Introduction
Several analyses of normalized difference vegetation
index (NDVI) data derived from satellite remote sensing have pointed to a
positive trend in gross primary productivity (GPP) and leaf area index (LAI)
of the northern high latitudes in the recent decades
. Warming has also
occurred over this time. Global mean surface air temperatures increased by
0.2 to 0.3 ∘C over the past 40 years, with warming
greatest across northern land areas around 40–70∘ N
. Precipitation increases have also been
observed over both North America and Eurasia over the past century
. describe the
co-occurrence of these climatic and ecosystem changes. Here we investigate
increasing GPP of terrestrial ecosystems of northern Eurasia and determine
the relative attribution arising through changes in several geophysical
quantities, hereinafter referred to as “environmental variables”, as they
potentially drive observed temporal changes in vegetation productivity.
GPP is a physical measure of the rate of
photosynthesis, or the rate at which atmospheric CO2 is fixed by
autotrophic (generally green) plants to form carbohydrate molecules.
Photosynthesis, being a biological process, is regulated by several
environmental factors. Productivity is highest at the optimum temperature,
though this optimum can be modified by cold or warm acclimation
. Water availability also affects plant
hydraulics and chemistry by controlling the nutrient uptake through shoot
transportation . Increasing atmospheric
CO2 concentration increases GPP by biochemical fertilization for
C3 plants and increasing water use efficiency for both C3 and C4
plants .
Simplified land cover for northern Eurasia for year 2007 overlaid with the spatial distribution of the 10 flux tower sites whose GPP
(gross primary productivity) data were used to validate the GPP data derived from satellite NDVI (normalized difference vegetation index).
For our statistical analysis, we show the distribution of two fundamental types of vegetation types: (i) herbaceous, i.e. without woody stems,
which includes tundra in the north and grasslands (Eurasian Steppe) to the south, and (ii) wooded, i.e. plants with wood as its structural tissue,
which includes the boreal forests appearing in the middle and extending from the western to the eastern boundary. This land cover map has been
derived from the Moderate Resolution Imaging Spectroradiometer (MODIS) Type 5 land cover product . The details of the flux tower
sites are listed in Table .
There is both direct and indirect evidence of increasing productivity across
the northern high latitudes. Flask- and aircraft-based measurements show that
the seasonal amplitude of atmospheric CO2 concentration across the
Northern Hemisphere has increased since the 1950s, with the greatest
increases occurring across the higher latitudes . This
trend suggests a considerable role of northern boreal forests, consistent
with the notion that warmer temperatures have promoted enhanced plant
productivity during summer and respiration during winter
. Observed at eddy covariance sites,
net ecosystem exchange (NEE), the inverse of net ecosystem productivity
(NEP), is a strong function of mean annual temperature at mid- and high
latitudes, up to the optimum temperature of approximately
16 ∘C, above which moisture availability overrides the
temperature influence . Other studies have found vulnerabilities in
ecosystems of North America as well as Eurasia from warming-related changes in hydrological patterns
, thereby highlighting the importance of
precipitation. With warming, low-temperature constraints to productivity have relaxed
. Tree-ring data suggest that black
spruce forests have experienced drought stress during extreme warmth
. Over northern Eurasia, precipitation trends have
complicated the relationship between temperature and productivity, as the
increasing moisture constraints have made northern Eurasia more drought-sensitive . Increasing atmospheric CO2
concentration is another factor, as CO2 fertilization has been
demonstrated through observations, models, and FACE (free-air CO2
enrichment) experiments .
Cloudiness or shade can strongly influence vegetation productivity
, particularly over northern Eurasia
. Disturbances through forest fires also affect vegetation
productivity by destroying existing vegetation and allowing for regeneration
.
