Volume 34, Issue 19
Hydrology and Land Surface Studies
Free Access

Diverse responses of vegetation phenology to a warming climate

Xiaoyang Zhang

Xiaoyang Zhang

Earth Resources Technology, Inc. at NOAA/NESDIS/STAR, Camp Springs, Maryland, USA

Search for more papers by this author
Dan Tarpley

Dan Tarpley

Center for Satellite Applications and Research (STAR), NESDIS, NOAA, Camp Springs, Maryland, USA

Search for more papers by this author
Jerry T. Sullivan

Jerry T. Sullivan

Center for Satellite Applications and Research (STAR), NESDIS, NOAA, Camp Springs, Maryland, USA

Search for more papers by this author
First published: 11 October 2007
Citations: 229


[1] Warming climates have been widely recognized to advance spring vegetation phenology. However, the delayed responses of vegetation phenology to rising temperature and their mechanisms are poorly understood. Using satellite and climate data from 1982 to 2005, we reveal a latitude transition zone of greenup onset in vegetation communities that has diversely responded to warming temperature in North America. From 40°N northwards, a winter chilling requirement for vegetation dormancy release is far exceeded and the decrease in chilling days by warming winter temperature has little impact on thermal-time requirements for greenup onset. Thus, warming spring temperature has constantly advanced greenup onset by 0.32 days/year. However, from 40°N southward, the shortened winter chilling days are insufficient for fulfilling vegetation chilling requirement, so that the thermal-time requirement for greenup onset during spring increases gradually. Consequently, vegetation greenup onset changes progressively from an early trend (north region) to a later trend (south region) along the latitude transition zone from 40–31°N, where the switch occurs around 35°N. The greenup onset is delayed by 0.15 days/year below 31°N.

1. Introduction

[2] Global surface temperature, which has increased by 0.6°C in the past three decades [Hansen et al., 2006], has altered vegetation phenological shifts. Numerous field species-specific measurements have shown the advances in timing of both flowering and budburst in northern mid-high latitudes [e.g., Schwartz, 1998; Menzel and Fabian, 1999; Kramer et al., 2000; Penñuelas and Filella, 2001]. Similarly, long time series of satellite data have detected the acceleration in the greenness of canopy in vegetation communities between 45°N and 70°N [Myneni et al., 1997; Zhou et al., 2001]. These accelerations in spring phenology have been widely interpreted as the vegetation responses to rising temperature. However, field observations have also demonstrated that spring vegetation phenological events have delayed in a certain amount of vegetation species across Europe and North America during past decades [Bradley et al., 1999; Fitter and Fitter, 2002; Menzel and Fabian, 1999]. For vegetation communities at regional scales, long-term spring vegetation events retrieved from the station-climate-based Bowen ratio along the United States (US) East Coast display delayed shifts in southern regions of Virginia and Carolinas (around 35°N) [Fitzjarrald et al., 2001]. Although the delayed trends in some vegetation species are likely associated with elevated CO2 and N in field-controlled experiments [Cleland et al., 2006], little is known about the controls of rising temperature on the reduction of advance rate and the delay of vegetation greenup onset at continental scales.

[3] We investigated the diverse responses of vegetation phenology to warming climates and the related mechanisms. Specifically, we detected the onset of vegetation greenup using satellite data during last 24 years in North America (NA). We further investigated the spatial patterns in interannual trends of greenup onset and verified the trends using field-based lilac first bloom dates. Finally, we linked the interannual phenological trends to daily land surface temperature to reveal the mechanisms by which phenology diversely responds to warming climates.

2. Methodology

2.1. Phenological Detection From Satellite Data

[4] We collected the Normalized Difference Vegetation Index (NDVI) derived from AVHRR (Advanced Very High Resolution Radiometer) between 1982 and 2005 in North America to examine the long time series of vegetation phenology. The NDVI in the Global Vegetation Index (GVIx) product was produced by NOAA NESDIS (National Environmental Satellite, Data, and Information Service) with a temporal resolution of seven days (weekly composite) and a spatial resolution of 0.036° (about 4 km at equator) [Kogan et al., 2003; Donahue et al., 2005]. In this dataset, the impacts of the AVHRR sensor drift on the red band and near infrared band were calibrated using the updated post-launch coefficients [Wu, 2004; Gutman et al., 1995].

