2.1. Dataset

Three data sources were used for this research, they include: (i) temporal composites of MERIS MTCI; (ii) GlobCover 2009 land cover map; and (iii) a classification of the European biogeographical regions.

The MERIS MTCI is a ratio of the difference in reflectance between band 10 and band 9 and the difference in reflectance between band 9 and band 8 of the MERIS standard band setting. Bands 10, 9, and 8 correspond to wavelengths 753.75 nm, 708.75 nm, and 681.25 nm, respectively [21]. The MTCI is a surrogate of canopy chlorophyll content, and therefore provides a good estimation of photosynthetically active period of vegetation growth. We used composites of MERIS MTCI data at 1 km spatial resolution from 2002 to 2012. Composite periods were equal to eight days for the period from 2002–2007 and ten days for 2008–2012. This dataset was supplied by the European Space Agency and processed by Airbus Defence and Space. MTCI data were composited from the MERIS L2 product using an arithmetic mean and flux conservation resampling algorithm. The arithmetic mean is less sensitive to temporal biases and produces consistent images in both the spatial and temporal domains.

Land cover information was acquired from the Global Land Cover Classification (GlobCover2009) dataset at 300 m spatial resolution [43]. Currently, the GlobCover2009 dataset, derived from MERIS data, is the most detailed and recent global land cover map available with a hierarchical thematic legend consisting of 22 classes. Globcover2009 was selected for its greater consistency with MERIS MTCI time-series and its high geolocational accuracy (<150 m) [44].

Homogeneous biogeographical regions were used to stratify the European landscape at an aggregate level of ecological complexity. The map of biogeographical regions was based on the official delineations used in the Habitats Directive (92/43/EEC) and for the EMERALD Network set up under the Convention on the Conservation of European Wildlife and Natural Habitats (Bern Convention). This map categorizes Europe into 11 relatively homogeneous regions at a map scale of 1:1,000,000.

2.2. Phenology Estimates

The time-series MTCI data for every estimation period (i.e., year) used 1.5 years of data (from October in the previous year to the month of July in the next year) to estimate both the OG and EOS for every pixel. This length was used because the annual pattern spans across calendar years and hence complete information about LSP would not be obtained using single year data. The yearly values of OG and EOS were estimated for each image pixel of the study area using the methodology described in Dash, Jeganathan and Atkinson [18]. This methodology consists of two major procedures: data smoothing and extraction of phenological metrics.

The MERIS MTCI is a ratio of the difference in reflectance between band 10 and band 9 and the difference in reflectance between band 9 and band 8 of the MERIS standard band setting. Bands 10, 9 and 8 correspond to wavelengths 753.75 nm, 708.75 nm, and 681.25 nm, respectively. The MTCI time-series was smoothed using the Fourier method [45,46]. Although, there is a lack of general consensus regarding the robustness of smoothing methods and its performance can vary depending on data, Atkinson, et al. [47] showed that Fourier methods have a competitive performance when applied to MERIS MTCI data and require the specification of one parameter only, the number of harmonics for reconstruction. The first six Fourier components (mean + five harmonics) were used because they reproduced satisfactorily the original annual growth profile, obtaining a good fit between raw and smoothed MTCI time-series. Figure 1 shows some examples of raw MTCI and smoothed time-series as well as the extracted phenology metrics for different land cover types.

Figure 1. Inter-annual variation in MTCI values and phenology metrics for major natural vegetation types. (a) Broadleaved deciduous forests; (b) needleleaved evergreen forests; (c) grasslands; (d) shrublands.

Click here to enlarge figure

Figure 1. Inter-annual variation in MTCI values and phenology metrics for major natural vegetation types. (a) Broadleaved deciduous forests; (b) needleleaved evergreen forests; (c) grasslands; (d) shrublands.

The first derivative (difference between one composite value and its immediate preceding neighbour) of MTCI time-series was computed to locate the annual maximum. This peak was defined as a point on the phenological profile where the curvature changes the sign before and after a negative and positive trend of five successive composites, respectively. Next, the derived data were searched backward and forward departing at the maximum annual peak to estimate the OG and EOS, respectively. OG was defined as a valley at the beginning of the growing season point (a change in derivative value from positive to negative) and EOS was defined as a valley point occurring at the decaying end of a phenology cycle (a change in derivative value from negative to positive) [18]. The searching function stops when a valley peak greater than one-fifth of the annual maximum peak is reached. The composite number identified during the phenology estimation procedure was then converted to Julian days, assigning the date of the last day of the composite period. Figure 1 shows some examples of MTCI time-series for different land covers and the estimated phenology metrics.

The length of the season was equal to the number of days between OG and EOS. The phenological information over homogeneous pixels was grouped and analysed in terms of each natural vegetation land cover and different biogeographical regions. The statistical significance of the differences in OG and EOS between different biogeographical regions was assessed through a two-tailed t-test. A median phenological map was produced from the 10 years of time-series data to characterise the spatial variation of phenology across Europe. The 10-year time-series analysed overcomes some limitations in the satellite sensor data quality for a specific year and minimises the inter-annual variation in LSP metrics.