The role of temperature and precipitation in the positive trend of GPP of
northern high latitudes, especially northern Eurasia, has not been firmly
established. Few studies have examined the effect of CO2
concentration, cloudiness, and forest fires. Of these environmental variables,
CO2 concentration is unlike the others, given its long atmospheric
lifetime (∼100–300 years; ). Thus,
CO2 concentration is assumed to be more spatially uniform. As
a result, any statistical analysis using this variable will not be comparable
with the other variables. We consequently do not analyse the influence of
CO2 concentration. While some studies have focused on terrestrial
ecosystems of the pan-Arctic
or the high latitudes of
North America , few studies
have investigated the relative role of different environmental variables on
increasing GPP of northern Eurasia. Therefore, we assess in this study how
vegetation productivity trends in northern Eurasia are influenced by the
environmental variables air temperature, precipitation, cloudiness, and forest
fire. Objectives are to (1) calculate the long-term trend of both GPP and the
environmental variables, (2) assess the magnitude of the effect of the
environmental variables on GPP, (3) identify the seasonality of the
variables, and (4) identify the regions of northern Eurasia where the variables
boost or reduce GPP. Exploiting the availability of long-term time series
observation-based data we perform a spatially explicit grid point statistical
analysis to achieve the above objectives.
Biome property look-up table (BPLUT) for GPP algorithm with ERA-Interim and NDVI as inputs. The full names for the University of Maryland
land cover classes (UMD_VEG_LC) in the MOD12Q1 data set are evergreen needleleaf forest (ENF), evergreen broadleaf forest (EBF), deciduous needleleaf
forest (DNF), deciduous broadleaf forest (DBF), mixed forests (MF), closed shrublands (CS), open shrublands (OS), woody savannas (WS), savannas (SVN), grassland (GRS), and croplands (Crop).
UMD_VEG_LC
ENF
EBF
DNF
DBF
MF
CS
OS
WS
SVN
GRS
Crop
FPAR_scale
0.8326
0.8565
0.8326
0.8565
0.84455
0.7873
0.834
0.8437
0.8596
0.8444
0.8944
FPAR_offset
0.0837
-0.0104
0.0837
-0.0104
0.03665
-0.0323
-0.0107
-0.0183
-0.0044
-0.0297
-0.0517
LUEmax
0.001055
0.00125
0.001055
0.00125
0.001138
0.00111
0.00111
0.001175
0.001175
0.0012
0.0012
(kgC m-2 d-1 MJ-1)
Tmn_min (∘C)
-8
-8
-8
-6
-7
-8
-8
-8
-8
-8
-8
Tmn_max (∘C)
8.31
9.09
10.44
9.94
9.5
8.61
8.8
11.39
11.39
12.02
12.02
VPDmin (Pa)
500
1800
500
500
500
500
500
434
300
752
500
VPDmax (Pa)
4000
4000
4160
4160
2732
6000
4455
5000
3913
5500
5071
Results and discussion
Verification of satellite-derived GPP
The GIMMS-GPP and VIP-GPP, as well as their ensemble
mean (GPPsat), are individually verified against the flux-tower-based GPP data
using Pearson's correlation coefficient, percent bias, and the
Nash–Sutcliffe normalized statistic. Scatter plots
(Fig. ) show that GPP derived from the satellite
NDVI records is generally higher than the tower-based GPP at the flux tower
sites that have comparatively lower productivity (and vice versa). Moreover, the
agreement is stronger at lower-productivity sites than at higher-productivity
sites. Though Table lists all of the
verification statistics, we focus primarily on the annual GPPsat results
for
the rest of the study. The correlation coefficients are all positive and high
(0.7 for annual GPPsat); percent bias is predominantly negative (18.3 %); and
since all the values of the Nash–Sutcliffe efficiencies are above zero
(0.33), we conclude that the satellite NDVI-derived values are a more
accurate estimate of GPP than the observed mean for the respective flux tower
sites. Spatially explicit verification of GPPsat reveals that the correlation
is high and statistically significant for almost the entire study area
(Fig. a). GPPsat shows a general underestimation
in the boreal forests of the western parts of northern Eurasia and
overestimation in the Eurasian steppes to the south of the study area
(Fig. b).
