Land surface anomalies preceding the 2010 Russian heat wave and a link to the North Atlantic oscillation

The Eurasian wheat belt (EWB) spans a region across Eastern Ukraine, Southern Russia, and Northern Kazakhstan; accounting for nearly 15% of global wheat production. We assessed land surface conditions across the EWB during the early growing season (April–May–June; AMJ) leading up to the 2010 Russian heat wave, and over a longer-term period from 2000 to 2010. A substantial reduction in early season values of the normalized difference vegetation index occurred prior to the Russian heat wave, continuing a decadal decline in early season primary production in the region. In 2010, an anomalously cold winter followed by an abrupt shift to a warmer-than-normal early growing season was consistent with a persistently negative phase of the North Atlantic oscillation (NAO). Regression analyses showed that early season vegetation productivity in the EWB is a function of both the winter (December–January–February; DJF) and AMJ phases of the NAO. Land surface anomalies preceding the heat wave were thus consistent with highly negative values of both the DJF NAO and AMJ NAO in 2010.


Introduction
The 2010 Russian heat wave was caused by an unprecedented atmospheric blocking event that persisted from early July to mid-August. Although various aspects of this blocking event have been studied (Dole et al 2011, Matsueda 2011, Sedláček et al 2011, Lau and Kim 2012, Trenberth and Fasullo 2012, comparatively little attention has been focused on conditions prior to the heat wave; with the exception of Miralles et al (2014) who found soil-moisture deficits across most of Southwestern Russia at the heat wave's onset.
In fact, Russia experienced atypical weather for more than six months leading up to the heat wave. First, the winter of 2009/2010 was unusually cold (figure 1); attributable to an anomalously negative phase of the North Atlantic oscillation (NAO, Osborn 2011)-the lowest recorded value in a 60 year NOAA record (figure 2(a)). The severe winter of 2009/2010 was then followed by a warmer-than-normal early growing season. In May of 2010, positive land surface temperature anomalies began to emerge in Russia (figure 1). This temperature reversal coincided with persistence of the anomalously negative NAO through the early season, April-May-June (figure 2(b)). By June, the extent of land surface temperature anomalies South of Moscow was comparable to the height of the heat wave in August (figure 1).
In this study, we examined the unusually warm early season leading up to the Russian heat wave, but not the heat wave itself. Using satellite imagery from the moderate resolution imaging spectroradiometer (MODIS), aboard the AQUA and TERRA satellites, we assessed land surface conditions immediately prior to the heat wave, and over a longer term period from 2000 to 2010. To explain land surface anomalies preceding the Russian heat wave, and longerterm regional trends, we focused on statistical teleconnections involving both winter and early season phases of the NAO.
Our study area encompassed a region we term the Eurasian wheat belt (EWB; see outline in figure 1). The EWB accounts for nearly 15% of global wheat production (US Department of Agriculture (USDA) Foreign Agricultural Service 2010) and is projected to make an increasing contribution to global food security (Fischer et al 2005, World Bank 2009, Lioubimtseva and Henebry 2012. However, climate change has contributed to an estimated 3-5-fold increase in the probability of heat waves as severe as the 2010 event Coumou 2011, Otto et al 2012). Compared globally; Lobell et al (2011) found that reductions in agricultural productivity attributable to global warming were most severe in Russia. It is thus critically important to identify those broad-scale atmospheric processes that influence the EWB's climatic vulnerability.

Study area
The EWB spans an arc of fertile soils across Southern Russian, Northern Kazakhstan, and nearly all of Ukraine. We delineated boundaries of the EWB using the United Nations Food and Agricultural Organization World Soil Map (United Nations Food and Agriculture Organization (FAO) 2003). The Western and central portions of the EWB coincide with the 'Black-Earth Region' North of the Black Sea, where darkcolored mollisols (Chernozems) are well-suited to small-grain cultivation. The Eastern EWB is characterized by related mollisols-Kastanozems and Phaeozems. We defined the Northern boundary of the EWB based on transitional soils (Greyzems and Luvisols) occupying the ecotone between steppe to the South and temperate forest to the North (United Nations Food and Agriculture Organization (FAO) 2003).

