SEARCH

SEARCH BY CITATION

Keywords:

  • carbon cycle;
  • climate change;
  • global warming;
  • northern ecosystems;
  • temperature;
  • vegetative period

Abstract

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Data and Methods
  5. 3. Results
  6. 4. Discussion
  7. 5. Conclusions
  8. Acknowledgments
  9. References
  10. Supporting Information

[1] Gridded daily temperature from 1950 to 2011 and atmospheric CO2concentration data from high-latitude observing stations and the CarbonTracker assimilation system are used to examine recent spatiotemporal variability of the thermal growing season and its relationship with seasonal biospheric carbon uptake and release in the Northern Hemisphere. The thermal growing season has lengthened substantially since 1950 but most of the lengthening has occurred during the last three decades (2.9 days decade−1, p < 0.01 for 1980–2011), with stronger rates of extension in Eurasia (4.0 days decade−1, p < 0.01) than in North America (1.2 days decade−1, p > 0.05). Unlike most previous studies, which had more limited data coverage over the past decade, we find that strong autumn warming of about 1°C during the second half of the 2000s has led to a significant shift toward later termination of the thermal growing season, resulting in the longest potential growing seasons since 1950. On average, the thermal growing season has extended symmetrically by about a week during this period, starting some 4.0 days earlier and ending about 4.3 days later. The earlier start of the thermal growing season is associated with earlier onset of the biospheric carbon uptake period at high northern latitudes. In contrast, later termination of the growing season is associated with earlier termination of biospheric carbon uptake, but this relationship appears to have decoupled since the beginning of the period of strong autumn warming during the second half of the 2000s. Therefore, owing to these contrasting biospheric responses at the margins of the growing season, the current extension in the thermal growing season length has not led to a concomitant extension of the period of biospheric carbon uptake.

1. Introduction

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Data and Methods
  5. 3. Results
  6. 4. Discussion
  7. 5. Conclusions
  8. Acknowledgments
  9. References
  10. Supporting Information

[2] The plant growing season is the period of the year when plants are able to grow and assimilate carbon from the atmosphere. In temperate and boreal regions of the Northern Hemisphere, it begins in spring with increasing temperatures and solar radiation, the melting of snow, eventual thawing of the soil organic horizons, and the start of photosynthesis. It terminates in autumn as temperatures and solar radiation decrease, soils refreeze, and photosynthesis ceases [Kimball et al., 2004; Euskirchen et al., 2006]. Therefore, temperature anomalies in spring and autumn affect the timing and duration of the growing season [Linderholm, 2006], which in turn control the seasonal onset and ending of the ecosystem carbon uptake period (number of days with net CO2 uptake from the atmosphere) in these regions [Goulden et al., 1996; Churkina et al., 2005; Barr et al., 2009; Richardson et al., 2010]. Rising temperatures during recent decades have resulted in a widely reported pattern of earlier and longer-lasting growing seasons, observed in a range of vegetational and biophysical indicators and from local to continental scales [Keeling et al., 1996; Myneni et al., 1997; Menzel and Fabian, 1999; Zhou et al., 2001; Linderholm, 2006; Piao et al., 2006; Christidis et al., 2007; Piao et al., 2007]. The greater rate of change observed in the beginning of the growing season is thought to be a response to rapid spring warming, and earlier snowmelt and soil thaw [Myneni et al., 1997; Smith et al., 2004; Barr et al., 2009], while the smaller change in termination date is likely connected with lower rates of autumn warming [Christidis et al., 2007] and the influence of other environmental effects on autumn phenology and growth cessation [Suni et al., 2003; Kimball et al., 2004; Hänninen and Tanino, 2011].

[3] Changes in timing and length of the growing season affect the timing and magnitude of carbon assimilation from local to hemispheric scale [e.g., Goulden et al., 1996; Keeling et al., 1996; Baldocchi and Wilson, 2001; Churkina et al., 2005; Barr et al., 2009]. Empirical and modeling studies show that warmer and earlier springs stimulate early season gross ecosystem productivity (GEP) more than ecosystem respiration (ER), leading to earlier onset of net ecosystem carbon uptake and enhanced net ecosystem productivity (NEP = GEP − ER) [Goulden et al., 1996; Keeling et al., 1996; Randerson et al., 1999]. Increased early season ecosystem carbon uptake due to warmer spring conditions has been found to account for the enhanced amplitude and advanced timing observed in the seasonal cycle of atmospheric CO2 concentration [Keeling et al., 1996; Randerson et al., 1999; Nemani et al., 2003].

[4] In contrast to spring, warmer autumns and the associated delay in senescence tend to produce larger increases in ER than GEP [Piao et al., 2008; Vesala et al., 2010], although autumnal ecosystem responses are still poorly understood. Autumnal carbon assimilation of high-latitude forest ecosystems tends to be light-limited [Suni et al., 2003] and therefore warmer autumn temperatures may only marginally increase photosynthesis. On the other hand, soil respiration during autumn is stimulated by warmer soil temperatures and deeper thawing depths in permafrost regions [Randerson et al., 1999]. In spite of this, several studies have found that respiratory losses are insufficient to offset concurrent GEP gains in a range of temperate and boreal ecosystems [Richardson et al., 2010; Dragoni et al., 2011]. Nevertheless, autumnal ecosystem carbon losses associated with warmer autumns and prolonged growing season have accelerated autumn CO2build-up in the atmosphere during recent decades, reducing the length of the period of net biospheric carbon uptake as inferred from the annual cycle of atmospheric CO2 [Piao et al., 2008]. As a result, longer growing seasons are not always associated with increased annual NEP as respiratory losses resulting from prolonged autumn warmth can be larger than concurrent GEP gains [Piao et al., 2007, 2008]. Further autumn warming and potential delay in the growing season may offset the increases in carbon uptake associated with an earlier growing season and reduce the ability of northern ecosystems to capture carbon dioxide from the atmosphere.