Satellite-derived vegetation indices have been evaluated using a variety of
techniques. Using tree-ring width measurements as a proxy for productivity,
examined its relationship with NDVI from AVHRR instruments and found the correlation to be highly
variable across the sites, though consistently positive. Remarkably strong
correlations were observed in comparisons of GIMMS3g NDVI to aboveground
phytomass at the peak of summer at two representative zonal sites along two
trans-Arctic transects in North America and Eurasia .
From comparison of production efficiency model-derived NPP to the
stand level observations of boreal aspen growth for the 72 CIPHA (Climate
Impacts on Productivity and Health of Aspen) sites, the correlation was found to
be positive. LUE algorithms similar to the one used in this study for the
generation of GPP data sets from satellite NDVI produce favourable GPP results
relative to daily tower observations, with a strong positive correlation
. Evaluating the uncertainties in the
estimated carbon fluxes computed using a similar LUE-based GPP model,
concluded that the uncertainty in LUE
(ε) characterization is the main source of simulated GPP
uncertainty. GPP simulation errors under dry conditions are increased by an
insufficient model vapour pressure deficit (VPD) representation of soil water
deficit constraints on canopy stomatal conductance and ε
. It was also found that the GPP model does
not consider the response of ε to diffuse light due to canopy
clumping and shaded leaves .
Temporal changes in GPP
Across the study domain, regionally averaged GPPsat exhibits a trend of
2.2 (±1.4) gCm-2month-1decade-1.
Figure a displays the annual GPP trend map.
Increases are noted across most of the region except for a small area in the
north-central part of the region, just east of the Yenisey River. The largest
increases are located in the western and south-eastern part of the region.
Over half (69.1 %) of the study area exhibits a statistically
significant positive trend (95 % significance level), while
0.01 % of the area has a statistically significant negative trend.
Uncertainty in the ensemble mean GPP is illustrated by the coefficient of
variation map (Fig. b). The highest
uncertainty is noted in the north-central and the south-western part of the
region. The yearly increase in annual GPP for both GIMMS-GPP (red) and
VIP-GPP (blue; Fig. c) reveal the difference
between the two data sets, which is highest at the beginning of the study
period. The nature of increase in GPP is also different for the two data
sets, with the rise in one being more linear than the other. A possible explanation
for the differences in the two data sets is discussed in
Sect. . Examining the seasonality of GPP trends
(of GPPsat; Fig. ), we find that the summer trend
is greatest among all other seasons. This implies that the response of GPP to
environmental changes is greatest at the peak of the growing season. While
the productivity of the region is predominantly increasing, there are clearly
certain areas each season with decreasing productivity.
The GPP increase described here is consistent with the results of
, who also noted considerable interannual and spatial
variability, with many areas demonstrating decreased greenness and lower
productivity. Using a process-based model (LPJ-DGVM) to perform
a retrospective analysis for the period of 1982–1998,
found, after accounting for the carbon loss due to autotrophic respiration,
that boreal zone NPP increased by 34.6 g C m-2yr-1, which
is comparable to our estimate. The higher GPP trend in summer
(Fig. ), especially over the northern Eurasia
portion of the domain, suggests that the vegetation of this region is
predominantly cold-constrained, a finding described in other recent studies
.
Box plot showing grid distributions of seasonal GPP trends for
GPPsat. The GPP trends are expressed in g C m-2month-1
10 yr-1. The black band and middle notch represent the 2nd
quartile or median; box extents mark the 25th (1st quartile) and 75th (3rd
quartile) percentiles. Whiskers extend from the smallest non-outlier value to
the largest non-outlier value. The colours, green, red, orange, and grey
represent spring, summer, autumn, and annual seasonal trends respectively. As
described in Sect. , GPP trends for winter have not
been assessed in this study.
