Changes in the phenology of physical and ecological variables associated with
climate change are likely to have significant effect on many aspects of the
Baltic ecosystem. We apply a set of phenological indicators to multiple
environmental variables measured by satellite sensors for 17–36 years to
detect possible changes in the seasonality in the Baltic Sea environment. We
detect significant temporal changes, such as earlier start of the summer
season and prolongation of the productive season, in several variables
ranging from basic physical drivers to ecological status indicators. While
increasing trends in the absolute values of variables like sea-surface
temperature (SST), diffuse attenuation of light (Ked490) and
satellite-detected chlorophyll concentration (CHL) are detectable, the
corresponding changes in their seasonal cycles are more dramatic. For
example, the cumulative sum of 30 000 W m
Satellite-derived data sets used here to produce phenological indicators.
Estuarine areas world-wide are experiencing rapid changes due to anthropogenic pressures of both local and global nature (Cloern et al., 2016). The ecosystem of the brackish Baltic Sea has been under anthropogenic stress for many decades (Elmgren, 1989, 2001). Both direct effects, such as input of nutrients and various pollutants, and indirect effects through climate change are important. Time series of environmental variables show variability at multiple scales but separating the effects of natural climate variability from the effects of anthropogenic climate change is difficult. We apply a set of phenological indicators to multiple environmental variables, mostly from satellite sensors, to detect changes in the environment. It appears that phenological indicators are very sensitive in detecting environmental change. We show significant changes in the seasonality of both the physical drivers and in ecological indicators of the Baltic Sea. Increasing trends are detectable in the absolute values of sea-surface temperature (SST) and water turbidity, but changes in their seasonal cycles are more clear-cut. The seasonality of a number of abiotic and biotic variables for which data are available has changed drastically during the last few decades. These changes are likely to have major effects on many aspects of the Baltic Sea ecosystem.
The data sets used in this analysis and
their sources are shown in Table 1. Surface incoming shortwave irradiance
(SIS) data were produced from geostationary Meteosat satellites (Mueller et
al., 2009, Müller et al., 2015) using a climate version of the Heliosat
algorithm (Cano et al., 1986, Beyer et al., 1996) and obtained from the
Satellite Application Facility on Climate Monitoring (CM SAF,
Summary of proposed climate indicators.
SST data produced by NOAA and NASA were obtained
from
Near-surface wind data were assembled from two sources. Data for 1987–2011
are version 3.5a from the Cross-Calibrated Multi-Platform (CCMP) Ocean
Surface Wind Components (Atlas et al., 2008), available from
Satellite-detected coefficient of attenuation of diffuse downwelling light at
490 nm (Ked490) and near-surface chlorophyll concentration (CHL) are
produced by the ESA Ocean Colour Climate Change Initiative project (Lavender
et al., 2015) using satellite data archives of NASA's SeaWiFS and MODIS-Aqua
sensors and ESA's MERIS sensor. Version 2.0 data sets were downloaded from
Fraction of cyanobacteria accumulations (FCA, Kahru et al., 2007; Kahru and Elmgren, 2014) is a form of presenting the frequency of cyanobacteria accumulations that is normalized to the number of unobstructed satellite views of the sea surface. This normalization is needed as clouds often cover the sea surface and make it impossible to detect accumulations.
For the analysis of changes in the annual cycle we use the following
phenological indicators (Table 2):
Day of year when a threshold value is first reached (DF Day of year when a threshold value is last reached (DL Duration between the first and last reaching of a threshold value (DD Day of year when the annual maximum occurs (DM Count of days above a threshold value (DC
These indicators can be applied to different satellite-derived or in situ
variables. Satellite versions of these indicators were spatially averaged
over the area of interest – e.g., the whole Baltic Sea or parts of it (Fig. 1).
The nomenclature of indicators uses two characters showing the type of
indicator (e.g., DF, DL, etc.), followed by the variable type (e.g., SST for
sea-surface temperature), and followed by the threshold value (e.g., 16 for
16
Maps of the study areas.
