ESTIMATING PHENOLOGY METRICS FROM SENTINEL-2 TIME SERIES IN FOREST SITES

The Life MODERn NEC project has been created thank to the directive "NEC" (National Emission Ceiling, 2016/2284) of the European Union. The objective of the project is to monitor emissions of atmospheric pollutants (sulphur, nitrogen, organic compound, ammonia, and particulate matter) and to assess the impact on water and terrestrial ecosystems. Monitoring of terrestrial ecosystems is carried out by in situ sampling of indicators for air quality, atmospheric deposition, crown condition and phenology, ecosystem chemistry, ground vegetation, tree growth, meteorological variables, ozone injury, and soil solution. The project has identified six sites in Italy where in situ data are systematically collected. The six sites are part of the ICP forests (International Co-operative Programme on Assessment and Monitoring of Air Pollution Effects on Forests) level II plots, that comprises a total of 31 sites distributed over the Italian forests. In this work, all sites (31) were analysed thanks to the advantages offered by remote sensing technologies able to provide synoptic view over large territories. We exploited time series of Copernicus Sentinel-2 (S2) multi-spectral satellite images to estimate phenological metrics of the investigated sites. Phenological metrics are very important parameters to determine the health status of the forests and to identify changes induced by pollutants; specifically, we aim at pointing out changes in the timing and vigour of the plant's annual cycle. Monitoring vegetation phenology with field surveys can be time and manpower consuming because the operator has to visit sites several times during the year to collect data and observations. The use of remote sensing technologies could reduce the effort involved in field measurements and the synergy between remote and field data could increase the accuracy of the metrics. The satellite Sentinel-2 constellations provide multi-spectral images with high spatial resolution and short revisit time which is very important to observe the phenological phases in fast-changing environments. We focused on the 31 sites and used time series of spectral indices derived from Sentinel-2 imagery for the period 2016-2022. Since sites are distributed in all of Italy, downloading and processing all S2 tiles to extract time series of the indices could be quite a resource-intensive process. In order to reduce processing time, spectral indices were extracted from S2 images in Google Earth Engine (GEE). Every site is identified by a set of two coordinates of the central point of the plot that is object of field surveys and measurements in the LIFE MODERn (NEC) project. Methods to identify the phenological metrics (green-up, maturity, senescence and dormancy) are based on a double sigmoid function that was fitted to the time series of the daily vegetation index. From the sigmoid, the metrics were calculated using the derivatives of the curve. Processing is done using the R package "sen2rts" (Ranghetti, 2012). This package takes as input S2 time series, it reduces the noise that could be present in the time series and then fits a double logistic curve and extracts metrics. In the data preparation phase, the pixels under cloud or shadow condition have been removed based on quality layer of the S2 Level 2A product, and then the smoothing parameter has to be regulated based on the index and the annual oscillation. The first index we tested was NDVI which is widely recognized as a suitable indicators of the vegetative annual cycle of plants. Thanks to the red and NIR (near infrared) bands, it is possible to track the increment of biomass and photosynthesis activity, which can be translated into the phenological status of the plants. This is more evident for deciduous broadleaved vegetation, that is characterized by a clear intra-annual seasonality of NDVI; instead, seasonal cycles of evergreen vegetation might be more problematic to identify. These plants retain their leaves for more than one year and their photosynthetic activity is strictly regulated but chlorophyll is always present; therefore, the seasonal variability of NDVI values is less evident. The photosynthetic activity is regulated during the year by pigments like the carotenoids in the evergreens, so it could be tracked using an index specifically designed to be sensitive to them. The result of this work is a dataset of the phenological metrics (expressed as dates-DOY Day Of the Year) for the analysed forest sites. Results were evaluated based on available databases of phenology metrics and/or by photointerpretation. The estimated metrics and their change through the years, could be used to evaluate the difference in the annual cycle of plants eventually attributable to pollutants. The tested methodology and tools could be exploited to expand to the Italian level I sites of the ICP project. The level I plots are more than 250, so this remote sensing-based approach could allow an affordable and fast way to monitor a large number of sites through time. The processing time and memory usage on local machines are limited thanks to the download of NDVI time series from GEE, however, the double logistic fitting steps is still quite demanding and, for this reason, we foresee for future applications over large areas the full implementation of the process chain in GEE environment, to make feasible processing of a large number of points/sites thanks to cloud computing resources.