Temporal changes in the environmental variables
The regionally averaged air temperature increase is nearly monotonic and the
distributions displayed in Fig. a show that the
region has a predominantly positive trend for all parts of the growing
season. Warming is highest in autumn. A statistically significant increase in
temperature is noted for approximately half of the region. The greatest
increases are found in the north-eastern and south-western parts of the
region (maps not shown). Unlike temperature, precipitation does not exhibit a
sustained increase over the study period. While the regional median trend for
precipitation is highest for spring (Fig. b), the
range of trends for this region, from minimum to maximum, is highest for
summer. The fraction of the region experiencing significant increases in
annual precipitation is about 3 times the area experiencing significant
decreases. The significant positive trends are located in the north-eastern
and western parts (mainly boreal forests) of the domain, while significant
negative trends are located in the west-central (boreal forests) and
south-eastern (steppes) parts of the region (maps not shown). Along with the
regional averages of other environmental variables,
Table reveals the regional average of cloudiness,
which shows a negative trend. However, similar to precipitation, the spatial
standard deviation is very high, implying a high spatial variability in
cloudiness trends across the region. Unlike precipitation, a greater fraction
of the region is experiencing significant decreasing cloudiness or
a significant clear-sky trend (Fig. c). Compared to
the rest of the region, annual cloudiness shows higher negative trends in the
southern parts of the study area (maps not shown). Burnt area exhibits
significant trends, both positive and negative, over only 1 % of the
region, with the total yearly burnt area for the study area increasing from
15.9 to 17.1 million hectares from 1997 to 2010. The negative trend of the
regional mean (Table ;
Fig. d) is not significant.
Change in the environmental variables over the period of study
represented by seasonal trends. Panels (a–c) show distribution of
2 m air temperature, precipitation, and cloud cover respectively for
the period 1982–2010, and panel (d) illustrates seasonal trends of total
burnt area for the period 1997–2011. The temperature, precipitation, and
cloud cover data are taken from the Climatic Research Unit (CRU TS 3.21)
data set . Burnt area data from the Global Fire Emissions
Database (GFED; ).
Trend statistics for annual monthly averages of environmental variables. The first and second columns list the fraction of the
region with significant (95 % significance level) positive trends and negative trends respectively. The third column is the
regional mean trend of the variables per decade. The fourth column is the coefficient of variation, estimated as the distribution mean
divided by the standard deviation.
Environmental
Positive trend
Negative trend
Trend 10 yr-1
Coefficient
driver
(% of area)
(% of area)
(regional mean)
of variation
Temperature
50.9 %
0 %
0.39 ∘C
0.53
Precipitation
15.2 %
4.5 %
0.61 mmmonth-1
3.0
Cloudiness
7.9 %
16.9 %
-0.18 % of grid cell
4.2
Burnt area
0.7 %
0.3 %
-0.88 ha
20.6
Recent studies have reported similar changes in these environmental
variables. For the period of 1979 to 2005, found
temperature trends over the region range from 0.3 to 0.7 ∘Cdecade-1, and for most regions of the higher latitudes,
especially from 30 to 85∘ N, significant positive precipitation trends
have occurred. Contrary to the cloud cover trend we find here, studies
reported in AR4 suggest an increase in total cloud cover since the middle of
the last century over many continental regions, including the former USSR and
western Europe . The large spatial
variability in the gridded cloud cover trends (Table )
may explain the disagreement. Burnt area data, representing fire disturbance,
is dissimilar from the other environmental variables in that it spans only
14 years of the 29-year study record, and it is spatially
non-uniform, involving only a fraction of the total study area. This
limitation makes it difficult to assess impacts on vegetation productivity
. While the model used to generate the satellite NDVI-derived GPP data does not account for CO2 fertilization directly, the
fertilization effect may be partially represented through associated changes
in NDVI. As stated in Sect. , we do not analyse atmospheric
CO2 concentration due to its spatial homogeneity.
Attributing GPP changes to environmental variables and assessing seasonality
Annual GPP is affected by more than one environmental
variable. To study the impact of an individual environmental variable, we
eliminate the impact of other variables by performing partial correlations.
With the temporal range of the fire data (GFED) being a fraction of that of the
other environmental variables, it is not possible to compute the partial
correlation. Consequently, we are unable to assess the effects of only fire by
eliminating the effects of the other variables. Moreover, fires have been
found to be significantly correlated with annual GPP (GPPsat) for only a small
fraction (1.7 to 3.4 % depending on season) of the entire study area. The
impact of fires on annual GPP for the region is therefore ignored in this
study.
