Abstract
Potential climate drivers of Arctic tundra vegetation productivity are investigated to understand recent greening and browning trends documented by maximum normalized difference vegetation index (NDVI) (MaxNDVI) and time-integrated NDVI (TI-NDVI) for 1982–2015. Over this period, summer sea ice has continued to decline while oceanic heat content has increased. The increases in summer warmth index (SWI) and NDVI have not been uniform over the satellite record. SWI increased from 1982 to the mid-1990s and remained relatively flat from 1998 onwards until a recent upturn. While MaxNDVI displays positive trends from 1982–2015, TI-NDVI increased from 1982 until 2001 and has declined since. The data for the first and second halves of the record were analyzed and compared spatially for changing trends with a focus on the growing season. Negative trends for MaxNDVI and TI-NDVI were more common during 1999–2015 compared to 1982–1998.
Trend analysis within the growing season reveals that sea ice decline was larger in spring for the 1982–1998 period compared to 1999–2015, while fall sea ice decline was larger in the later period. Land surface temperature trends for the 1982–1998 growing season are positive and for 1999–2015 are positive in May–June but weakly negative in July–August. Spring biweekly NDVI trends are positive and significant for 1982–1998, consistent with increasing open water and increased available warmth in spring. MaxNDVI trends for 1999–2015 display significant negative trends in May and the first half of June.
Numerous possible drivers of early growing season NDVI decline coincident with warming temperatures are discussed, including increased standing water, delayed spring snow-melt, winter thaw events, and early snow melt followed by freezing temperatures. Further research is needed to robustly identify drivers of the spring NDVI decline.
1. Introduction
Three decades of remotely sensed normalized difference vegetation index (NDVI) data document an overall increase in Arctic tundra vegetation greenness (Myneni et al 1997, Jia et al 2003, Walker et al 2009, Xu et al 2013, Ju and Masek 2016) but the trends display considerable spatial variability. NDVI represents vegetation productivity as measured by aboveground biomass (Tucker and Sellers 1986, Shippert et al 1995, Stow et al 2004) and has a strong relationship with biomass throughout the panarctic tundra (Epstein et al 2012). Panarctic tundra vegetation greening is associated with increases in summer warmth (Jia et al 2003, Hope et al 2005) that are, in large-part, driven by summer sea ice retreat along Arctic coasts (Bekryaev et al 2010, Bhatt et al 2010, Dutrieux et al 2012). Climate variability over the Arctic Ocean plays an important role in tundra vegetation productivity because so much of the Arctic tundra biome is located near Arctic Ocean coastlines (Walker et al 2005).
Trends covering the period 1982–2015 are overall positive for summer open water, summer warmth index (SWI, the sum of the monthly mean temperatures above 0 °C from April to September), MaxNDVI (peak NDVI) and time-integrated NDVI (TI-NDVI, sum of biweekly NDVI values above 0.05 from May–September) (figure 1). Increased open water during summer (figure 1(a)) resulting from sea ice retreat along the Arctic coast is consistent with warming land-surface temperatures (figure 1(b)) and increased vegetation greenness (figure 1(c) and (d)). A more detailed examination shows that not all regions have positive trends, for example, there is an area of cooling in western Eurasia, which is broadly co-located with maxNDVI and TI-NDVI declines (figure 1). While sea ice decline has continued over the satellite record, summer land surface temperatures and vegetation productivity measures have not continued to simply increase (Epstein et al 2015) (figure 2). Time series of SWI for the panarctic, North American Arctic and Eurasian Arctic reveal large interannual variability and low-frequency variability that is offset between North America and Eurasia (figure 2(a)). The Eurasian SWI decreased from 2004–2010, while in North America SWI declined from 1998–2005, but both show SWI increases since 2010 (figure 2(a)). The MaxNDVI time series shows a steady increase over the record (figure 2(b)) while TI-NDVI trends (figure 2(c)) have flattened or even declined (e.g. Eurasia) over the last decade (Bhatt et al 2013). This suggests that likely multiple processes influence vegetation productivity beyond secular greening associated with increased summer warmth.
Figure 1 Percent change from 1982–2015 for (a) May–August open water, (b) SWI (April–September), (c) MaxNDVI, and (d) TI-NDVI. Units of percent change in all panels. Panels have been updated from Bhatt et al 2013 and Epstein et al 2016.