[5] A recent study based on updated satellite-derived Normalized Difference Vegetation Index (NDVI) data found that the hemispheric extension of the growing season between 2000 and 2008 was dominated by delayed termination, rather than earlier onset [Jeong et al., 2011]. During this period, autumn senescence was further delayed by 2.3 days presumably due to a recent amplification of late-season warming, whereas spring green-up further advanced by only 1.8 days due to a reduction of spring warming rates compared with previous decades.Zhu et al. [2012] used a previous version of the same satellite dataset and a network of phenological observations through the period 1982–2006 to investigate changes in the growing season in North America, where the lengthening has been driven almost exclusively by a trend toward delayed termination since the 1980s [Piao et al., 2007]. They found that the strongest delay in senescence occurred during the last four years (2003–2006) of their study period. The overall trends toward delayed senescence and extended growing season since 1982 were not statistically significant if these four years were excluded from trend analysis, indicating the contribution of the anomalies during this period.

[6] This apparent amplification of autumn warming and delayed termination of the growing season during recent years may have important implications for the carbon balance of northern ecosystems and the biosphere. Furthermore, in comparison with the 1980s and 1990s, temporal and spatial changes in the growing season during the 2000s are still very uncertain because few studies span the complete decade. In this study, we examine recent spatial and temporal variation of the thermal growing season in a longer-term context and their correlation with the timing of biospheric carbon uptake across the Northern Hemisphere (north of 30°N). Using an updated dataset of gridded daily temperature, we characterize the spatiotemporal variability of the duration and the timing of the start and end of the thermal growing season, based on a 5°C threshold, over the period 1950–2011. We then investigate the relationship between these parameters and the timing and duration of the period of biospheric carbon uptake based on in-situ measurements of atmospheric CO2concentration at four high-latitude observing stations and simulated CO2 concentrations from the CarbonTracker assimilation system [Peters et al., 2007]. This work complements earlier large-scale studies of the growing season as it spans some recent years not previously analyzed, thus helping to reduce current uncertainties in phenological changes over the past decade and their potential impact on the carbon cycle.

2. Data and Methods

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Data and Methods
  5. 3. Results
  6. 4. Discussion
  7. 5. Conclusions
  8. Acknowledgments
  9. References
  10. Supporting Information

2.1. Temperature Data and Thermal Growing Season

[7] We used the HadGHCND gridded dataset (2.75° × 3.75°) of maximum and minimum daily temperatures [Caesar et al., 2006] to derive daily mean temperature data over the period 1950–2011 at valid grid points on land areas north of 30°N, excluding Greenland. Missing values in the mean temperature field, located principally after the year 2007 and at grid points in Siberia, Japan and Scandinavia (Figure S1 in the auxiliary material), were imputed using the regularized expectation maximization algorithm with truncated total least squares [Schneider, 2001]. Given the large size of our data matrix, this regularization scheme (RegEM-TTLS) was preferred because it is computationally faster than the alternative scheme based on ridge regression (RegEM-Ridge). Grid points with more than three years of missing data were discarded from analysis. For convenience, data for February 29th in leap years were excluded, leaving 365 daily values for each year. Earlier versions of this dataset have been used in previous studies of the climatological growing season [Christidis et al., 2007] and extreme events [Caesar et al., 2006; Hamilton et al., 2012].

[8] The TIMESAT package [Jönsson and Eklundh, 2004] was used to compute the date of the start (SOS) and end (EOS) of the growing season for each grid point and year. First, each seasonal cycle was smoothed using a double logistic fit to reduce noise associated with high day-to-day variability. Then, the dates for SOS and EOS were obtained from the points where the smoothed cycle intersected a fixed temperature threshold of 5°C, which is a widely used value for determining the boundaries of the thermal growing season at mid and high latitudes [Frich et al., 2002]. The length of the thermal growing season (GSL) was calculated as the difference between EOS and SOS. Warmer regions like western Europe and the Middle East were masked out from further statistical analysis because our temperature threshold was too low to derive continuous annual time series of thermal growing season parameters. Temporal trends in the SOS, EOS and GSL parameters, over the period 1980–2011, were computed using the non-parametric Mann-Kendall trend test with a trend free pre-whitening procedure [Yue et al., 2002] as implemented in the zypR library. This 32-year period was chosen to make our results more comparable with studies based on satellite data.

2.2. Biospheric Carbon Uptake Period

[9] The seasonal cycle of atmospheric CO2 concentrations can be used as an integrated measure of the net carbon exchange between the terrestrial biosphere and the atmosphere [Keeling et al., 1996; Heimann et al., 1998]. We used the annual cycle of the atmospheric CO2 concentration to estimate the timing and duration of the biospheric carbon uptake period at Alert (82°N), Point Barrow (71°N), Ocean Station M (66°N) and Shemya Island (53°N) observing stations [Cooperative Atmospheric Data Integration Project, 2011]. Three of these stations are located in the Arctic and one in the north Pacific so their seasonal cycles primarily reflect changes in carbon exchange (NEP) across northern terrestrial ecosystems with little influence of ocean exchange, fossil fuel emissions and tropical biomass burning [Randerson et al., 1997; Piao et al., 2008].

[10] To extract the seasonal cycle from the monthly records, we used the curve-fitting procedures developed byThoning et al. [1989] and implemented in the CCGVU routine (available at ftp://ftp.cmdl.noaa.gov/ccg/software/). This algorithm decomposes the CO2time series into a long-term trend and seasonal and interannual components using a polynomial function and a series of annual harmonics combined with a digital filtering technique based on Fast Fourier Transform (FFT) and low-pass filters. We ran the routine using a quadratic polynomial to fit the trend, four annual harmonics for the seasonal component, and short and long frequency cutoff parameters of 100 and 650 days in the FFT low-pass filtering in order to remove high-frequency noise and isolate the trend component. These settings have been commonly used in previous studies of long-term changes in the CO2 seasonal cycle [e.g., Thoning et al., 1989; Buermann et al., 2007].