The regional median partial coefficient of determination (R2) for
significant values (Table ) suggests that the summer
values of the environmental variables have the highest influence on annual
GPPsat. The contrast between summer and the other seasons is strongest for
temperature, highlighting the importance of summer temperatures to annual
productivity. Figure reveals that the relationships
between annual GPP and the environmental variables are not completely
explained by simple correlation (R2), as the distributions of partial
correlations provide more information about the interaction. Considering only
significant correlations (Fig. ), we find that
increasing temperatures predominantly increase GPP. The relationship between
precipitation or cloudiness and GPP, on the other hand, leads to a
predominantly bi-modal distribution, with both positive and negative effects.
Other than spring, areas demonstrating significant negative partial
correlations appear to be larger than the areas of significant positive
partial correlations. Among the environmental variables assessed, temperature
has the highest partial coefficient of determination
(Table ). Moreover, unlike precipitation and
cloudiness, temperature has a predominantly positive relationship with annual
GPP. These relationships imply that, over recent decades, low temperatures
have been the major constraint for GPP in northern Eurasia.
Bean plots of the multi-modal distribution for significant (95 %
significance) partial correlation between annual de-trended GPP (GPPsat) and
the values of each de-trended environmental variable after eliminating the
influence of the other variables. A bean plot is an alternative to the box
plot and is fundamentally a one-dimensional scatter plot. Here it is
preferred over a box plot as it helps to show a multi-modal
distribution. The thickness of a “bean” is a function of the frequency of the
specific value – that is, the thicker a “bean” is for a value, the relatively
higher the number of grid points having that value. The values shown are
the Pearson's correlation coefficients which are based on the linear least-squares trend fit. Correlation values range from -1 to +1. Values closer
to -1 or +1 indicate strong correlation, while those closer to 0
indicate weak correlation. The colour of the box indicates the season of the
environmental variable being investigated (annual: grey; spring: green; summer: red; autumn: amber). The short horizontal black lines for each
“bean” is the median of that distribution.
Similar results were reported by , who concluded that
satellite-derived vegetation indices show an overall benefit for summer
photosynthetic activity from regional warming and only a limited impact from
spring precipitation. The dominant constraint of temperature was described by
, who found the same constraint to be decreasing. However,
our results contrast with those of , who concluded that at
the continental scale of Eurasia, vegetation indices in summer are more
strongly regulated by precipitation, while temperature is a relatively
stronger regulator in spring and autumn. Regarding the dominance of
temperature as a regulator, concluded that, over the last
decade, Eurasia has been more drought-sensitive than other high-latitude
areas.
Since GPP trends are highest in summer (Fig. ), the
peak of the growing season, we are interested more in the impact of the
environmental variables during summer on annual GPP since the terrestrial
vegetation is likely to be more responsive to variations in summer
environmental conditions relative to other seasons. Spatial analysis helps to
elaborate on the results shown in Table and
Fig. . Assessing the partial significant correlation of
annual GPP and summer temperature
(Fig. a;
Table ), we find that areas with a positive
correlation (62 % of the area) are concentrated to the north and east of the
region, which include both tundra and boreal forest areas. Negative
correlations occur across 2 % of the region, largely in the south within the
Eurasian steppes. For other parts of the year (maps not shown for spring and
autumn correlations but distributions represented in
Fig. ), significant negative correlations become more
spatially disperse, while significant positive correlations are limited to
the centre and west of the region for spring, becoming more disperse in
autumn. Determining the partial correlation between annual GPP and summer
precipitation, Fig. b reveals that the
areas of significant positive correlations (4 % of area) are scattered over
the southern part of the study area (steppes vegetation), while the
significant negative correlations (16 % of area) are scattered across the
north (tundra and boreal). Correlations for spring precipitation with annual
GPP (maps not shown) are predominantly positive, while that for autumn
precipitation is predominantly negative. The spatial correlations for summer
cloudiness and summer precipitation are similar
(Fig. c), though the area under
significant correlation is comparatively less. Negative correlation areas are
about 9 times more extensive than positive correlation areas
(Table ). Compared to summer, the area under
significant positive correlation is higher for spring, while the area under
negative correlation is higher for autumn (maps not shown).