Download figure:
Standard image High-resolution image
Figure 2 Time series of (a) SWI (°C month), (b) MaxNDVI (unitless), and (c) TI-NDVI (unitless) for each growing season from 1982–2015. Each panel displays time series for the Arctic, Eurasia, and North America. Note that there is a different scale for the two y-axes in panel (a) where Arctic is given on the left axis while Eurasia and North America are represented on the right axis.
Download figure:
Standard image High-resolution imageTundra vegetation productivity is primarily controlled by temperature (Bliss 1997, Chernov and Matveyeva 1997, Callaghan et al 2004) where increased warmth leads to increased biomass. The heterogeneity in recent trends suggests a closer examination of moisture, which also plays a role in plant productivity. Ground-based observations along the North American (NAAT) and Eurasian (EAT) Arctic Transects (Walker et al 2012b) highlight differences in vegetation due to temperature, moisture and substrate factors. The NAAT and EAT have similar temperature regimes and plant distributions for a given bioclimate subzone (Walker et al 2005) but the EAT receives more annual precipitation and has more moss biomass in subzones B-E (Walker et al 2012b). Climate research based on global modeling consistently shows that as the climate warms the total snowfall decreases but it increases at higher latitudes and altitudes where below-freezing temperatures prevail in winter (Krasting et al 2013, Kapnick and Delworth 2013).
Increasing snow at high latitudes has not yet been clearly documented in contemporary observations for the panarctic but several recent observationally-based studies show increases in snow amounts in Eurasia (Bulygina et al 2009, Cohen et al 2012), Abisco Sweden (Kohler et al 2006), and northern Alaska (Urban and Clow 2014, Clow 2014, Cherry et al 2014, Vikhamar-Schuler et al 2010). Increased transport of moist static energy in a CO2 enhanced climate is a primary mechanisms by which the Arctic atmospheric column warming is amplified (Alexeev et al 2005). Zhang et al (2012) found that moisture transport into Eurasia increased during winter in the NCEP/NCAR Reanalysis, which is consistent with increased snow amounts. An analysis of remote sensing data sets from 2000–2010 found that a sea ice concentration decline of 1% leads to a 0.36%–0.47% increase in cloud cover (Liu et al 2012). Some of this moisture is expected to result in increased precipitation as shown in an investigation of deuterium excess measurements where sea ice decline coincides with an increase in high-latitude precipitation originating in the Arctic (Kopec et al 2016). Reduced sea ice imposed in a regional model in the Beaufort Sea leads to increased precipitable water indicating that increased moisture is available in the atmospheric column to form clouds and precipitation (Bieniek et al 2015). Therefore, once climate warming has melted a sufficient amount of sea ice, hydro-climate processes likely play a larger role as drivers of vegetation productivity than when the ice first began to decline and contributed to large local land surface warming.
The new contribution of this study is to document panarctic trends within a season over the period 1982–2015 for the GIMMS NDVI3g for Arctic tundra together with landsurface temperatures, sea ice concentration, and ocean-heat content. We focused particularly on seasonality of trends to investigate the times of year that are changing the most, and relating these to interactions between the vegetation and associated climate drivers. Recent work shows that the warming and greening trends over panarctic tundra have slowed down (Walker et al 2012a, Bhatt et al 2013, Epstein et al 2015) and are more heterogeneous. This study compares the seasonality and trends during 1982–1998 with those from 1999–2015. These periods were chosen by determining the timing of trend change (1998 +/− 7 yr) in the panarctic SWI time series based on a parametric, nonlinear regression technique called 'breakfit regression' (Mudelsee 2010 2009). Exact break points vary considerably over Arctic tundra domain (Bhatt et al 2013), however, 1998 shows up as a global temperature peak due to the strong El Niño, is midway through the data record, and is consistent with the timing of the global 'climate hiatus' (Easterling and Wehner 2009). Based on this reasoning, the analysis is split into two periods as follows: 1982–1998 and 1999–2015 to investigate variations in NDVI and its climate drivers.
2. Data and methods
2.1. Data
Special Sensor Microwave Imager (SSM/I) sea ice concentrations (Comiso and Nishio 2008) and Advanced Very High Resolution Radiometer (AVHRR) radiometric surface temperature are investigated over the 1982–2015 period. The term sea ice concentration is used in this publication to describe sea ice area in units of percentage area covered. The portion of a pixel that is not covered by sea ice is denoted as open water (i.e. 100% − sea ice concentration). Summer open water (OW) is constructed by summing up the weekly May through August open-water percentage (units are % open water area within a pixel) and then dividing by the number of weeks to produce an average weekly OW amount for the summer.
