastal and wetland vegetation communities drawn from ISPRA habitat map of the area.

Results and Discussion

The vegetation community classification algorithm was run using Sentinel-2 derived features as input, i.e. WAVI and NDFI, as well as surface reflectance at peak of season conditions, under two different classification tests: test A, i.e. validating the results separately for 2016 (independent validation) and 2017 (temporal transferability test), and test B, i.e. merging 2016 and 2017 samples. The 2016 and 2017 (test B, fig.1) maps were produced only for natural vegetation, water and wetland areas of Venice lagoon, while all the other land cover classes were excluded using the CORINE Land Cover map.

Figure 1: Vegetation community type maps derived from S2 time series using the integrated 2016-2017 data (test B), representing vegetation cover in Venice lagoon: a) 2016 season; b) 2017 season.

The comparison between accuracy assessment results derived from the independent 2016 validation set and the 2017 (test A) shows a generally good accuracy for the 2016, with OA=80.6% and Kappa=0.771 calculated over the independent validation set. When the rules implemented starting from 2016 training set are applied to the same features derived from 2017 Sentinel-2 data, overall accuracy decreases to 65.5%. This suggest that inter-annual variations in input features must be considered if temporal transferability of the method is targeted. The temporal inconsistency highlighted from test A outcomes was tackled by running the classification experiment under test B conditions (Tab.1). The accuracy of the vegetation community type maps produced for 2016 and 2017 are generally good and highly consistent, i.e. with OA=78.9% (Kappa=0.751) and OA=79.1% (Kappa=0.754) for 2016 and 2017, respectively, and differences in per-class accuracies between the years lower than 0.10. Even if performance over 2016 is slightly lower than in test A conditions, test B provided more robust results, that bear higher chances of reliability if applied in different seasons.

Table 1: Confusion matrix (number of 10×10 m pixels) of the vegetation community type maps derived from S2 time series using the test B scheme, showing overall and per-class accuracy metrics calculated on the independent validation sets for 2016 (upper panel) and 2017 (lower panel).

The spectral reflectance comparison between S2 and PRISMA shows that spectra derived from PRISMA after atmospheric correction with ATCOR are generally darker by around 10-20 % than S2 derived surface reflectance data, but this reflectance underestimation is apparently consistent across the visible to shortwave infrared range. Regarding the potential performance of PRISMA in discriminating different vegetation communities, the high spectral content of PRISMA bands provides a notable boost in separability of coastal and wetland vegetation types (based on ISPRA habitat map legend) clearly highlighted in Table 2.

Table 2: Separability matrix showing J-Mdist scores based on spectral reflectance derived from PRISMA scene of 25/08/2019 over Arborea study area for selected coastal and wetland vegetation communities (from ISPRA habitat map of Sardinia). Lower left triangle shows the J-Mdist scores calculated using PRISMA spectral bands, and upper right triangle shows the J-Mdist scores calculated resampling PRISMA data to S2 bands.

Future perspectives

The results of this research activity demonstrated that spectral information derived from S2 time series can be effectively used for assessing the status of coastal and wetland vegetation as primary indicator of ecosystem conditions. Vegetation community type maps derived for the years 2016 and 2017 have generally provided a reliable picture of Venice lagoon vegetation, with an overall accuracy around 80%. A preliminary test of PRISMA capabilities of our vegetation targets revealed the potential of hyperspectral data in enhancing spectral discrimination capabilities of different coastal and wetland vegetation communities compared to multispectral ones, thus opening concrete possibilities of improvement for the proposed approach in the near future.

Publications and Presentations

  • Villa, P; Braga, F (2020). Mapping coastal and wetland vegetation communities using multi-temporal satellite data. EUROLAG 9 conference, 20-24 January 2020, Venice, IT.
  • Villa, P; Giardino, C; Mantovani, S; Tapete, D; Vecoli, A; Braga, F (2021). Mapping Coastal and Wetland Vegetation Communities using Multi-temporal Sentinel-2 data. XXIVth ISPRS Congress, 04-10 July 2021. Accepted into conference program.

References

  1. Villa, P., Bresciani, M., Bolpagni, R., Pinardi, M., Giardino, C. (2015). A rule-based approach for mapping macrophyte communities using multi-temporal aquatic vegetation indices. Remote Sensing of Environment, 171, 218-233.
  2. Louis, J., Debaecker, V., Pflug, B., Main-Korn, M., Bieniarz, J., Mueller-Wilm, U., … Gascon, F. (2016). Sentinel-2 Sen2Cor: L2A Processor for Users. In Living Planet Symposium, 740, 91.
  3. Villa, P., Mousivand, A., Bresciani, M. (2014). Aquatic vegetation indices assessment through radiative transfer modeling and linear mixture simulation. International Journal of Applied Earth Observation and Geoinformation, 30, 113-127.
  4. Boschetti, M., Nutini, F., Manfron, G., Brivio, P. A., Nelson, A. (2014). Comparative analysis of normalised difference spectral indices derived from MODIS for detecting surface water in flooded rice cropping systems. PloS one, 9(2), e88741.
  5. Quinlan, J. R. (1996). Learning decision tree classifiers. ACM Computing Surveys (CSUR), 28(1), 71-72.
  6. Witten, I. H., Frank, E., Hall, M. A., Pal, C. J. (2016). Data Mining: Practical machine learning tools and techniques. Morgan Kaufmann.
  7. Richter R. and Schläpfer, D. (2019). “Atmospheric / Topographic Correction for Satellite Imagery”, DLR report DLR-IB 565-01/2019, Wessling, Germany, 210.