Spectrophenology from Sentinel 2 data: definition of reference curves for forest ecosystem mapping

Monitoring the spectral response of habitat types obtained by medium resolution satellite imagery can provide relevant information about the distribution and temporal dynamics of vegetation types through environmental gradients and geographic scales. The combination of the orbital cycle and bandwidth of Sentinel 2 sensor provides data every 5 days in mid-latitudes, which allows high-resolution temporal monitoring related to ecosystem phenology cycles and possible changes of habitats conservation status. In this context, the objective of this study was to obtain average spectrophenological curves for main tree and shrub formations present in Cantabria, as well as the analysis and characterization of phenological metrics that allow characterizing distributions and determining particular vegetation functions across landscape continuum. For the definition of the curves, we have processed all the historical data available from the MSI sensor of Sentinel 2 A and B satellites in order to generate a high temporal resolution time series of the NDVI index. Results have been temporarily averaged for a monthly representation after applying a manual filter of clouds and shadows that maximizes the quality of the data at the pixel level and a temporal smoothing of the series that filters out anomalous and missing values. NDVI values have been extracted at the pixel level for a sample of 230 points located with a subpixel GPS precision and whose floristic composition has been characterized by botanists in the field. Seasonal phenological curves for the years from 2017 to 2020 and annual average curves have been obtained, determining their variability and the degree of representativeness for each habitat type of interest. Finally, the equations of the functions adjusted to the average phenological curves by plant formation have been obtained by applying Fourier analyses. Results show particular features for each of the habitat type due to the existence of specific functional traits related to phenological variations between their average multiannual curves. These results highlight the interest of using spectrophenological time series of satellite data for defining reference curves of habitat types. These curves allow, on the one hand, identifying the spatial distribution as well as structural and function variations of different ecosystems, and on the other hand, having at hand an early detection system of anomalous phenological responses of vegetation communities that can be related to natural or anthropic disturbances in a context of Global Change.