The remotely-sensed surface temperature data have been corrected through effective cloud-masking techniques and calibration through the utilization of in-situ surface temperature data. Surface temperatures from the surface heat budget in the Arctic (SHEBA) experiment conducted in the central Arctic from October 1997 through September 1998 and 2 m air temperatures from meteorological stations were used to calibrate the AVHRR data. Details of this procedure can be found in Comiso (2003). SWI was calculated as the sum of average April to September monthly surface temperatures above freezing at each pixel and is in units of °C months. Note that SWI calculated using ground surface temperatures is warmer than SWI calculated using 2 m air temperature (Raynolds et al 2008).
Remotely-sensed NASA GIMMS (Global Inventory Modeling and Mapping Studies) bi-weekly maximum NDVI data (Pinzon and Tucker 2014) from 1982–2015 are derived from AVHRR sensors on NOAA-7 through NOAA-18 satellites. The NDVI3g product corrected discontinuities in the GIMMS NDVI north of 72°N and permitted the first comprehensive analysis of NDVI trends in the High Arctic (Bhatt et al 2010). Details about the calibration process are available in Pinzon and Tucker (2014). This study used approximately 12 km resolution NDVI data to more closely match the grids used for sea ice and surface temperature. The maximum NDVI (MaxNDVI) is the highest summer NDVI value, representing peak vegetation photosynthetic capacity, and serves as an indicator of tundra biomass (Shippert et al 1995, Walker et al 2003a). The time-integrated NDVI (TI-NDVI) is the sum from May to September of biweekly values above a threshold value of 0.05, low enough to capture the often abrupt snowmelt. TI-NDVI incorporates the length of the growing season and phenological variations, better represents gross primary production than MaxNDVI (Tucker and Sellers 1986) and is better correlated with climate parameters (Bhatt et al 2010).
The Panarctic Ice Ocean Modeling and Assimilation System (PIOMAS) dataset (Zhang and Rothrock 2003, Steele et al 2011) provides biweekly ocean heat content data for 1988–2013, where sea surface temperature and sea ice concentration were assimilated into the model simulation. The heat content was calculated by vertically integrating the density of the ocean times the specific heat capacity times the ocean potential temperature minus a reference temperature of −2 °C from the surface to 100 m depth (or less depending on depth of ocean). The data were then area-averaged within a 100 km buffer of the coast of each tundra region.
The NCEP/NCAR reanalysis (Kistler et al 2001) provided sea level pressure (reanalysis) and the ERA-Interim (Dee et al 2011) 2 m air temperature for this study. Reanalyses are constructed by ingesting a variety of observations (i.e. meteorological station data, atmospheric soundings, sea surface temperature, satellite observations) into a weather forecast model to fill in data sparse regions and create a gridded 'observational' data set. These data sets have been invaluable for investigating climate variability processes over the past several decades.
2.2. Analysis methods
The scale of interest in the study is the Arctic non-alpine tundra and its continental divisions of North America and Eurasia. The analysis employs time series averaged over oceanic regions within 100 km of the Arctic coastline and over the full tundra domains at elevations < 300 m. The least-squares-fit method was used to determine the trends of open water, SWI, MaxNDVI, and TI-NDVI in the spatial presentation (figure 1). Spatial trends are shown as a magnitude change for Open Water, SWI, TI-NDVI, and MaxNDVI over the 34 years period and are based on the pixel size of the given data set. The statistical significance of correlations and trends was assessed using the two-tailed Student's t-test at the 95% or greater level. Climate data variability in the Arctic displays large-amplitude multi-decadal variability along with trends (Polyakov et al 2013) which reduces the degrees of freedom due to large autocorrelations from year-to-year. For significance testing the reduced degrees of freedom were calculated using a lag-1 autocorrelation method outlined by Santer et al (2000).
3. Results
3.1. Spatial trends
SWI trends vary regionally from 1982–1998 (figure 3(a)) with a strong warming trend over North America, particularly northern Canada, in the region between 90 and 120 °W, and a distinct cooling trend in West Siberia between 50 and 120 °E. The trend is more mixed from 1999–2015 (figure 3(b)). Trend magnitudes for SWI from 1982–1998 (first period) show increased warmth over most of the Arctic except western Eurasia, where the trends are negative (figure 3(a)). SWI for 1999–2015 (second period) displays patches of negative trends throughout Eurasia and Alaska while the Canadian High Arctic and Greenland show mainly warming trends (figure 3(b)). The overall weakening of SWI trends since 1999 suggests that additional processes are currently operating in the Arctic tundra.