[11] The output of this procedure is a time series of smoothed and detrended annual cycles of the original data. The downward (upward) zero-crossing date of CO2 was determined as the day when the atmospheric CO2 annual cycle crosses the zero line from positive (negative) to negative (positive) values during spring (autumn). A sensitivity test showed that the relative temporal variations in these parameters of the CO2annual cycle are not affected by the choice of the parameter values used in the curve-fitting and filtering procedure (Figure S2 in theauxiliary material).

[12] The downward (spring: SZC) and upward (autumn: AZC) zero-crossing dates have often been used to derive the timing and duration of the net carbon uptake period (CUP) for each year [e.g.,Piao et al. 2008], though this is not strictly correct. If the annual cycle of CO2concentration at a high-latitude measuring station were controlled only by biospheric fluxes, then the net CUP would run from the spring maximum in concentration to the late summer minimum, rather than from the downward to upward zero-crossing times. Nevertheless, we use the zero-crossing times and their difference as surrogates for the onset, termination and duration of the net CUP for two reasons. First, to allow easier comparisons with earlier work. Second, because the timing of a zero-crossing can be determined more accurately than the timing of a maximum or minimum, especially if they are broad and superimposed with shorter-term variability. We use the term biospheric carbon uptake rather thannet biospheric carbon uptake because of this issue, though we expect that an earlier maximum in CO2concentration (e.g. due to the earlier onset of the net CUP) should be associated with an earlier spring downward zero-crossing (and similarly for autumn). In other words, if the shape of the seasonal cycle does not change significantly, then a relative change in the phase of the cycle identified at one point (e.g. the maximum) will be matched by relative phase changes at all other points (e.g. zero-crossings).

[13] In order to examine the reliability of the four CO2observing stations, we used column-averaged CO2mole fraction from the latest release of CarbonTracker (CT2010) to derive spatial fields of zero-crossing dates for the period 2000–2009. CarbonTracker is a reanalysis of the recent global surface land and ocean fluxes and the corresponding atmospheric CO2 mole fractions estimated by assimilating surface flask measurements from the NOAA/ESRL network and tall tower measurements using an Ensemble Kalman Filter technique [Peters et al., 2007, 2010]. The underlying atmospheric transport model TM5 [Krol et al., 2005] with 25 vertical layers is driven by meteorological data from the European Centre for Medium Range Weather Forecasts (ECMWF). The global data of carbon dioxide mole fraction are available at 3-hour time steps with 3° × 2° horizontal resolution and 25 vertical levels. We used daily column-averaged mole fraction data for the free-troposphere (850–500 hPa or 1.2–5 km), corresponding to levels 5 through 10 of the TM5 model before 2005 and levels 6 through 10 since 2006 due to an improvement in the vertical resolution. CO2 concentration gradients in this layer result from the exchange between the atmosphere and the land surface before subsequent transport and atmospheric mixing by weather systems. The same procedure explained above was applied to daily CO2 fields in order to extract the timing and length of the biospheric CUP for every grid box.

[14] Quantitative comparisons between the corresponding parameters of the thermal growing season and biospheric carbon uptake were made using linear correlation analysis. The statistical significance of correlations was estimated using a non-parametric random phase test with 1000 Monte-Carlo simulations [Ebisuzaki, 1997]. This method ensures that the significances are appropriate given the degrees of freedom associated with each of the time series being correlated.

3. Results

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Data and Methods
  5. 3. Results
  6. 4. Discussion
  7. 5. Conclusions
  8. Acknowledgments
  9. References
  10. Supporting Information

3.1. Spring and Autumn Temperatures

[15] Spring and autumn land temperatures have risen by about 1°C during the study period, but most of the warming has occurred during the last three decades (Figures 1a and 1b). On a mean hemispheric scale, temperatures in spring and autumn have risen at a similar rate of around 0.40°C per decade over the period 1980–2009 (Table 1). However, there is considerable spatial variability in the temperature trends. Eurasia has warmed significantly in both spring (0.50 ± 0.41°C decade−1) and autumn (0.41 ± 0.18°C decade−1), whereas in North America the warming has been significant only during autumn (0.42 ± 0.26°C decade−1). In general, since 1980 spring surface warming has been significant across Europe, northern Canada and most of the mid and high latitudes of Eurasia, while spring temperatures have either fallen or not risen significantly in the region located between 30°N and 70°N in central and western North America (Figure 1c). Autumn has also warmed significantly across mid and high latitudes of Eurasia and most of the boreal regions in North America but not in Europe (Figure 1d).

image

Figure 1. (a and b) Temporal variability and (c and d) Mann-Kendall linear trends of spring (March–April) and autumn (September–October) mean surface air temperature from the HadGHCND dataset. Area-averaged time series north of 30°N are given for North America (NAM, red), Eurasia (EURA, blue) and Northern Hemisphere (NH, black). Trends are for the period 1980–2011 and grid boxes with significant trends (p < 0.05) are indicated by black crosses. The period 2000–2011 is highlighted with gray shading in the left panels. In the right panels, grid boxes excluded from analysis due to permanent ice cover and missing data are shown in gray.

Download figure to PowerPoint

Table 1. Mann-Kendall Linear Trends (±1 s.e.) in the Hemispheric and Continental Area-Averaged Series of SOS, EOS, GSL and Spring (March–April) and Autumn (September–October) Temperatures Between 1980–2011a
RegionGrowing Season (Days · Decade−1)Temperature (°C · Decade−1)
SOSEOSGSLSpringAutumn
  • a

    Bold typeface indicates trends that are significantly different from zero at p < 0.05, and an asterisk indicates that the coefficient is also significant at p < 0.01.