3.1. Spatial Variation in Phenological Metrics

Figure 2 shows the OG, EOS, and LS of the vegetation of the pan-European continent. OG dates are widely distributed from -10 days (approximately 21 December) to 150 days (30 May): later OG dates were observed at higher latitudes and vice versa. Earlier OG is associated mainly with rainfed croplands in the south of Europe and grasslands in Ireland and the southwest of England and Wales. On the other hand, later OG occurs in Fennoscandia and the north of Russia and in mountainous areas such as the Carpathians or the Caucasus. The progressive patterns in phenophases from low to high latitudes are attributed mainly to temperature variation [48], but also to the photoperiod and winter chilling requirements for some temperate tree species [5]. In terms of longitudinal variation, the OG of the Atlantic and Mediterranean basin regions is earlier than the OG of the Continental region. This is likely due to different vegetation responses to climatic conditions (i.e., temperature and precipitation) [39]. This large variability in phenophases could be explained not only by the difference in climatic conditions between coastal and interior continental regions, but also by the effect of changes in altitude associated with the mountain ridges.

The EOS dates range from 150 days (30 May) to 390 days (25 January), and their spatial distribution is generally the reverse of that of the OG dates for natural vegetation (Figure 2b). The later OG is generally coincident with a delayed EOS. However, this is only applicable to regions of natural or semi-natural vegetation, whereas cropland areas are characterised by earlier OG and EOS. Hence, regions with a later EOS exhibit longer length than regions with an earlier OG (Figure 2c). Note that there exist areas in Western Europe mainly associated with crops which present earlier dates for both OG and EOS.

Figure 2. Phenology maps of Europe. (a) Onset of greenness (OG); (b) end of senescence (EOS); (c) length of the season (LS).

Click here to enlarge figure

Figure 2. Phenology maps of Europe. (a) Onset of greenness (OG); (b) end of senescence (EOS); (c) length of the season (LS).

Figure 3. Standard deviation of phenology estimates. (a) Spatial distribution of different land cover types (DF); (b) onset of greenness (OG); (c) end of senescence (EOS).

Click here to enlarge figure

Figure 3. Standard deviation of phenology estimates. (a) Spatial distribution of different land cover types (DF); (b) onset of greenness (OG); (c) end of senescence (EOS).

Figure 4. Boxplots showing the distribution of OG, EOS, and LS for each vegetation type and biogeographical region (box, interquartile range; horizontal line, median; vertical lines, minimum and maximum). Box plots are shown ordinally from earliest to latest date and resume the information about the timing of phenophases from 2003 to 2011. Map shows the spatial distribution of the European biogeographical regions delineated in the Habitats Directive (92/43/EEC). Colours in the map correspond to the colours of boxplots. Alpine (Alp), Anatolian (Ana), Black Sea (B. Sea), Continental (Cont), Mediterranean (Med), Panonian (Pan), and Steppic (Step).

Click here to enlarge figure

Figure 4. Boxplots showing the distribution of OG, EOS, and LS for each vegetation type and biogeographical region (box, interquartile range; horizontal line, median; vertical lines, minimum and maximum). Box plots are shown ordinally from earliest to latest date and resume the information about the timing of phenophases from 2003 to 2011. Map shows the spatial distribution of the European biogeographical regions delineated in the Habitats Directive (92/43/EEC). Colours in the map correspond to the colours of boxplots. Alpine (Alp), Anatolian (Ana), Black Sea (B. Sea), Continental (Cont), Mediterranean (Med), Panonian (Pan), and Steppic (Step).

LS durations are given in Figure 2c. Shorter growing seasons (approximately three months) are associated with rainfed croplands in West Europe, and forests in boreal and mountainous areas. Maximum LS values (approximately one year) appear in the Atlantic basin associated with grasslands (see Figure 1d). Atzberger, Klisch, Mattiuzzi and Vuolo [41], Jeong, Ho, Gim and Brown [39], and Zhang, Tan and Yu [42], and Atzberger, Klisch, Mattiuzzi and Vuolo [41] found similar dates and patterns of variability for all phenophases. However, it should be noted that some differences can be attributed to the inclusion of crops in the present map.

The spatial distribution of spring phenophases (i.e., OG) is in agreement with the expected pattern of European classification of biogeographical regions and the GlobCover2009 land cover map. Interestingly, the spatial variability at a local scale (high frequency component) is determined by the type of land cover, whereas at a larger scale (low frequency component) the variability in phenophases matches the pattern of the biogeographical regions.