Medians of the distributions of the relative partial significant contribution (R2 – 95 % significance) of each de-trended
environmental variable (except fire) of each season to the interannual variability in de-trended annual GPP (GPPsat). In each case the
total contribution may not add up to 100 %. In these cases the factors behind the unexplained attribution are not identified.
Environmental variable
Annual
Spring
Summer
Autumn
Temperature
26.1 %
26.5 %
37.7 %
19.9 %
Precipitation
22.9 %
20.7 %
20.7 %
17.9 %
Cloudiness
18.9 %
18.3 %
19.3 %
18.8 %
Spatial distribution of statistically significant (95 %
significance level) partial correlation between de-trended annual GPP
(GPPsat) and de-trended summer values of environmental variables (a) temperature, (b) precipitation, and (c) cloud cover.
Negative correlations are shown with shades of red and positive correlations are shown in shades
of blue.
Box plots of the distribution for correlation between de-trended
values of each environmental variable. The location of the box and in
particular the median on the y axis, on either side of the zero line,
reveals the predominant sign of the correlation.
The negative correlations for temperature and positive correlations for
precipitation and cloudiness in the southern grasslands (Eurasian steppes)
are not surprising, as these grasslands are relatively dry compared to other
biomes in the broader region. In this part of the study area, increasing
temperatures in summer may lead to greater water stress
. Decreasing precipitation would
increase water stress. Moreover, increasing cloud cover would tend to lead to
a higher probability of rain , thus relieving water
stress induced by warming in this relatively dry area. The cause of the
negative correlations in the north is unclear. The relationship may be
attributable to the predominantly positive relationship between cloud cover
(equivalent to inverse of sunshine duration) and precipitation
(Sect. ). In the light-limited and relatively
colder north, an increase in cloud cover could, on the one hand, cause a decrease
in direct radiation and increase in diffuse radiation, which may increase GPP
through higher LUE (). However, an increase
in cloud cover could decrease total solar radiation and, in turn,
productivity .
Recent studies have shown similar relationships to those found here.
showed that, across the pan-Arctic basin, while
productivity increased with warming, increasing drought stress can offset
some of the potential benefits. However, concluded
that while GPP was significantly higher during warm years for the pan-Arctic,
the same was not true for the Eurasian boreal forests, which showed greater
drought sensitivity. Positive impacts of warming on GPP have been suggested
in warming experiments . However, decreasing
growing-season forest productivity, represented as a decline in “greenness”
across northern Eurasia, may be a reflection of continued summer warming in
the absence of sustained increases in precipitation
.
Connection between annual GPP of northern Eurasia (GPPsat) and summer values of environmental variables shown as percentage of
the study area with statistically significant (95 % significance level) positive and negative partial correlation coefficients.
Environmental
GPP (ensemble mean)
variable
positive
negative
Temperature
61.7 %
2.4 %
Precipitation
3.9 %
15.9 %
Cloudiness
1.3 %
9.5 %
Relationships among individual environmental variables
Environmental variables are not independent of one another. We examine
correlations among the de-trended individual variables to better understand
their interactions. Figure shows
distributions of the correlations. The temperature–precipitation correlation
is predominantly negative, indicating that increases in precipitation did not
accompany recent warming. Significant negative trends are located in the
southern parts of the study area (steppes) as well as the boreal forests at
the western and eastern ends of the region. These changes may be leading to
increasing water stress, evidence of which is noted in a subset of the
region. Indeed, approximately 2.4 % of the area in the southern
parts of the study area (Fig. a) shows
significant negative partial correlation between annual GPP (GPPsat) and
summer temperature. The relationship between temperature and cloud cover is
similarly predominantly negative. Spatially, however, the significant negative
correlations are located in the central and western parts of the region.