Figure 3 Magnitude change trend for SWI based on AVHRR land surface temperatures for (a) 1982–1998 and (b) 1999–2015 and based on 2 m air temperature from ERA-Interim Reanalysis for (c) 1982–1998 and (d) 1999–2015. Units are in magnitude change in °C month over 17 years.
Download figure:
Standard image High-resolution imageSWI trends based on 2 m air temperature from the European Center Reanalysis (ERA-Interim) for 1982–1998 (figure 3(c)) compare favorably with AVHRR based land surface temperature trends (figure 3(a)). In contrast, SWI trends from 1999–2015 based on the ERA-Interim show general warming over the tundra region (figure 3(d)) while AVHRR-based SWI shows areas of negative trends in Eurasia (figure 3(b)) suggesting that land surface and air temperature trends are diverging in recent years. Time series of SWI based on land surface and 2 m air temperature is shown in figure 4 for North America (figure 4(a)) and Eurasia (figure 4(b)). As expected, the air temperature based SWI is cooler on average than the land surface based SWI but displays similar interannual variability (correlation is about 0.6 for both regions). The trends have a similar slope in North America but are quite different in Eurasia where the air temperature based SWI is warming faster than land surface temperature based SWI (figure 4(b)). One possible local explanation for these diverging trends is the idea that a denser plant canopy with a thicker organic layer will reduce the warming of the soil (Walker et al 2003b). This idea is supported by Circumpolar Active Layer Monitoring (CALM) network observations which show that from 1996–2013 the summer N-factors (ratio of ground surface to air temperatures) in bioclimate subzones C, D, and E have been decreasing (Shiklomanov and Streletskiy 2015). This is likely to be one of multiple processes operating that prevents simple interpretations of the long-term observational data. Finally, it is plausible that this discrepancy could be an artifact of the data, caused either by changes over time in the number and location of data sources for the ERA-interim data, or issues with calibrating a long-term satellite record (Urban et al 2013).
Figure 4 Time series of SWI (°C month) based on 2 m air temperature from the ERA-Interim reanalysis (black) and AVHRR landsurface temperature (red) for (a) North America and (b) Eurasia.
Download figure:
Standard image High-resolution imageOverall, the panarctic tundra maxNDVI trends display more pixels with negative trends in the recent period compared to the earlier period (figure 5). For MaxNDVI, trends for 1982–1998 (figure 5(a)) are positive (> 0.02) for 54% of the pixels and negative (< −0.02) for 6% of the pixels. Considering only pixels with significant (p < 0.05) trends, 14% are positive for 1982–1998. For the 1999–2015 period (figure 5(b)) trends are positive for 55% of the pixels and negative for 29% of the pixels. Significant trends are positive in 31% and negative in 14% of the pixels for 1999–2015. MaxNDVI declines are limited to the southern edges of the tundra during the period from 1982–1998 (figure 5(a)) but are found in western Eurasia, southwest Alaska, and the northern Canada in the period from 1999–2015. The positive MaxNDVI trends have become stronger in the 1999–2015 period compared to the earlier period.
Figure 5 Magnitude change trend for MaxNDVI for (a) 1982–1998 and (b) 1999–2015. Magnitude change trend for TI-NDVI for (c) 1982–1998 and (d) 1999–2015. Units are in magnitude change over 17 years for all panels.
Download figure:
Standard image High-resolution imageTI-NDVI magnitude trends are generally positive from 1982–1998 and largely negative for the 1999–2015 period with consistent negative trends in Eurasia in agreement with figure 2(c). Over the period 1982–1998 (figure 5(c)), 81% of the pixels have positive trends (> 0.05) and 5% have negative trends (< −0.05). Considering only pixels with significant (p < 0.05) trends, 17% are positive for 1982–1998. From 1999–2015 (figure 5(d)), 25% of the pixels have a positive trend while 67% have a negative trend. Significant trends for 1999–2015 are positive in 7% and negative in 34% of the pixels. Variations within a season are discussed next to understand the timing of TI-NDVI declines and possible climate drivers of these declines.
3.2. Changing seasonality
Sea-ice concentration in the 100 km Arctic coastal zone has decreased more than 10% from climatology (1982–2015) between late May and November (figure 6(a)). Figure 6 displays weekly sea ice concentration climatology (blue bars) and trends (grey bars) for the full period 1982–2015 (figure 6(a)), the early period of 1982–1998 (figure 6(b)) and the recent period from 1999–2015 (figure 6(c)). From 1982–1998, the sea ice declines were larger (> 10% per 17 years) during spring and smaller in fall. Negative trends were larger (> 10% per 17 years) during fall for 1999–2015 than for 1982–1998. This analysis shows that over the satellite record, the sea ice concentration decreased first in spring followed by large declines in the fall. This is consistent with warmer atmospheric temperatures driving earlier spring breakup followed by an extended open-water season, increased heat storage in the ocean and a delayed freeze up in the fall.
