The indicator addresses trends in land surface productivity derived from remote sensing-observed time series of vegetation indices. The vegetation index used in the indicator is the Plant Phenology Index (PPI). The PPI is based on the MODIS Nadir BRDF-Adjusted Reflectance product (MODIS MCD43 NBAR). The product provides reflectance data for the MODIS ‘land’ bands (1-7) adjusted using a bi-directional reflectance distribution function. This function models values as if they were collected from a nadir-view, to remove so-called cross-track illumination effects. The PPI is a new vegetation index optimised for efficient monitoring of vegetation phenology. It is derived from radiative transfer solution using reflectance in visible-red and near-infrared (NIR) spectral domains. The PPI has a linear relationship with the canopy green leaf area index (LAI) and its temporal pattern is highly similar to the temporal pattern of gross primary productivity (GPP) estimated by flux towers at ground reference stations. The PPI is less affected by the presence of snow than commonly used vegetation indices, such as the Normalized Difference Vegetation Index (NDVI) or the Enhanced Vegetation Index (EVI).
The product is distributed with 500-m-pixel size (MODIS Sinusoidal Grid) with an 8-day compositing period.
Methodology for indicator calculation
The PPI time series is affected by noise due to, for example, atmosphere and remaining cloud influence, resulting in some spikes and outlier values. Since large spikes and outliers might significantly affect further function fitting, they first have to be removed from the data. This is done in an initial filtering process, further described in the Timesat software manual.
After outlier removal, the next step in the analysis is the determination of the number of growing seasons. This is based on a harmonic function fit (sine-cosine functions) to the data. The presence of a second season is established by evaluating the amplitudes of the first and second components of the harmonic fit. Presence of noise in the data complicates the decision on whether the given secondary maximum represents a true growing season or not. Therefore, an amplitude threshold is used to remove seasons that are smaller than the given threshold. A detailed description of the determination of the number of growing seasons is found in the Timesat software manual.
After the number of growing seasons has been determined, double logistic functions are fitted to the data from each pixel. This is done to generate smooth continuous functions that describe each individual growing season well. It is assumed that most of the noise included in the PPI (or any other vegetation index) results in negative bias of the values. Therefore, iterative adaptation of the logistic functions to the upper envelope of the data is applied in the following step. The function fit is performed on the PPI data. Values less than the first function fit are then considered influenced by noise and thus less important, so their weights are decreased for the next iteration of the function fitting.
Phenological metrics (and other parameters describing the character of the given growing season) are finally extracted from the fitted function data. The following parameters are extracted for each growing season detected to determine productivity:
- Start of season (SOS): date of the start of the season defined as the date when the PPI has increased to the 20% level of the average annual PPI amplitude (Jin et al., 2017). The average annual PPI amplitude is the difference between the average peak level and the average base level for each pixel.
- End of season (EOS): date of the end of the season defined as the date when the PPI drops under the 20% level of the average annual PPI amplitude .
- Large integral: integral of the fitted function between the start and end of the season.
- Small integral: integral of the differences between the fitted function and the base level from start to end of the season.
Seasonal amplitude is calculated as the difference between the fitted curve maximum and the base level. The SOS and EOS points on the curve are then given as the fraction of the amplitude, i.e. the date when the fitted curve reaches/drops below the defined percentage of the seasonal amplitude. For this indicator, a level of 20% of the seasonal PPI amplitude was used as the SOS and EOS detection thresholds.
The output of the process is a productivity metric for each year of the time series 2000-2016 (17 years) covering the EEA-38 territory (the 27 EU Member States plus Iceland, Lichtenstein, Norway, Switzerland and Turkey, and six collaborating countries, Albania, Bosnia and Herzegovina, North Macedonia, Serbia and Kosovo) and the United Kingdom. The spatial resolution of the productivity data set is at a pixel size of 500 m × 500 m. To address changes in productivity, a linear regression was fit to the productivity time series of each grid cell of the data set. As the PPI, and consequently also productivity, is a dimensionless measure, the change was expressed as the R value of the linear regression model instead of the slope of the fitted model. The R value, i.e. the coefficient of determination, is expressed between -1 and 1. It shows how close the data are to the fitted regression line. In general, the higher the value the better the model fits the data. The advantage of the R value is that unlike the slope of the linear regression model the R value can be compared across bioclimatic regions and between various ecosystems. The change was also expressed as the relative growth. The relative growth is expressed in percentage and is calculated as the change between the first and the last year of the time series in proportion of the first year’s productivity value.