Grid-cell-wise correlations between precipitation and cloud cover are
predominately positive, with the significant correlations spread out across
the region. As described in Sect. , the correlations
between precipitation and cloud cover help to explain why spatial
distributions of the correlation coefficients of precipitation and cloud
cover with GPP are similar. documented a positive
relationship between sunshine duration (equivalent to the inverse of cloud
cover) and vegetation greenness. While increasing cloud cover leads to an
increased probability of precipitation, and thus reduces water stress, it
also reduces the sunshine duration and hence GPP. According to
Table , regional mean precipitation has a positive
trend, while cloudiness has a negative trend. However,
Fig. reveals the predominantly
positive correlation between these two variables. This apparent contradiction
is because the long-term trends are calculated for the actual values, while
the correlation analysis is performed after de-trending (removing long-term
trends) the variables.
Consistent with our results, found that, in the boreal
and tundra regions of Alaska, NPP decreased when it was warmer and dryer and
increased when it was warmer and wetter. They also described how colder and
wetter conditions also increased NPP. concluded that while,
globally, annual GPP for boreal forests is significantly higher in warmer
years, the relationship does not hold true for Eurasian boreal forests, which
they identify to be more drought-sensitive. For this reason, regional GPP
variations are more consistent with regional wetting and drying anomalies, as
we note for the south-western part of the study region. In this study we
assessed only GPP. Other carbon cycle processes such as autotrophic and
heterotrophic respiration and disturbances may not be responding in a similar
manner. Additional studies are required before extrapolating these results to
other carbon cycle components.
Conclusions
The ensemble mean of the GPP data sets derived from GIMMS3g and
VIP NDVI data indicates that vegetation productivity generally increased
across northern Eurasia over the period 1982 to 2010, with a significant
increase for as much as 69.1 % of the region. A significant decrease
in GPP occurred across only 0.01 % of the region. We note some
disagreement in the nature and magnitude of the increasing GPP among the two
data sets. The regional mean trend for the ensemble mean GPP is
2.2 (±1.4) g C m-2month-1decade-1. The regional
analysis is consistent with results of prior studies which have suggested
that air temperature is the dominant environmental variable influencing
productivity increases across the northern high latitudes. Examining partial
coefficients of determination (R2), we find that the summer values of
temperature, precipitation, and cloudiness have the highest influence on
annual GPP. Considering the regional median of partial significant R2
values, summer air temperature explains as much as 37.7 % of the
variation in annual GPP. In contrast, precipitation and cloudiness explain
20.7 and 19.3 % respectively. A significant positive partial
correlation between summer air temperature and annual GPP is noted for
61.7 % of the region. For 2.4 % of the area, specifically
the dryer grasslands in the south-west, temperature and GPP are inversely
correlated. Precipitation and cloudiness during summer also impart a
significant influence, showing areas with both positive and negative
significant partial correlation with annual GPP. Fire has a very small
effect, with only up to 3.4 % of the region showing significant
correlation, and consequently the impact of fire on GPP was ignored for the
subsequent analysis. The spatial analysis reveals that the statistical
relationships are not spatially homogeneous. While warming likely contributed
to increasing productivity across much of the north of the region, the
relationship reverses in the southern grasslands, which are relatively dry.
That region exhibits increasing GPP, but with warming accompanying increased
moisture deficits potentially restricting continued productivity increase.
This result demonstrates that vegetation has been resilient to drought
stress, which may be increasing over time.
We recommended that this study be followed up with experiments conducted
using process-based models in which a single forcing variable independent of
the others is manipulated. If feasible, multiple models should be used in
order to quantify the uncertainty due to differences in model
parameterization. Depending on emissions, population, and other forcing
scenarios, rates of change in the environmental drivers such as air
temperature and precipitation may be different than those found in this
study. Thus it is critical to examine future scenarios of change across the
region to better understand terrestrial vegetation dynamics under the
respective model simulations. Environmental drivers influence other elements
of the carbon cycle beyond the individual plant. In order to determine how
terrestrial carbon stocks and fluxes have changed in the recent past, or may
change in the near future, all aspects of the carbon cycle should be
investigated in the context of changes in overarching climate influences.