Figure 3 shows the standard deviation of LSP estimates over the 10-year period. The maximum deviation of EOS is larger than for OG in general terms (up to 90 days vs. 60 days, respectively). This could be due to either higher uncertainties in the estimates of autumn phenophases or a greater complexity of autumn phenology. All phenophases showed a similar spatial pattern in standard deviation which is driven mainly by land cover type. Rainfed crops in France and the United Kingdom experienced the greatest variability (mostly due to different crop types and management practices), followed also by crops in the Steppic biogeographic region (see Figure 4). Forests generally presented a very small variability, usually around five days. However, northern latitude forests presented a high variability in OG and a low variability in EOS. Although further research on this phenomenon in northern forests is required, it could be related to the photoperiod having a more important role in the senescence of northern forests (considering that the photoperiod is more stable in time than temperature).

3.2. Characterisation of the Phenology of the Main Biogeographical Regions in Europe

LSP varies greatly over broad geographic gradients, according to climate zone and vegetation type [5]. Figure 4 shows the boxplot computed from homogeneous pixels of the four major natural vegetation types for the different European biogeographical regions. In general, grasslands present the earliest OG values, starting around the middle of February, followed by shrublands for which OG starts from the first week of March, deciduous trees the second week of April and needle-leaved evergreen forests from the same date as deciduous trees until mid-May, depending on the biogeographic region. A similar pattern can be observed for the end of senescence. However, there is greater variability within biogeographical regions. EOS is reached first by the continental grasslands around the middle of July, followed by Mediterranean evergreen forests at the end of August, shrublands from the second week of November, deciduous forests from the middle of November, and finally, evergreen forests, which can last until the end of the year. Regarding LS, Atlantic grasslands present the longest growing season, which is equal to 302 days, followed by the southern evergreen forests with LS over 250 days, and the shrublands and evergreen forests of the remaining biogeographical regions which vary between 184 (Boreal) and 245 (Atlantic) days. Shrublands and deciduous forests have similar growing season lengths (around 220 days) with slight variation depending on the biogeographical region.

Different levels of response to environmental drivers can be observed according to the vegetation type. For broadleaved deciduous forests and shrublands, the phenological variations are relatively stable in all regions, whereas needle-leaved evergreen forests or grasslands present greater variability between biogeographical regions and phenophases. The variation of OG of deciduous trees is grouped into two categories: Alpine regions and northern latitudes, such as Atlantic, Continental, and Steppe regions starting during the first week of April; and the southern biogeographical regions (Black Sea, Mediterranean, and Panonian), which start four days later. EOS dates are more diverse and ranged from 16 November for Alpine, Steppe, and Continental regions until 30 November and 2 December for the Black Sea and Atlantic Regions, respectively. In terms of LS, it can be observed that deciduous trees of regions with higher values of EOS produce slightly longer growing seasons (Atlantic and Black Sea), equal to 240 and 232 days, respectively, perhaps due to the oceanic influence keeping the temperature warmer for a relatively longer period. The timing of grasslands was estimated for the Atlantic and Continental regions only, as large homogeneous sites were not found for the remaining biogeographical regions. Grasslands present similar OG dates in both regions. However, there are important discrepancies regarding the EOS and LS times. Continental grasslands have a median EOS date equal to 196 Julian days (15 July), more than four months earlier than Atlantic grasslands (26 December). This results in a LS equal to five-and-a-half months for Continental grasslands against a LS of almost a year for Atlantic grasslands (11.8 months). Additionally, the variance in estimated EOS for Continental grasslands was significantly greater than for other categories. This can be due to the wide longitudinal range of the Continental region, but also to greater difficulty in estimating the autumnal phenophases in temperate regions [5].

Needle-leaved evergreen forests presents a latitudinally controlled phenological pattern which is mediated by the influence of relief. OG starts first in the south, around 14 April for Mediterranean and Anatolian regions, whereas the rest of the biogeographical regions are from one-to-four weeks delayed according to a latitudinal gradient. For Alpine evergreen forests, OG starts later than Boreal evergreen forests, with OG dates equal to 16 May and 6 May, respectively. This difference can be due to the influence of Alpine Tundra on the estimation of dates for the Alpine region. However, the EOS for Boreal evergreen forests is earlier than for any other biogeographical region (4 November), resulting in the shortest LS (184 days). The EOS for Steppe evergreen forests occurs 12 days later than for Boreal forests, followed by the Continental and Atlantic regions. The EOS for the Mediterranean, Anatolian, Black Sea, and Alpine regions spans the start of the calendar year occurring in mid-January. The LS of the three first regions mentioned before are quite similar, with LS values of nine months (277, 269, and 277 days, respectively). However, the length of the season in Alpine evergreen forests is shorter due to a much-delayed OG (7.8 months), which can be influenced by the timing of snowmelt and the photoperiod in the Alpine Tundra ecosystem [49].

Shrubland phenophases are quite similar between biogeographical regions, but have greater variation within a region. OG occurs from 4 March to 20 March, and EOS occurs from 8 November to 30 November. The LS for shrublands varies from 7.3 to 8.5 months. A clear latitudinal gradient could not be found for shrublands, presenting a high variance in phenophase date. This variation, especially in the Mediterranean region, could be associated to differences in elevation and precipitation, together with heterogeneity in the landscape [24].