Northern Hemisphere1.42 ± 0.62*1.50 ± 0.77*2.86 ± 1.31*0.43 ± 0.24*0.46 ± 0.17*
North America0.53 ± 1.201.39 ± 1.211.15 ± 2.110.03 ± 0.480.42 ± 0.26*
Eurasia2.25 ± 0.93*1.59 ± 0.63*4.00 ± 1.14*0.50 ± 0.41*0.41 ± 0.18*

3.2. Variability of the Thermal Growing Season

[16] During our study period there is a general long-term trend toward earlier and longer-lasting growing seasons until the first half of the 2000s, when an unprecedented period of delayed termination began and further extended the growing season by an amount similar to the advance in the beginning (Figures 2a–2c). This recent delay in termination of about 4 days resulted from a sustained increase in autumn temperatures of around 1°C during the second half of the 2000s (Figure 1b). The combination of earlier SOS with the recent shift toward delayed EOS has resulted in the longest growing seasons during our study period, particularly in Eurasia (Figure 2c). At the hemispheric scale, the growing season has extended by a week since 2005, following an initial extension of about 5 days during the 1990s. It has extended at a rate of 2.86 ± 1.31 days per decade since 1980, as a result of significant trends toward advanced SOS (−1.42 ± 0.62 days decade−1, p < 0.01) and delayed EOS (1.5 ± 0.77 days decade−1, p < 0.01) during the same period (Table 1). The extension has been larger in Eurasia (4.0 ± 1.14 days decade−1, p < 0.01) owing to significant advances in SOS and delays in EOS, whereas it has not been significant in North America (1.15 ± 2.11 days decade−1, p > 0.05) due to smaller changes in timing (Table 1).

image

Figure 2. (a–c) Temporal variability and (d–f) Mann-Kendall linear trends of the start (SOS), end (EOS) and length (GSL) of the thermal growing season. Area-averaged time series north of 30°N are given for North America (NAM, red), Eurasia (EURA, blue) and Northern Hemisphere (NH, black). Linear trends are for the period 1980–2011 and grid boxes with significant trends (p < 0.05) are indicated by black crosses. The period 2000–2011 is highlighted with gray shading in the left panels. In the right panels, grid boxes excluded from analysis due to permanent ice cover, missing data or discontinuous time series of growing season parameters are shown in gray.

Download figure to PowerPoint

[17] The spatial patterns of trends in GSL over the period 1980–2011 show that the extension of the growing season has been significant across most of Asia, eastern Europe, Alaska and the far north of Canada (Figure 2f). Significant advance in SOS has driven the extension across most of Asia and Europe (Figure 2d), whereas in Alaska and northern Canada the extension has been dominated by significant trends toward delayed EOS (Figure 2e). Although not statistically significant, the growing season length has decreased in the region undergoing spring and autumn cooling in central and western North America because of a moderate delay in SOS and no change in EOS. Thus the continental average in EOS (Figure 2b) is dominated by trends in Alaska and northern Canada.

3.3. Carbon Uptake Period and Growing Season

3.3.1. Observing Stations

[18] Most CO2 observing stations began recording during the second half of the 1980s and all of them have data spanning the period 1987 and 2009 (Table 2), which was chosen as the common period of analysis. The timing of CO2zero-crossing dates and the resulting length of the biospheric CUP vary considerably from year to year (Figures 3b–3d), but there are no significant trends during the period 1987–2009 (Table 2). In a longer term context, Point Barrow shows a clear trend toward advanced SZC (−1.5 ± 0.84 days decade−1, p < 0.001) and AZC (−1.6 ± 1.53 days decade−1, p< 0.05) since 1972, but the length of the CUP does not show any significant concurrent long-term trend as a result of relatively parallel advances in the zero-crossing dates.

Table 2. Mann-Kendall Linear Trends (±1 s.e.) of Zero-Crossing Dates and Length of CUP at our Four Monitoring Stations From the Global NOAA-ERSL Air-Sampling Network Between 1987–2009a
StationLocationPeriodZero-Crossing Dates (Days · Decade−1)CUP
Spring (SZC)Autumn (AZC)(Days · Decade−1)
  • a

    None of the trends is significantly different from zero at p < 0.05.

ALT82.5°N–62.5°W1986–2009−0.50 ± 1.60−0.03 ± 2.680.38 ± 2.50
BRW71.3°N–156.6°W1972–2009−0.82 ± 1.880.68 ± 3.360.66 ± 3.97
STM66.0°N–2.0°E1983–2009−1.14 ± 1.36−3.11 ± 3.40−1.63 ± 3.43
SHM52.7°N–174.1°E1987–2009−1.79 ± 2.12−0.87 ± 2.89−0.01 ± 2.72
image

Figure 3. (a) Interannually detrended seasonal cycle of atmospheric CO2and (b) time series of spring zero-crossing dates, (c) autumn zero-crossing dates, and (d) length of the carbon uptake period at high-latitude monitoring stations (ALT, BRW, STM and SHM) of the global NOAA-ERSL air-sampling network.