A detailed description of the methodology for calculating the productivity metric can be found in the Timesat software manual.
Methodology for gap filling
Addressing ecosystem services and their complex interactions calls for a coherent approach to understanding the coupled human-environment system.
In November 2013, the European Parliament and the Council adopted the 7th EAP, ‘Living well, within the limits of our planet’. The degradation of ecosystems is recognised as one of the major threats to the provision of ecosystem services, biodiversity and Europe’s resilience to climate change and natural disasters within the 7th EAP, in priority objective 1, paragraph 23. Priority objective 5, paragraph 66, states that environmental monitoring is one of the cornerstones of the Union’s environmental policy, and, within priority objective 5, paragraph 71, the mapping and assessment of ecosystem services are recognised as a necessary basis for developing the most appropriate responses to environmental change.
The 7th EAP is intended to help guide EU action on the environment and climate change up to and beyond 2020. It highlights that ‘Action to mitigate and adapt to climate change will increase the resilience of the Union’s economy and society, while stimulating innovation and protecting the Union’s natural resources’. Consequently, several priority objectives of the 7th EAP refer to climate change adaptation.
In February 2021, the European Commission presented the EU adaptation strategy package. The new strategy sets out how the European Union can adapt to the unavoidable impacts of climate change and become climate resilient by 2050. One of the objectives of the strategy is to ensure better-informed decision-making, which will be achieved by bridging knowledge gaps and further developing the European climate adaptation platform (Climate-ADAPT) as the ‘one-stop shop’ for adaptation information in Europe. Climate-ADAPT has been developed jointly by the European Commission and the EEA to share knowledge on (1) observed and projected climate change and its impacts on environmental and social systems and on human health, (2) relevant research, (3) EU, transnational, national and sub-national adaptation strategies and plans, and (4) adaptation case studies.
In May 2020, the EU adopted a Biodiversity Strategy to 2030, related to protecting and restoring nature. The strategy states that ‘the biodiversity crisis and the climate crisis are intrinsically linked. Climate change accelerates the destruction of the natural world through droughts, flooding and wildfires, while the loss and unsustainable use of nature are in turn key drivers of climate change’. Droughts are negatively affecting agricultural ecosystems and food security, the resilience of forest ecosystems and the ability of green urban spaces to protect people against heat waves. In particular, the impacts of extended droughts on ecosystems need to be assessed because they can lead to significant loss of vegetation productivity and irreversible damage to the condition of ecosystems and can lead to land degradation.
One area of uncertainty related to the choice of the fitting function in Timesat. In the scientific literature, several fitting methods have been used and proposed. In this analyses, logistic functions were chosen since they are well founded in the scientific literature , and in a recent study have been found to be one of the most robust methods for regional phenology estimation.
Another source of uncertainty is the detection of the SOS and EOS points on the seasonal vegetation profile. To address appropriate levels for SOS and EOS, we analysed estimates of GPP (gross primary productivity) from ground-measured data from carbon flux towers from the international Fluxnet network. This was done to evaluate if there was any significant difference in the SOS and EOS estimated from these measurements between different land cover classes. The analysis did not indicate any clear separability between the classes. This was because of high variability in the GPP data, and hence there was no basis for making individual choices for different land cover classes or climate zones. Therefore, a fixed threshold of 20% of the annual PPI amplitude was used in the indicator assessment process. This level was chosen based on the analyses in Jin et al. (2017). It cannot be ruled out that the use of a single threshold across Europe may have introduced some uncertainty.
Data set uncertainty
The data set represents several plant functional types aggregated with pixels of 500 m × 500 m. Therefore, the data set can be used at only the ecosystem level, indicating productivity changes of main plant functional types. As opposed to filed measurement, remote-sensing products measure vegetation light absorption from a satellite at heights of several hundred kilometres, which might introduce bias due to atmospheric disturbances.
No uncertainty has been specified.