Figure 6 Arctic weekly sea ice climatology (blue bars) and trend (grey bars) for 100 km zone for (a) 1982–2015, (b) 1982–1998, and (c) 1999–2015. Weekly trends statistically significant at the 95% level or greater are marked with an asterisk on the x-axis. Sea ice is in units of percent concentration and trend is percent concentration over time period of panel.
Download figure:
Standard image High-resolution imageOceanic heat content in the 100 km coastal zone has increased in the Arctic (figure 7(a)) throughout the year and peak positive trends occur in late September over the 1988–2013 period. Figure 7 displays biweekly oceanic heat content climatology (orange bars) and trend (grey bars) for the period 1988–2013 in the Arctic (figure 7(a)), North America (figure 7(b)) and Eurasia (figure 7(c)). The heat content reaches its annual maximum value in early September for North American and Eurasia but the peak positive trends are in late August in North America and in late September in Eurasia. The seasonal cycle of Eurasian fall sea ice freezeup is later than in North America and the sea ice decline is larger in fall (not shown). This is consistent with the later oceanic heat-content trend peak in Eurasia compared to North America. The ice-reduced warm ocean is a source of heat and moisture for the Arctic atmosphere in the autumn. Sea ice reduction is considered to be the underlying cause of near-surface Arctic atmospheric warming during fall (Screen and Simmonds 2010) and heat provided by the ocean further supports this idea. The positive trends in heat content are late in the growing season and could potentially impact vegetation productivity the following year by hastening spring sea ice melt. September ocean heat content in the 100 km coastal zone is correlated with sea ice concentration the following spring at −0.56 (same season correlation is −0.78), suggesting that there is memory in the system that may provide a possible link between the coastal ocean and terrestrial regions.
Figure 7 Arctic biweekly ocean heat content (orange bars) and trend (black bars) for 100 km zone from 1988–2013 for (a) Arctic, (b) North America, and (c) Eurasia. Biweekly trends statistically significant at the 95% level or greater are marked with an asterisk on the x-axis. Heat content is in units of MJ m−2 and trend is MJ m−2 over 26 years (1988–2013).
Download figure:
Standard image High-resolution image
Figure 8 Arctic weekly land surface temperature climatology (red bars) and trend (grey bars) for the full tundra for (a) 1982–2015, (b) 1982–1998, and (c) 1999–2015. Weekly trends statistically significant at the 95% level or greater are marked with an asterisk on the x-axis. Temperature is in units of °C and trend is °C over time period of panel.
Download figure:
Standard image High-resolution imageSummer low elevation (< 300 m) tundra land surface temperature increases have slowed down in recent years (figure 8). Figure 8 displays land surface temperature climatology (red bars) and trends (grey bars) for the full period 1982–2015 (figure 8(a)), the early period of 1982–1998 (figure 8(b)) and the recent period from 1999–2015 (figure 8(c)). From 1982–1998, there was strong warming in late winter–early summer (Mar–mid June) and in Oct–Nov. The late winter trends switched to strong negative values over the period 1999–2015. During 1982–1998, land surface temperatures display generally positive trends throughout the year with positive trends in May and November (figure 8(b)). Trends from 1999–2015 display modest positive values in May and June and weak negative trends from mid-July to the end of summer (figure 8(c)). These panarctic summer cooling trends are evident in individual panarctic tundra regions except those around Greenland (Bhatt et al 2013). The recent declines in land surface temperature may reflect local processes, as was shown in the idealized model study by Bieniek et al 2015, but likely primarily result from the global climate hiatus that coincides with the 1999–2015 period and is not expected to continue much longer (Roberts et al 2015).
Figure 9 Arctic biweekly NDVI (green bars) and trend (grey bars) for full tundra for (a) 1982–2015, (b) 1982–1998, and (c) 1999–2015. Biweekly trends statistically significant at the 95% level or greater are marked with an asterisk on the x-axis. MaxNDVI is unitless and trend is change over time period.