[5] We detected vegetation phenology from each annual time series of NDVI data after removing cloud and snow contaminations. Specifically, we first identified the pixels with snow or cloud contaminations using information in the quality assurance flag of the GVIx product. These contaminated data were replaced using a moving-window average based on the two nearest neighbours with valid data [Zhang et al., 2006]. Second, the snow contamination was further minimized because of uncertainty of the snow flag. To do this, we calculated a time series of annual minimum NDVI values for each pixel from 1982–2005. The median value in the minimum NDVI series in a pixel was then extracted as the representative of background NDVI value which was used to replace the smaller values in an annual time series. Third, the annual time series of NDVI data were fitted against the day of year (DOY) using piecewise logistic functions [Zhang et al., 2003]. Finally, the functions were applied to automatically detect the phenological transition date in the onset of vegetation greenup [Zhang et al., 2003, 2006]. Note that this algorithm treating each year's NDVI data individually overcomes residual errors in the satellite inter-calibrations.

[6] The trend of interannual variation in greenup onset was calculated using a simple linear regression slope for each pixel. The regression analysis was conducted using the time series of greenup onset dates after excluding the maximum and the minimum values to minimize the uncertainty in the detected phenological dates. Further, zonal averages for the phenological trends were computed across 0.25° latitude bands in the humid climate regions after arid and semiarid regions (mainly distributed at western US) were excluded using the Köppen climate classification map produced by the Food and Agriculture Organization (FAO) and croplands were removed using land cover data generated using the MODIS (MODerate Resolution Imaging Spectroradiometer) data [Friedl et al., 2002]. Note that croplands might change slightly during 1982–2005.

2.2. Field Observations of Lilac First Bloom Dates

[7] The field-measured lilac data were used to verify the satellite-based phenological trends. We collected lilac (Syringa chinensis and Syringa vulgaris) first leaf and first bloom dates observed in fields across the United States [Schwartz and Caprio, 2003]. This spatially-distributed lilac dataset includes more than 1100 locations with different observation time periods starting as early as 1950's.

[8] Interannual variations in lilac first bloom dates were calculated. To approximately match the period of AVHRR data, the lilac observations that stopped before 1980 were not taken into account. In addition, lilac first leaf was not investigated here because measurements were missed in most of the years. Thus, lilac first bloom dates in the remaining locations were used to calculate the interannual variation trends using a linear regression method. The interannual trends were then averaged for each 0.5° latitude zone to determine their dependences on latitude.

2.3. Land Surface Temperature

[9] Land surface temperature (LST) data were acquired to investigate the responses of spring phenology to temperature. We obtained 3-hourly LST data at a spatial resolution of 32 km between 1981 and 2005 from the NCEP (National Centers for Environmental Prediction) North America Regional Reanalysis (NARR) [Mesinger et al., 2006]. This dataset was produced using a fixed assimilation/forecast model and is the most accurate and consistent long-time series of dataset that covers the entire NA [Mesinger et al., 2006].

[10] The interannual variations in LST were computed. The 3-hourly land surface temperature values were first averaged to generate daily data. We then calculated winter chilling days (number of days with daily temperature ≤5°C) from August to next July and thermal-time requirement (TTR, degree days with temperature >5°C from January 1 to the greenup onset) for the occurrences of vegetation greenup. Using the same procedure (simple linear regression) as used in processing vegetation greenup onset data, we also computed the trends of interannual variations in chilling days and thermal-time requirements for each pixel and the zonal average across NA.

2.4. Thermal Time-Chilling Model

[11] To investigate how warming temperature might influence dates of vegetation greenup at a continental scale, we employed a thermal time-chilling model. The mechanism of this model suggests that a critical duration of winter chilling should be required to release vegetation bud dormancy and a critical spring thermal time (temperature summation above a base value) to initiate spring greenup [Sarvas, 1974; Cannell and Smith, 1983; Chuine and Cour, 1999]. If the buds are inadequately chilled during winter, they will remain partially dormant in spring and require a large thermal time to trigger budburst [Murray et al., 1989]. In other words, vegetation is assumed to respond to decreased duration of previous chilling by increasing the requirements of temperature forcing to trigger spring greenup. The model is expressed as an exponential function:
equation image
where TTR is the thermal time requirement (degree days), Cdis the number of chilling days, and α, β and γ are coefficients.