Download figure to PowerPoint

[19] The interannual variability of zero-crossing dates is more coherent between stations in spring than in autumn (Figure 3), indicating a larger spatial footprint of variations in biospheric carbon exchange during spring. Over the period 1987–2009, variations in the timing of SZC at the four stations are strongly and positively correlated with hemispheric anomalies in SOS (r = 0.66 to 0.70, p < 0.05), despite a strong continental contrast in the magnitude of the correlations (Table 3). This indicates that earlier SOS is associated with earlier biospheric carbon uptake. Figures 4a–4dillustrate the remarkable agreement between time series of SZC and hemispheric SOS over the full length of the individual stations. Note that the agreement is strong not only for interannual anomalies but also for longer-term decadal features, especially in the longer station records. For instance, it is clear that the hemispheric trend toward earlier SOS is consistent with the long-term variability of SZC at Point Barrow since 1972 (r = 0.64,p < 0.05; Figure 4b). The continental contrast in the correlations was further examined by a spatial correlation analysis (Figures 5a–5d), which reveals that the stations share a very similar SOS spatial footprint with significant correlations across most of Eurasia and poor correlations over North America. This suggests that the springtime biospheric signal in the atmosphere at high northern latitudes is dominated by temperature-induced variations in the timing of biospheric uptake over Eurasia. Regression analysis between SZC for the four stations combined and SOS averaged over Eurasia during 1987–2009 indicates that the timing of net spring carbon uptake at high-northern latitudes advances by 0.61 ± 0.09 days for each day of advance in SOS over Eurasia. The slope of the relationship is practically zero when using SOS averaged over North America, consistent with the poor correlations observed inTable 3. Point Barrow, in Alaska, is the only station where SZC is moderately correlated with SOS in North America (r = 0.30, p > 0.05), although over a small region near the station (Figure 5b).

Table 3. Correlations Between the Timing and Duration of the Biospheric Carbon Uptake Period at Each Monitoring Station and the Timing and Duration of the Climatological Growing Season Averaged Over the Northern Hemisphere (NH), North America (NAM) and Eurasia (EURA) Between 1987–2009a
StationSOS vs SZCEOS vs AZCGSL vs CUP
NHNAMEURANHNAMEURANHNAMEURA
  • a

    Correlations in bold are significant at p < 0.05 and the asterisk indicates that the coefficient is also significant at p< 0.01. The significance of the correlations was estimated using a non-parametric random phase test with 1000 Monte-Carlo simulations.

ALT0.660.010.740.61*0.59*0.360.280.51*0.08
BRW0.650.300.520.410.43*−0.23−0.24−0.320.02
STM0.70−0.030.71−0.120.12−0.210.180.350.07
SHM0.61−0.200.82−0.26−0.25−0.230.010.420.32
image

Figure 4. (a–d) Comparison of hemispherically averaged time series of SOS, (e–h) EOS and (i–l) GSL with SZC, AZC and length of the biospheric CUP at each monitoring station. EOS and GSL time series are plotted inverted to aid visual comparison with AZC and CUP. The correlation between the series over their common period is indicated in each plot, and significant correlations (p < 0.05) are indicated by an asterisk. The vertical gray shading in Figures 4e–4l highlight the period of strong autumn warming and delay in EOS between 2005–2011.

Download figure to PowerPoint

image

Figure 5. (a–d, e–h) Correlation of the time series of timing and (i–l) duration of the biospheric carbon uptake period at each monitoring station (black dot) with gridded fields of timing and duration of the thermal growing season for the period 1987–2009. The significance of the correlations at each grid box was estimated using a non-parametric random phase test with 1000 Monte-Carlo simulations and grid boxes with significant correlations (p < 0.05) are indicated by black crosses. Grid boxes excluded from analysis due to permanent ice cover, missing data or discontinuous time series of growing season parameters are shown in gray.

Download figure to PowerPoint

[20] In contrast to spring, variations of zero-crossing date in autumn are negatively correlated with hemispheric and continental anomalies in EOS (Table 3). Correlations are slightly stronger and more significant in North America than in Eurasia, but the continental contrast in the magnitude of correlations is not as strong as in spring. Practically all stations show a negative correlation, with two out of four stations having statistically significant correlations with EOS averaged over North America and the hemisphere (r = −0.41 to −0.61, p< 0.05). This means that delayed EOS or prolonged thermal growing season is associated with earlier termination of biospheric carbon uptake at high northern latitudes. A comparison between the time series of AZC and hemispheric-scale EOS shown inFigures 4e–4h, however, indicates that the shift toward delayed EOS during the second half of the 2000s is not associated with an increased advance in AZC as might be expected. Indeed, it appears that during this recent period there is a decoupling in the relationship between AZC and EOS. Before 2005, the interannual and longer-term variability of EOS and AZC agree relatively well. The hemispheric trend toward delayed EOS is coherent with a long-term advance in AZC at Point Barrow since 1972 (r = −0.39,p < 0.05 for 1972–2009; Figure 4f). Spatial correlations show predominantly negative correlations over large regions of Eurasia and North America but, unlike spring, the spatial footprints of the stations are quite variable and display fewer common regional features(Figures 5e–5h).

[21] As a result of the opposite biospheric responses to variations at the edges of the thermal growing season, the length of the biospheric CUP is not consistently correlated with GSL (Table 3; Figures 4i–4l). However, GSL averaged over North America is negatively and significantly correlated with CUP at Alert, in the eastern Artic (r = −0.51, p < 0.01), and Shemya Island, in the North Pacific (r = −0.42, p < 0.05). The spatial correlation maps for these stations show that this is due to a relatively large area of significant association centred in the northeastern part of the continent (Figures 5i and 5l). Overall, the results show that GSL does not equate with biospheric CUP because of the contrasting biospheric responses at the margins of the thermal growing season.

3.3.2. CarbonTracker

[22] In order to examine the spatial footprint of the relationships between the timing of the thermal growing season and biospheric CUP at the four CO2 observing stations, we used CO2zero-crossing fields based on free-tropospheric CO2concentrations as represented by CarbonTracker between 2009 and 2010. The interannual variability of zero-crossing dates at the observing stations and co-located grid boxes of CarbonTracker agree better in spring than in autumn (Figure S4 in theauxiliary material). This suggests that, at least at the location of the observing stations, the seasonal cycle simulated by CarbonTracker tends to better capture the greening signal in spring and the concurrent onset of biospheric carbon uptake than it does the process of senescence and carbon release in autumn. However, a comparison of variations in the timing of the thermal growing season at continental and hemispheric scales with zero-crossing dates averaged over high northern latitudes shown inFigure 6, confirms most of the features seen in the correlations based on observing stations. The timing of high-latitude SZC from CarbonTracker is positively correlated with anomalies in SOS averaged over Eurasia and the hemisphere but poorly correlated with SOS over North America. This is consistent with the results based on the observing stations shown inTable 3, and reinforces our observation that temperature-driven variations in biospheric activity over Eurasia dominate the biospheric signal in the atmosphere at high northern latitudes during spring.

image

Figure 6. (a–c) Comparison of continental and hemispheric time series of SOS, (d–f) EOS and (g–i) GSL with zero-crossing dates and length of biospheric CUP derived from CT2010 and averaged over latitudes north of 55°N. Time series of EOS and GSL are plotted inverted to aid visual comparison with AZC and biospheric CUP. Correlations are indicated in each plot and significant coefficients (p < 0.05) are indicated by an asterisk.