Download figure:
Standard image High-resolution imageThe 1982–2015 biweekly NDVI displays large positive trends during the peak of the growing season and weak negative trends in May and late September (figure 9(a)). Figure 9 displays biweekly NDVI climatology (green bars) and trends (grey bars) for the full period 1982–2015 (figure 9(a)), the early period of 1982–1998 (figure 9(b)) and the recent period from 1999–2015 (figure 9(c)). NDVI biweekly trends are strikingly different between the 1982–1998 and 1999–2015 period. During the 1982–1998 period (figure 9(b)), the NDVI increases were largest in spring indicating an earlier greenup, which is consistent with spring warming (figure 8(b)). NDVI trends for 1999–2015 show declines in the early and late parts of the growing season and positive trends during the peak season (figure 9(c)). Spring temperatures trends are positive during the 1999–2015 period, so a lack of warmth cannot explain negative NDVI trends.
Figure 10 Magnitude change for sea level pressure (slp) for (a) 1982–1998 and (b) 1999–2015. Data are from the NCEP/NCAR Reanalysis (Kistler et al 2001). Panel (a) has units of hPa and panel (b) has units of hPa per 17 years.
Download figure:
Standard image High-resolution image
The initial story of sea ice decline leading to increased summer warmth and vegetation productivity (Bhatt et al 2010) now needs additional processes to explain the trends and variations evident in the recent data. The seasonality and time-series analysis indicate that since 1999 summer land surface temperatures have decreased in some parts of the Arctic, and biweekly spring NDVI has declined leading to declines since about 2001 in TI-NDVI. We next discuss how to understand the vegetation productivity trends in the context of ecological and climate drivers.
4. Discussion
4.1. Possible ecological drivers of the decline in spring-time NDVI
Panarctic trends of decreasing NDVI in May and June during 1999–2015 (figure 9(c)) represent the largest contribution to the TI-NDVI declines since 1999. This indicates that early-growing-season processes are likely causing the NDVI declines. Several possibilities exist for the decline in springtime NDVI values.
4.1.1. Deeper, later-melting snowpack
The role of the timing of snow melt needs to be considered here because changes in the timing of snow melt strongly impact plant phenology (Zeng and Jia 2013) and especially the timing of greenup (Walker et al 1993 1999, Wahren et al 2005, Borner et al 2008, Torp 2010, Callaghan et al 2012, Bienau et al 2014, Bjerke et al 2014, Bjorkman et al 2015). A logical cause of the observed decline in May–June NDVI would be that increased winter snowfall may be leading to later snow melt causing delayed phenology and decreased early-season NDVI (Walker et al 1993, Ellebjerg et al 2008). However, the June snow-cover extent (SCE) for the northern hemisphere has been declining at increasingly rapid rates since 2005, and was −17% per decade for the period 1979–2015 (Derksen et al 2015). Thus, the springtime negative trend in NDVI is striking but not easily explained in light of reported sharply declining snow cover in North America and Eurasia in April, May, and June.
Bieniek et al (2015) document increases in Alaska springtime snow-water equivalent based on the GlobSnow reanalysis data (Takala et al 2011), which provides an inferred explanation for springtime MaxNDVI declines although the exact timing of snowmelt is not available in this data set. For this study we tried unsuccessfully to extend the GlobSnow conclusions for Alaska to the panarctic domain and since the results were fairly noisy they were not included. Long-term data sets able to identify the timing of snow melt specifically for the panarctic region are needed to test the hypothesis that snow is responsible for the declines in spring NDVI.
We suspect that major differences in the timing of melt are occurring in the warmer (early-melt) and colder (later-melt) parts of the Arctic, but we are aware of only one detailed analysis of station temperature, precipitation, and snow data along a coherent south-north transect in the Arctic. Vikhamar-Schuler et al (2010) investigated long-term climate data from four stations (Nadym, Salekhard, Tarko-Sale, and Mare-Sale) in the Yamalo-Nenets Autonomous Okrug, Russia. Reindeer herders, who migrate 1500 km between the Nadym region and Mare-Sale, are concerned about the increasing occurrence of severe weather, including rain-on-snow events that affect their reindeer's access to forage (Forbes et al 2016). The analysis showed major differences in the spring melt in the three southern stations (Nadym, Salekhard, Tarko-Sale) compared to the northernmost station at Mare-Sale. All the stations experienced warming from the periods 1961–1990 and 1979–2008 (overlapping 30 year period Salekhard experienced a 33% increase in precipitation between 1900 and 2008. Mare-Sale and Salekhard showed similar trends in snow cover between the 1930s up to the early 1980s. Then Salekhard experienced a considerable and persistent reduction in the snow season, mainly because of earlier melting in the spring, while the snow season at Mare-Sale after a short drop was prolonged. The reason for this difference is attributed by the authors to a combination of differences in the effect of the springtime temperature increase and differences in the precipitation patterns. This single example should not be used to infer similar circumpolar changes along south–north climate gradients, but it does suggest that more attention needs to be directed at the changes in the timing of spring snowmelt across the Arctic tundra climate gradient.