[12] The thermal time-chilling models were generated using zonal average data for each individual year and long-term average, separately. The maxima curvature change rate along each thermal time-chilling curve was calculated to determine the turning point from fully to insufficiently winter-chilled regions. The geo-locations of turning points were determined by associating with latitude using the linear relationship between chilling day and latitude in a local area.

3. Results

3.1. Spatial and Temporal Trend in Greenup Onset

[13] The spatial patterns indicate that the onset of greenup occurs in early March around 30°N, pushes northwards gradually, and reaches tundra regions (65–70°N) by June. The northward shift exhibits an average gradient of 3.4 ± 0.34 days per degree of latitude in natural vegetation (excluding croplands and semiarid regions). As expected, this dependence in greenup onset on latitude is consistent interannually although the magnitude varies slightly.

[14] Figure 1 displays the phenological trends of interannual variations during 1982–2005. The greenup onset has advanced in 70% of areas while it has delayed in 30%, which are the results calculated using Lambert Azimuthal Equal-Area Projection. The change rate in 80% of the region ranges from −0.7 (negative value for advance in greenup onset) to 0.3 days/year (positive value for delay).

Details are in the caption following the image
Interannual change rate in greenup onset from 1982 to 2005, where negative value represents an advance trend while positive value denotes a delay trend. The white color indicates the areas without good data or no changes.

3.2. Phenological Transition Zone

[15] The zonal average in interannual change rate over NA (Figure 1) for natural vegetation reveals a significant spatial pattern in delay and advance trends of greenup onset. The satellite-based zonal phenological trend exhibits three distinctive regions (Figure 2a). First, the greenup onset has advanced with a relatively constant rate of 0.32 ± 0.07 days/year (mainly ranging 0.2–0.4 days/year) between 40°N and 70°N. This trend is qualitatively identical to the previous findings from both satellite data and field measurements but the magnitude is slightly different from previous studies [Fitter and Fitter, 2002; Zhou et al., 2001]. Second, the change rate varies linearly from −0.43 to 0.24 days/year between 40°N and 31.5°N, which reflects a phenological transition zone where vegetation greenup onset switches from an advance trend to a delay trend gradually. The switching occurs around 35.5°N. Third, the greenup onset has been delayed in 31.5°N southwards with an average rate of 0.15 ± 0.05 days/year.

Details are in the caption following the image
(a) Zonal variation in the interannual change rate of phenology (day/year) with latitude in North America. The solid brown curve is satellite-based greenup onset; the green asterisks are field-based lilac first bloom; and the linear green line is the fitted line between latitude and lilac first bloom. (b) The thermal time-chilling model (the solid curve) established from zonal average data (the circles) from 1982–2005. The dashed line is latitude corresponding to different chilling days. The solid green dot (identified using maxima curvature change rate) represents the turning point which separates the fully and insufficiently winter chilled regions for greenup onset. TTR is the thermal time requirement (degree days) and Cdis the number of chilling days.

[16] This phenological transition zone with diverse trends is supported by spring events derived from field species-specific observations. The lilac first bloom dates across the US, although not the same as the vegetation community-level greenup onset, have generally advanced in 35.5°N northwards while delayed southwards (Figure 2a). The interannual change rate varies significantly along latitude (P = 0.025) in the phenological transition zone. The fitted linear line of change rate against latitude matches very well with the variation in satellite-based phenological trends (Figure 2a).

3.3. Variation in Thermal Time-Chilling Curve

[17] The thermal time-chilling model explicitly describes the impacts of temperature on greenup onset. The average data of both satellite-based greenup onset and daily land surface temperature from 1982–2005 characterize the nature in the responses of greenup onset to both winter chilling fulfilment and TTR (Figure 2b). The thermal time-chilling curve demonstrates that TTR decreases rapidly with increasing chilling days and reaches an asymptotic value when the number of chilling days is longer than 135 days, which is identified using the maxima curvature-change rate along the curve and is referred as to the turning point (the green dot in Figure 2b). This identified chilling day corresponds to the latitude of 40.75°N (Figure 2b) which is coincident with the northern starting point of the phenological transition zone (Figure 2a). In the northern region, the number of chilling days is generally longer than 135 days and linearly increases along latitude (5.4 days per latitude degree, P < 0.0001) and TTR remains relatively stable with a value of 62.8 ± 11.6°C. This suggests that the number of chilling days far exceeds the chilling requirement so that TTR is reached quickly in warming climates to trigger the onset of greenup. In contrast, the chilling days are insufficient for vegetation dormancy release around 40°N southwards and is reduced rapidly along latitude (14.8 days per latitude degree, P < 0.0001). Consequently, TTR for vegetation budburst increases exponentially southwards with a rate of about 41.6 degree days per latitude degree.