Download figure to PowerPoint

[23] Also in agreement with the observing stations, Figure 6shows that variations in high-latitude AZC from CarbonTracker are negatively correlated with hemispheric and continentally averaged EOS. However, correlations are clearly stronger with EOS averaged over Eurasia. This autumnal continental contrast is not as clear at the observing stations (Table 3). On the other hand, some degree of disagreement between the anomalies of AZC and EOS during recent years (Figures 6d–6f) appears to be consistent with the apparent decoupling in the relationship between AZC and EOS observed in the stations during the second half of the 2000s (Figures 4e–4h), although this could also be due to the change in vertical resolution of the TM5 transport model in 2006. Thus, CarbonTracker zero-crossing data confirm the relationships found at the observing stations and suggest a dominant role of Eurasian biospheric activity on variations in CO2 concentration in the atmosphere not only during spring but also during autumn.

[24] Figure 6 also shows a continental contrast in the relationship between the length of the biospheric CUP and GSL. The correlation is negative and stronger with GSL averaged over North America and positive but weak with GSL averaged over Eurasia. Although the relationship between GSL and length of the biospheric CUP was not consistent among the stations, this negative correlation over North America agrees in part with significant negative correlations seen at Alert and Shemya Island (Table 2).

4. Discussion

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Data and Methods
  5. 3. Results
  6. 4. Discussion
  7. 5. Conclusions
  8. Acknowledgments
  9. References
  10. Supporting Information

4.1. Recent Changes in Temperature and Growing Season

[25] Northern Hemisphere land surface temperature has increased by 0.32 ± 0.08°C per decade over the period 1979–2005, with the largest increases during winter and spring [Trenberth et al., 2007]. Rising temperatures during this period have resulted in a large-scale pattern of earlier and longer-lasting growing seasons seen across many vegetational and biophysical indicators [Keeling et al., 1996; Myneni et al., 1997; Menzel and Fabian, 1999; Zhou et al., 2001; Linderholm, 2006; Piao et al., 2006; Christidis et al., 2007; Piao et al., 2007]. Recent updates of the instrumental temperature record show that the rate of warming slowed between 2005 and 2008, where global surface temperatures declined by around 0.2°C, but increased again during 2009 and 2010 [Kaufmann et al., 2011]. This resulted in little overall warming during the past decade.

[26] We found that during this period the warming signal has weakened in spring but strengthened in autumn (Figure 1). On a hemispheric scale, autumn temperatures during the past decade have been the warmest since 1950 and have increased by up to about 1°C above the mean from 2005 until the end of the record in 2011. This change in the pattern of seasonal warming has resulted in a symmetrical extension of the thermal growing season during the second half of the 2000s, in contrast to previous decades where the extension was dominated by earlier springs [e.g., Linderholm, 2006; Christidis et al., 2007]. On average, hemispheric SOS has advanced by 4.0 days since 2005 while EOS has been delayed by 4.3 days, resulting in a unprecedented extension of the growing season by 8.4 days (Figure 2). This result confirms recent findings of phenological studies based on updated satellite NDVI data showing reduced rates of advance in spring greening, but substantial delay in vegetation senescence and extended growing season during the past decade [Jeong et al., 2011; Piao et al., 2011; Zhu et al., 2012; Zeng et al., 2011].

[27] Previous studies have reported that the rate of extension of the nominal growing season between the 1980s and early 2000s has been more pronounced in Eurasia than North America [Piao et al., 2007], mirroring differences in magnitude and timing of seasonal warming. Earlier beginning due to greater warming in spring has driven the extension in Eurasia, whereas in North America the extension has been dominated by delayed senescence associated with greater autumn warming [Piao et al., 2007]. However, during recent years the extension of the thermal growing season in Eurasia has been caused by a combination of significantly advanced SOS and delayed EOS, resulting from significant spring and autumn warming trends (Table 1). An earlier modeling study by Piao et al. [2007], found that during the period 1980–2002 autumn delay did not contribute to the extension of the growing season in Eurasia as a result of insignificant autumn warming. The contrast with our results shows the importance of recent autumn warming in the overall trend. In North America, the rate of extension of the thermal growing season has been significant only at high-northern latitudes and driven exclusively by delayed EOS, resulting from significant autumn warming in this region. Unlike Eurasia, most of North America has experienced a slight spring cooling during recent decades, precluding any advance in SOS. This climate feature has been associated with reduced photosynthetic activity and reversal of a former spring greening trend in the region [Wang et al., 2011], which coincides with some degree of delay in spring green-up [Zhu et al., 2012] and timing of SOS (Figure 2d).

[28] Regionally, recent autumn warming and the associated delay in EOS have been stronger in central Asia, Alaska and eastern North America (Figures 1 and 2). A similar spatial pattern of seasonal warming was found by Jeong et al. [2011]using NCEP/NCAR reanalysis temperature and satellite NDVI data through 2008. Furthermore, they found that increased late-season temperatures in these regions have been associated with a concurrent shift toward delayed timing of vegetation senescence. In boreal North America,Zhu et al. [2012] found that the extension of the growing season based on NDVI data has been driven exclusively by delayed vegetation senescence, particularly from 2003 until the end of their dataset in 2006. Although they did not discuss the role of temperature changes in their results, we find that this is the only region of North America that has undergone significant autumn warming (Figure 1d) and extension of the thermal growing season during recent decades (Figure 2f). In northwest China, significant delay in EOS since 1980 (Figure 2e) is consistent with the findings of a recent study based on daily temperature from local meteorological stations, showing a significant contribution of autumn delay to the extension of the thermal growing season across the region [Jiang et al., 2011].