4.1.2. Increased water on the tundra
Recent studies have shown an increase in the amount of standing water in areas of ice-rich permafrost, throughout the Arctic, due to permafrost degradation (Liljedahl et al 2016, Jorgenson et al 2006, Farquharson et al 2016). Much of this water is in the form of small thermokarst ponds several meters to tens of meters in diameter, which are formed as the upper surface of the ice wedges melt and subside. For example, widespread and rapid thermokarst with extensive ponded freshwater occurred during the last decade at High Arctic sites on Ellef Ringnes, Prince Patrick, and Banks Island, Canada. A warming summer climate caused warmer ground temperature, which increased active layer depth, ice-wedge melt, subsequent ground subsidence, and ponding (Farquharson et al 2016). NDVI is sensitive to soil moisture and water, both of which decrease NDVI values and can affect trends (Raynolds and Walker 2016). Any increases in snow will also cause an increase in meltwater in the early part of the growing season, as there is little opportunity for surface water to be absorbed in the ground because of shallow thaw. Standing water on the surface will contribute to a cooler land surface temperature but we currently have no evidence that this effect is stronger in Eurasia than North America, which is what we would expect to be consistent with the SWI trends shown in figure 4 (i.e. weak land surface temperature based SWI trends in Eurasia).
4.1.3. Killing frosts due to midwinter or early-spring snow melt
Other possible causes of decreasing spring NDVI include early season die-off of vegetation caused by mid-winter thaws exposing the tundra to extreme cold during winter (Bjerke et al 2014), hard frosts in spring after the tundra has become snow free (Inouye 2008) and midwinter rain-on-snow events (Forbes et al 2016). The combined contribution of multiple anomalous winter and spring weather events and outbreaks of moths contributed a to a major decline in NDVI over a large area of maritime northern Norway in 2011–2012 in northern Norway (Bjerke et al 2014). We have also noted extensive browning events in northern Alaska in association with late frosts in 2015 and 2016, but these local events are difficult to invoke for the nearly circumpolar decline in TI-NDVI that we see since 1999.
4.1.4. Increased role of shrubs in spring browning
The shift in tundra vegetation from graminoid to deciduous shrubs may also contribute to the biweekly NDVI trends that are seen over the 1999–2015 period. Sweet et al (2014) documented a later budburst for tall deciduous vegetation compared to graminoid vegetation on the North Slope of Alaska because snow melt was delayed around the shrubs. In a subsequent study, Sweet et al (2015) found that shrubs extended the peak period without extending the growing season length. The continued increase during the peak with declines in spring of biweekly NDVI during 1999–2015 is consistent with increased deciduous shrub cover. This may be a valid explanation locally and regionally where shrub expansion has occurred, but it cannot explain the very widespread large-scale NDVI declines found in the GIMMS3g remote sensing record extending into the High Arctic where erect deciduous shrubs are absent.
4.2. Global climate drivers of Arctic trends
In addition to Arctic sea ice decline and changing cloud cover as climate drivers of tundra vegetation productivity, it is important to consider climate variations from lower latitudes, which transport moisture and heat poleward. For example, Ding et al (2014) analyzed annual mean climate observations and model data to argue that tropical forcing is responsible for the anomalous high over Greenland. Mean summer (June–August) sea-level pressure from the NCEP/NCAR Reanalysis is shown in figure 10(a) (update from figure 8(b) of Bhatt et al 2013) and is characterized by a high over the Beaufort Sea, low pressure over Eurasia and high pressure in North America. The sea-level-pressure (slp) difference pattern (1999–2015 minus 1982–1998) (figure 10(b)) displays a clear large-scale pattern showing that since 1999 slp has increased around Greenland and decreased over Eurasia and western North America. During summer, lower slp typically indicates increased cloudiness and cooling while higher slp indicates clearer skies and warming. These recent decreases of slp over the tundra region are consistent with generally lower SWI values. This suggests that the large-scale circulation likely contributes to the recent decline of summer warmth over most of the Arctic tundra.