The existence of trends and their significance was evaluated with the
nonparametric Sen slope (Sen, 1968) and the Mann–Kendall test using 95 %
significance level (Salmi et al., 2002). In parallel, 95 % confidence
limits of the least squares linear regression slope (as implemented in NMath
numerical libraries,
The radiation budget at the Earth's surface is a key variable affecting other
variables such as surface temperature, primary production, etc. The incoming
shortwave irradiance (SIS, W m
Time series of the daily surface incoming shortwave irradiance
(SIS, W m
Changing seasonality in cumulative surface incoming shortwave
irradiance (SIS, W m
As global time series of SST are available for
several decades (Casey et al., 2010), SST is a good variable for phenological
analysis. While SST itself is an important driver affecting the rates of
biological processes, its indirect effects through its influence on
stratification are probably more important. Near-surface stratification
suppresses vertical mixing in the water column and affects many ecological
processes, such as the flux of nutrients to the surface layer or the
accumulation of cyanobacteria near the water surface. As cyanobacteria growth
in temperate environments is enhanced by higher temperatures and near-surface
stratification, SST is an important environmental driver for these blooms
(Paerl and Huisman, 2008). Cyanobacteria growth typically accelerates when
SST exceeds 12
Changes in satellite-detected SST phenology in the Baltic Sea (area
indicated in Fig. 1a). The symbols and regression lines are (left to right):
first day when 12
It appears that while the springtime warming of the sea surface has become
significantly earlier (i.e., the slope of DFSST vs. year is negative),
particularly for colder temperatures, the time in summer when the higher
temperatures are reached has not changed and the trend is not significant for
any SST above 16
Wind speed and direction are important variables that affect the biology of the Baltic surface waters. In a previous study (Kahru et al., 2007) we found a correlation between the location of the major cyanobacteria accumulations either in the southern or northern half of the Baltic and the strength of wind in the northeasterly direction. We therefore examined the mean annual cycle in the strength of winds in the northeastern direction (NE-ward winds) and the possible changes to it. The annual cycle of the NE-ward winds has a minimum in April and a subsequent increase after that (Fig. 6a). We examined whether the onset of the summer increase in NE-ward winds has changed. While the last 7 years (2008 to 2015) show a steady shift towards earlier start of the summer wind increase (Fig. 6b), the large interannual variability precludes establishing a long-term trend and the changes in the starting time of the summer increase in NE-ward winds are not significant.
The coefficient of downwelling light at 490 nm (Ked490) is a commonly used
indicator of water transparency (e.g., Kratzer et al., 2003; Lee et al.,
2005) that can be estimated from a satellite sensor. It reflects both light
absorption – e.g., by colored dissolved organic
matter (CDOM) and phytoplankton pigments – and scattering of light
by particles. The timing of the increase in Ked490 has changed significantly
during the time period with reliable satellite data (1997 to present). The
summer period with high Ked490 corresponding to low water transparency has
become earlier and persisted longer into the fall (Fig. 7). Significantly,
these changes have been more pronounced for the higher values of Ked490 than
for the lower values of Ked490. For example, using the regression lines in
Fig. 7, we can estimate that from 1998 to 2013 the period with mean Ked490
over 0.1 m
Temporal changes in the start and end of Ked490
The satellite-derived surface concentration of CHL is an important measure of the amount of phytoplankton, but is known to have large errors in the Baltic Sea (Darecki and Stramski, 2004; Attila et al., 2013). This is due both to problems in the atmospheric correction procedures and in the separation of phytoplankton pigments from the CDOM. The standard CHL algorithm essentially measures total absorption of blue light by a mixture of phytoplankton pigments, non-algal particles and CDOM. The separation of these components is made difficult by the large and variable concentrations of CDOM that are typical of the Baltic Sea. While the satellite-derived CHL overestimates true near-surface CHL in the Baltic Sea due to the high concentration of CDOM (Darecki and Stramski, 2004), the phenology of CHL is still a meaningful indicator of the timing of phytoplankton blooms, as the early increase in turbidity is well correlated with phytoplankton spring bloom and the summer maximum in turbidity is correlated with the cyanobacteria blooms in the central Baltic Sea.