4.2. Biospheric Response

[29] Interannual variations in the timing of the seasonal cycle of atmospheric CO2at high northern latitudes integrate large-scale changes in boreal growing season and associated impacts on the balance between photosynthetic drawdown and respiratory release of CO2 by northern terrestrial ecosystems [Keeling et al., 1996; Randerson et al., 1997; Heimann et al., 1998]. The downward spring zero-crossing point of the detrended annual cycle of CO2 is used as a surrogate (see discussion in section 2.2) for the timing of the seasonal switch of the biosphere from a net carbon source to sink, while the upward autumn zero-crossing point is used similarly for the return to a net carbon source [e.g.Piao et al. 2008]. Thus, zero-crossing dates are indicative of variations in the period when the biosphere acts as a net carbon sink during the boreal growing season.

[30] Variations in our simple indices of thermal growing season (SOS and EOS) correlate significantly with variations in zero-crossing dates at high northern latitudes, but the sign of the relationship is opposite at the two margins of the growing season(Table 3). Earlier thermal growing seasons (warm springs) are consistently associated with earlier biospheric carbon uptake, while in contrast delayed termination of the thermal growing season (warm autumns) is generally associated with earlier net biospheric carbon release. This means that the length of the thermal growing season is not equivalent to the period of biospheric carbon uptake. These large-scale relationships between the temperature-defined growing season and the biospheric carbon uptake period are consistent among stations (Figure 4), and are confirmed by high-latitude (north of 55°N) daily fields of CO2concentration in the free-troposphere as simulated by the CarbonTracker assimilation system (Figure 6).

[31] A similar seasonal contrast between spring (March–May) and autumn (September–November) temperatures and the timing of CO2zero-crossing dates was found recently byPiao et al. [2008] in a set of ten observing stations with long and continuous CO2 records between 20°N and 82°N, including Alert and Point Barrow. They mechanistically attributed the negative correlation between temperature and CO2zero-crossing in autumn to a larger stimulation of ecosystem respiratory release than photosynthetic drawdown of CO2 by warmer temperatures and delayed vegetation senescence. This mechanism requires a higher temperature sensitivity of respiration than photosynthesis in autumn. They estimated a regional temperature sensitivity of 5 gC m−2 °C−1 and 2.5 gC m−2 °C−1 for ecosystem respiration and photosynthesis, respectively. On the other hand, they attributed the positive correlation between temperature and CO2zero-crossing in spring to a larger stimulation of photosynthesis than ecosystem respiration by warmer temperatures, in agreement with earlier studies [Keeling et al., 1996; Myneni et al., 1997; Randerson et al., 1997, 1999; Piao et al., 2007]. These contrasting ecosystem responses to temperature are the most likely cause of the correlations observed between our simple growing season indices and the timing of biospheric carbon uptake, as our threshold-based thermal growing season depends completely on spring and autumn temperatures.

[32] Several studies have reported significant trends toward earlier spring and autumn CO2zero-crossing dates during recent decades, with changes of around 2 to 4 days per decade and larger changes in autumn [Keeling et al., 1996; Randerson et al., 1997; Piao et al., 2008; Thompson, 2011]. The study by Piao et al. [2008]also highlighted a trend toward shorter net carbon uptake period from 1980 to 2002 owing to a larger advance in autumn zero-crossing as a result of rising autumn temperatures. In contrast to these studies, we find that variations in zero-crossing dates and biospheric carbon uptake period at the high-latitude observing stations are characterized by the absence of significant long-term trends during the last two decades (Table 2). This is likely due to differences in the time-frame of our study, as zero-crossing dates at the observing stations closely follow large-scale variations in the timing of the thermal growing season (Figure 4) and trend coefficients, even when not significant, are nearly all negative. Considering a longer time-frame between 1972 and 2009, Point Barrow shows significant long-term trends toward earlier zero-crossing in spring and autumn. Furthermore, results fromPiao et al. [2008]indicate that the magnitude of the trends in zero-crossing dates and carbon uptake period from 1980 to 2002 are stronger at mid latitudes and relatively weak at high latitudes. This may also explain in part the lack of significant trends in our results between 1987 and 2009.

[33] An examination of the temporal variability of zero-crossing dates and timing of the thermal growing season suggests that the relationship between the timing of autumn zero-crossing and the termination of the thermal growing season may have decoupled during the second half of the 2000s, in connection with the shift toward later termination of the thermal growing season. This apparent change in the relationship could be caused by some bias in our temperature dataset during recent years due to the imputation of missing temperature data in a fraction of grid boxes (Figure S1 in theauxiliary material). However, if true we would also expect to see a similar bias in spring and clearly this is not the case. Also, the magnitude of hemispheric temperature anomalies in spring and autumn are highly consistent with the CRUTS3.1 dataset (Figure S3 in the auxiliary material) [Mitchell and Jones, 2005]. Comparisons of HadGHCND with two other datasets representing changes in cold and warm days and nights also show good agreement on the global scale (L. Alexander, personal communication, 2012). Therefore, it seems unlikely that this apparent change in the relationship is associated with biases in the temperature data. It may be related to the influence of other natural or anthropogenic factors on summer or late growing season carbon fluxes that may have modulated the overall biospheric response to recent autumn warming. The role of the ocean and other factors (e.g., soil moisture, CO2and nitrogen fertilization, ecosystem disturbances) that could critically influence the autumn zero-crossing time during recent years warrants further research but is beyond the scope of this paper.