As the global climate has warmed and the hydrological cycle has intensified (Huntington and Billmire 2014), atmospheric moisture (Del Genio et al 1991) and poleward moisture transport have increased (Held and Soden 2006). How the increased global hydrological cycle is manifest in the Arctic is not well documented particularly during the growing season but Zhang et al (2012) have shown that winter time moisture transport into Eurasia has increased. Moisture effects on tundra vegetation productivity are receiving increased research efforts. Delidjakova et al (2016) found that increased onshore winds from Hudson Bay advected cool moist air over the tundra and resulted in increased vegetation productivity. Where climatological moisture levels are higher, such as in Western Eurasia compared to North America, biomass, particularly moss biomass, is larger in the moisture-rich areas for the same bioclimate subzone (Walker et al 2012b). Most of the atmospheric moisture available for precipitation in the Arctic is transported from lower latitudes, but a recent analysis of precipitation deuterium excess data documents an increase of moisture originating from the Arctic (Kopec et al 2016). Consistent with the idea of increased moisture from Arctic sources, cloud cover over the Arctic Ocean has increased as sea ice has decreased (Liu et al 2012). There is also evidence that precipitation has increased over the Arctic during the warm season. Kokelj et al (2015) conclude that the recent increase in the number and size of active thaw slumps over the Mackenzie basin is linked to precipitation increases. Changes in moisture transport from lower latitudes and increased local moisture need to be investigated further to understand recent vegetation productivity declines to better anticipate future biomass changes in the Arctic tundra.
5. Summary and conclusions
This study employs remote sensing data in the Arctic to explore the seasonality and update recent trends of sea ice and oceanic heat content in the coastal oceans and land surface temperatures, maximum NDVI and time-integrated NDVI over the tundra. The summary for the panarctic tundra is as follows. During the 1982–2015 period, summer sea ice declined while oceanic heat content increased. Summer warmth index (SWI) increased until mid-1990s and remained flat until the last two years when it began to increase again. MaxNDVI increased over the full period while TI-NDVI declined since the early 2000s. Comparing trends for 1982–1998 to 1999–2015 reveals that negative trends were more common during the latter period for MaxNDVI and TI-NDVI.
The climatology and trends of the weekly sea ice, biweekly ocean-heat content, weekly land surface temperature and biweekly NDVI were calculated to investigate the seasonality of changes in the panarctic. Sea-ice decline was larger during spring during the 1982–1998 period compared to 1999–2015, while fall sea ice decline was larger in the later period. Oceanic heat content is at its climatological maximum in early September for the panarctic and the largest positive trends occurred within a few weeks of this maximum. land surface temperatures are above zero Celsius from late May to late August and displayed positive trends most of the year during the 1982–1998 period and displayed negative trends most of the year except late spring to early summer during 1999–2015. The biweekly NDVI reached its climatological peak in the second half of July and during the 1982–1998 period the largest positive trends occur in spring, consistent with increased vegetation productivity early in the season. Biweekly NDVI trends from 1999–2015 displayed significant negative trends in May and the first half of June, suggesting that there are processes delaying green-up. There were also significant negative trends in September during the 1999–2015 period for the panarctic biweekly NDVI.
Our current understanding does not allow us to identify the drivers responsible for the recent declines in vegetation productivity. As sea ice decline has continued, other processes such as increased cloudiness may be coming into play and reducing summer temperatures. However, the early growing season declines in NDVI coincide with warming temperatures, not cooling, so require an alternate explanation. Numerous possible climate and ecological drivers of NDVI decline from across the panarctic tundra have been discussed in the manuscript: increased standing water, delayed spring snow-melt, winter thaw events, increased shrub cover, and early snow melt followed by freezing temperatures. Each of these pathways has been documented over fairly localized areas so cannot explain the large scale of the spatial TI-NDVI trends. Since most models predict Arctic greening, it is vital to understand the drivers of the recent browning (Phoenix and Bjerke 2016). A synthesis study that jointly considers vegetation type, permafrost conditions, elevation, as well as climate factors such as temperature, heat and moisture transport, and timing of snowfall and spring snowmelt is needed to better understand recent tundra vegetation productivity declines.
Acknowledgments
This work benefitted from fruitful conversations with Ross Brown, Chris Derksen, David Robinson, Walt Meier, Gus Shaver, and Stein Rune Karlsen. This study was supported by NASA/NEESPI Land Cover Land Use Change Initiative, Grant No. NNG6GE00A, NNX09AK56G and NNX14AD90G; National Science Foundation grants ARC-0902175 and ARC-1203506; and NASA NNH16CP09C—Arctic Boreal Vulnerability Experiment.