The time series of CHL in the central Baltic Sea (Fig. 8) shows the
well-known annual cycle with maxima corresponding to the summer cyanobacteria
and the spring diatom blooms. A statistically significant increasing trend
can be detected in the 1998–2013 time series of mean CHL in central Baltic
Sea (Fig. 8) with a slope of 0.067 mg m
Time series of the 5-day mean CHL (mg m
Cyanobacteria are a major component of the Baltic phytoplankton community
with some unique features. Their ability to fix nitrogen makes them important
in driving the nitrogen cycle and stimulating summer primary production (e.g.,
Larsson et al., 2001; Karlson et al., 2015). The propensity of the
co-dominant genus
Temporal change in the timing of cyanobacteria accumulations for the Baltic Sea. The circles connected with a solid line show the “center of timing” (modified after Kahru and Elmgren, 2014, by adding 2014 data). The red line is the estimated linear regression for the Baltic (area in Fig. 1a). The gray dotted lines show the mean start and end of the accumulations for the whole Baltic Sea. For reference, 1 July is year day 182 (183 in leap years), year day 200 is July 19 (in leap years, July 18).
We have applied a uniform set of phenological indicators to a number of variables ranging from physical drivers of the environment, such as incoming shortwave energy, sea-surface temperature and winds, to ecological (Ked490, CHL) and biological (cyanobacteria) components of the Baltic Sea ecosystem. Satellite-derived variables have the advantage of providing extended areal coverage instead of a point sample, with a regular and frequent sampling that is required for estimating phenological indicators. While satellite measurements using the visible and infrared spectrum are limited by cloudy periods, the effective sampling frequency is still much higher than with shipboard monitoring. For variables with large systematic errors such as satellite-detected CHL in the Baltic Sea, the phenological indicators are preferred as they have potentially less uncertainty compared to absolute values, particular during periods of rapid change like the phytoplankton spring bloom. However, uncertainty in timing becomes bigger during periods of slow change. The effect of missing days due to cloud cover is another factor causing errors in phenology. As we track the phenology of variables averaged over large areas (e.g., most of the Baltic Sea), the potential errors and quasi-random fluctuations are smoothed out. Systematic differences in missing data due to cloud cover (e.g., mostly southern or mostly northern areas) are another source of uncertainties in our phenological indicators and may be responsible for some of the interannual wiggles in the time series. Total uncertainties in our phenological indicators are complex and caused by errors in individual measurements, and the confounding effects of spatial and temporal compositing. However, the interannual trends in most variables are quite clear and that makes us confident that the errors are much smaller than the observed trends.
While significant trends can be detected in the absolute values of a number
of variables – e.g., the increasing trends of SST, Ked490 and CHL – the
phenological indicators often show more marked change and clearer trends,
even when trends in the mean values are questionable or nonsignificant. For
example, the time series of incoming shortwave irradiance (SIS, W m
The mean annual SST has increased during the last decades, in agreement with
the overall warming trend. However, changes in the timing of the annual SST
cycle are more drastic. As a result of the earlier warming and later cooling,
the period with SST above 16
Dramatic changes have occurred recently in the annual cycle of the
coefficient of light attenuation (Ked490), an indicator of water
transparency. The duration of the period with elevated but intermediate
Ked490 has somewhat increased but the duration of the period with high mean
Ked490 (over 0.4 m
The changes in the timing of Ked490 and CHL are related and reflect increased turbidity and decreased water transparency in the Baltic Sea. An obvious consequence of the increased Ked490 is that less light reaches depths below the surface. While this analysis was done for the central Baltic with small areas of benthic photosynthesis, we can assume that benthic communities in the coastal areas must also be experiencing significant reduction in light intensity due to the decreased water transparency. We can hypothesize that the increased turbidity and decreased water transparency are related to increased phytoplankton concentrations and increased bacterial production. Likely effects on the rest of the ecosystem including commercially important fisheries should be further evaluated.