[18] The analysis of daily temperature presents a trend of interannual variation in both chilling day and thermal time requirement (TTR) from 1982–2005 (Figure 3). The chilling day decreases in eastern NA with an average of 0.441 ± 0.243 days/year which is larger in northern areas than in southern regions although an increased trend appears in north-western regions (Figure 3a). On the other hand, TTR varies slightly without constant interannual trends in mid-high latitudes (Figure 3b). This result evidently suggests that variation in chilling days caused by warming temperature has limited impacts on the critical chilling requirement and has little effects on TTR in 40°N northwards. Consequently, increase in spring temperature accelerates the reach of TTR and hastens greenup onset becoming earlier as indicated in Figure 2a. Paradoxically, the increased rate of TTR dominates the eastern US from 40°N southwards, which is 4.65 ± 4.55 degree days per year in the region of 30–35°N except for a small region along the coast (Figure 3b). Evidently, the reduction in winter chilling days has accelerated the increase of the thermal time required for greenup onset. This causes the reduction of the advance rate in greenup onset and a delay trend in the phenological transition zone (Figure 2a).

Details are in the caption following the image
Interannual variation in LST from 1982 to 2005. (a) The change rate in chilling days; (b) cumulative temperature in degree days (°C) required for greenup onset. The white color indicates the areas without good data or no changes. The greenup onset in southwest is controlled by water forcing because of the dry climate, so that TTR for greenup onset is not discussed here. The agriculture region with the strong increase of TTR in the central US is influenced by human activities and is also excluded in this context.

[19] The thermal time-chilling curves for individual years provide a time series of turning points. By associating the turning points with latitude, we find that the turning point has shifted northwards with a rate of 0.1 (P = 0.07) latitude degrees per year (Figure 4). This indicates the northward shift of the phenological transition zone because the turning point separates the regions with fully filled and insufficient winter chilling days and corresponds to the northern start point of the transition zone as described in Figure 2b.

Details are in the caption following the image
Interannual variation in the turning point representing northern starting point of the transition zone from 1982–2005. The turning point (related to latitude) is identified using a maximum curvature change rate along the thermal time-chilling curve which is generated for each year individually.

4. Discussion and Conclusions

[20] The results in this study suggest that warming temperature advances greenup onset in mid-high latitudes but it delays the trends in low middle latitudes across North America. The requirement of thermal time is increased (by 4.65 degree days per year from 30–35°N) for the onset of vegetation greenup from 40°N southward because the number of winter chilling days which is shortened by 0.44 days/year becomes insufficient to fulfill vegetation chilling requirement. As a result, the greenup onset in the latitude transition zone between 40°N and 31°N has gradually switched from an advance trend of 0.32 days/year in 40°N northwards to a delay trend of 0.15 days/year below 31°N. The change from an earlier trend to a later trend occurs around 35°N. These phenological characteristics are well verified using long time series of field-measured lilac data. Moreover, the northern starting point of the phenological transition zone is also accurately reflected in the thermal time-chilling curve.

[21] The phenological transition zone has shifted northwards with a rate of 0.1 latitude degree per year. This pattern is likely to continue if the global surface temperature increases 0.75°C by 2020 (above year 2000 level) and at least 2–3°C by 2100 with the continuous rising of atmospheric CO2 and other greenhouse gases [Hansen et al., 1988, 2006]. Thus the greenup onset will continuously advance with climate warming in high latitudes because vegetation buds are always fully chilled. However, because TTR increases with the reductions of chilling days, the advance of greenup onset will slow down around 40°N and the timing of greenup onset between 35°N and 40°N may gradually change from advance to delay trends.


[22] We wish to thank Xiwu Zhan for comments, Wei Gou for AVHRR data, and Mark Schwartz for lilac data. The views, opinions, and findings contained in these works are these of the author(s) and should not be interpreted as an official NOAA or U.S. Government position, policy, or decision.