[34] Our correlation analysis indicates that biospheric activity over Eurasia dominates variations in CO2 concentration in the atmosphere at high northern latitudes during spring and apparently also during autumn. In part, this is due to the larger size of Eurasia but also may reflect continental differences in the strength of the carbon sink. During recent decades, the carbon sink has been stronger in Eurasia than in North America [Denman et al., 2007]. This has been associated with stronger vegetation greening trends and lengthening of the growing season in Eurasia because of larger rates of warming, particularly in spring [Zhou et al., 2001; Piao et al., 2007, 2011; Jeong et al., 2011].

[35] Further warming due to ongoing climate change is expected to result in longer active seasons until other factors (e.g., soil moisture, photoperiod) limit the extension. Climate model projections indicate that the length of the thermal growing season will increase by more than a month by the end of the twenty-first century, due to a major shift of the annual cycle toward higher temperatures [Tebaldi et al., 2006; Christidis et al., 2007; Meehl et al., 2007; Ruosteenoja et al., 2011]. In addition, delayed termination is expected to become a major contributing factor in modifying the length of the thermal growing season during the course of the century [Christidis et al., 2007]. This may result in increased rates of soil respiration in autumn and further decrease the length of the net carbon uptake period in northern ecosystems. However, our results suggest that the overall response of the biosphere to autumnal temperature changes is more complex compared with spring and can be influenced by other factors.

5. Conclusions

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Data and Methods
  5. 3. Results
  6. 4. Discussion
  7. 5. Conclusions
  8. Acknowledgments
  9. References
  10. Supporting Information

[36] The length of the thermal growing season has increased significantly since 1950, but most of the increase has taken place during the last three decades (1980–2011) and across Eurasia. Unlike previous studies, which had limited data coverage during the past decade, our results reveal a change from asymmetrical to symmetrical extension of the thermal growing season during the second half of the 2000s due to a shift toward later termination caused by unprecedented autumn warming. This result is consistent with the findings of recent studies based on updated NDVI data showing a shift toward delayed growing season in the Northern Hemisphere during the past decade. Our analyses also showed that earlier beginning together with further delay in termination during the second half of the 2000s has resulted in the longest thermal growing seasons since 1950.

[37] We have shown that earlier thermal growing season is associated with earlier onset of biospheric carbon uptake, whereas delayed termination of the thermal growing season, rather than being associated with prolonged biospheric carbon uptake, is associated with earlier termination of biospheric carbon uptake. Thus the length of the thermal growing season is not equivalent to the length of the biospheric carbon uptake period and the current extension in potential growing season length has not led to an extended biospheric carbon uptake period. This result is consistent with recent literature reports of increased ecosystem carbon losses during warm autumns and prolonged growing seasons. However, the strong delay in termination of the thermal growing season owing to unprecedented autumn warming during the second half of the 2000s was not associated with a concurrent increase in the advance of the termination of biospheric carbon uptake. This has led to an apparent decoupling in the relationship between autumn zero-crossing dates and the timing of the termination of the thermal growing season during the later part of the past decade.

Acknowledgments

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Data and Methods
  5. 3. Results
  6. 4. Discussion
  7. 5. Conclusions
  8. Acknowledgments
  9. References
  10. Supporting Information

[38] We thank Per Jönsson and Lars Eklundh for making available their TIMESAT code for use in our analyses. CarbonTracker (CT2010) data were kindly provided by Andy Jacobson from NOAA Earth System Research Laboratory. This research was supported by a scholarship from the Chilean Government to J.B. for doctoral studies under the program Formación de Capital Humano Avanzado of CONICYT. K.R.B. and T.M.M. acknowledge support from UK NERC (under grant NE/G018863/1). J.C. was supported by the Joint DECC/Defra Met Office Hadley Centre Climate Programme (GA01101). We thank the Associate Editor Peter Rayner and one anonymous reviewer for their constructive comments, which improved this article.

References

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Data and Methods
  5. 3. Results
  6. 4. Discussion
  7. 5. Conclusions
  8. Acknowledgments
  9. References
  10. Supporting Information

Supporting Information

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Data and Methods
  5. 3. Results
  6. 4. Discussion
  7. 5. Conclusions
  8. Acknowledgments
  9. References
  10. Supporting Information

Auxiliary material for this article contains four supplementary figures.

Auxiliary material files may require downloading to a local drive depending on platform, browser, configuration, and size. To open auxiliary materials in a browser, click on the label. To download, Right-click and select “Save Target As…” (PC) or CTRL-click and select “Download Link to Disk” (Mac).

Additional file information is provided in the readme.txt.

FilenameFormatSizeDescription
gbc1951-sup-0001-readme.txtplain text document2Kreadme.txt
gbc1951-sup-0002-fs01.epsPS document7265KFigure S1. Spatial and temporal distribution of missing data.
gbc1951-sup-0003-fs02.epsPS document96KFigure S2. Sensitivity of zero-crossing time anomalies to different frequency cutoffs in the low-pass filtering and number of annual harmonics used to extract the annual cycle from monthly CO2observations at Point Barrow.
gbc1951-sup-0004-fs03.epsPS document144KFigure S3. Comparison of time series of spring and autumn mean surface air temperatures from the HadGHCND and CRUTS3.1 datasets averaged over land regions north of 30 deg N.
gbc1951-sup-0005-fs04.epsPS document327KFigure S4. Comparison between observed (GLOBALVIEW2010) and CarbonTracker-based (CT2010) time series of SZC, AZC, and CUP at each of our CO2observing stations.
gbc1951-sup-0006-t01.txtplain text document1KTab-delimited Table 1.
gbc1951-sup-0007-t02.txtplain text document1KTab-delimited Table 2.
gbc1951-sup-0008-t03.txtplain text document1KTab-delimited Table 3.

Please note: Wiley Blackwell is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.