The phenological indicators that we applied to a number of satellite-detected variables show significant climate-related changes in the Baltic Sea ecosystem. It appears that the biological response to climate warming is amplified, compared to the rate of change in the physical forcing (SIS and SST). As satellite-derived CHL algorithms in the Baltic Sea are inaccurate and noisy (e.g., Darecki and Stramski 2004) – primarily due to the confounding influence of high concentrations of CDOM – trends in absolute values of CHL are difficult to interpret. For example, an apparently increasing trend in CHL could be influenced by increasing concentrations of CDOM or non-algal particles. However, the start of the annual increase in CHL is much more likely to reflect the increase in phytoplankton due to the known phytoplankton spring bloom and the dependence of photosynthesis on the annual cycle of light and stratification. Elevated concentrations of CDOM and non-algal particles are perhaps more likely to be partly responsible for extending the high CHL period in the fall.
Cyanobacterial blooms are a worldwide phenomenon associated with
eutrophication of lakes, reservoirs and estuaries, toxic contamination of
drinking water, and other undesirable effects. They cause major environmental
problems in the North American Great Lakes (Stumpf et al., 2012), in lakes in
China (Paerl et al., 2011) and in the Baltic Sea (Funkey et al., 2014; Kahru
and Elmgren, 2014). As cyanobacterial growth in temperate zones is enhanced
by higher temperatures and near-surface stratification, these blooms are
expected to become more frequent as a result of climate change (Pearl and
Huisman, 2008, 2009; Wiedner et al., 2007). Cyanobacterial growth typically
accelerates above 12
Even though the Baltic Sea is one of the marine areas in the world best
covered by observations, the frequency of long-term sampling is insufficient
for a confident detection of similar phenological indicators using
measurements at sea. Instead, comparisons can be made with mathematical
models simulating ecosystem dynamics over decades with high temporal
resolution (e.g., Eilola et al., 2011; Hense et al., 2013; Meier et al.,
2012). For instance, BALTSEM (BAltic Long-Term large Scale Eutrophication
Model, Savchuk et al., 2012) produces quite similar tendencies in the Central
Baltic Sea (Fig. 1b). The prolongation of the vegetative season from about
190 days in the beginning of 1970s to about 230 days in the middle of 2010s
has been accompanied by a tripling of the net primary production and shift of
the annual biomass maximum from spring to summer. Due to earlier warming and
delayed cooling, the duration of the period with simulated surface water
temperature exceeding 14
Phenological indicators are sensitive in detecting environmental changes that are often hardly detectable using absolute values of the respective environmental variables. Moreover, these indicators – e.g., the duration of the growth season – may have special ecological significance. Using these phenological indicators we show significant and in some cases drastic changes in the seasonality of the Baltic Sea. These changes are evident in multiple variables from purely physical to ecological to biological. For several ecologically important variables (Ked490, CHL) the length of the annual period of high values has increased by a factor of 2 or more during the last 2 decades. The analyses reported above are based on satellite data, meaning that the phenological analyses of this type can be made for most areas of the globe and not only for comparatively data-rich areas like the Baltic Sea.
Financial support was provided by the Swedish Research Council Formas and Stockholm University's Baltic Ecosystem Adaptive Management Program. During the writing of this paper M. Kahru was supported by Hanse-Wissenschaftskolleg (Delmenhorst, Germany). The authors thank Uwe Pfeifroth, Jörg Trentmann, and Christine Träger-Chatterjee for help in accessing SIS data, H. W. Paerl, R. P. Stumpf and C. Rolff for comments on the manuscript, and CM SAF, NASA Ocean Color Processing Group and ESA OC-CCI group for satellite data. Edited by: K. G. Schulz