Data
We assessed vegetation productivity using a remotely-sensed index sensitive to vegetated landscapes, the normalized difference vegetation index (NDVI). The NDVI exploits differences in spectral reflectance characteristic of green vegetation, i.e., low reflectance in the visible red wavelength (RED) and high reflectance in the near-infrared (NIR), and is calculated as: NDVI = (NIR − RED)/(NIR + RED). We calculated NDVI from MODIS imagery at 500 m spatial resolution using the nadir bi-directional reflectance distribution function adjusted reflectance (NBAR) product (NASA 2010b). MODIS-NBAR data is standardized to a nadir (vertical) viewing angle which for purposes of temporal analysis reduces noise associated with non-constant instrument view angle and atmospheric effects. It is distributed as 16 day rolling composites updated every eight days. To span the early growing season in the EWB, we used ten NDVI composites from mid-April to the last week in June.
Monthly values of the NAO index were obtained from the NOAA Climate Prediction Center (NOAA Climate Prediction Center 2011). This version of the NAO index is generated by rotated principal components analysis (Barnston and Livezey 1987). Empirical relationships between the NAO and surface air temperature were analyzed over the period 1951-2010 using seasonally-averaged surface (2 m) air temperature data from the NCEP/NCAR Reanalysis v1 (Kalnay et al 1996). A similar analysis with respect to precipitation was conducted using monthly precipitation data from the Global Precipitation Climatology Centre (GPCC) Reanalysis v5 (Rudolf et al 2010). In this case, GPCC data were not available for 2010. Plant water availability in the EWB was assessed using monthly values of the self-calibrated Palmer drought severity index (scPDSI) obtained from the NOAA Earth System Research Laboratory (NOAA Earth System Research Laboratory 2011, Dai 2011).

Mann-Kendall (MK) trend analysis
Geospatially referenced NDVI time series were analyzed using the MK trend test on values averaged over the early growing season. The MK trend test is a nonparametric method well-suited for identifying monotonic trends in time series that contain missing values and/or do not meet normality assumptions (Hirsch and Slack 1984). In this case, the MK test was used to assess the presence of monotonic trends over the period 2000-2010. Note that we did not estimate rates of change and make no inferences outside that period. Rather, our objective was to identify locations (500 m pixels) exhibiting directional change from 2000 to 2010 that was statistically distinguishable from random variation.

Results
The NDVI is positively correlated with the amount of photosynthetic biomass per unit area (green leaf area) and thus is an effective proxy for plant primary production (Tucker and Sellers 1986), especially in semi-arid settings like the EWB where leaf area and NDVI are strongly correlated (Fan et al 2009, Li andGuo 2012). As such, we use the terms 'NDVI' and 'primary production' synonymously. Where the unit of analysis is the NDVI, an interpretation of results is often best expressed in terms of plant primary production underlying the NDVI.
With respect to broad vegetation provinces, the EWB occupies a semi-arid transitional zone between forest and taiga to the North and the Caspian and Kazakh deserts to the South (Olson et al 2001). This transitional zone, consisting mainly of steppe and forest steppe, is reflected in a North-South declining gradient in mean early season NDVI (primary production) for our baseline period, 2000-2009 (figure 3(a)). A similar declining gradient is found from West to East across the region.
Comparison of mean early season NDVI in 2010 with the 2000-2009 baseline shows that substantial reductions in early season primary production occurred prior to the Russian heat wave (figure 3(b)). Negative land surface anomalies spanned nearly the entire EWB from central Ukraine Eastward, with highly negative anomalies concentrated in Northern Kazakhstan. Positive NDVI anomalies were concentrated in the far Western and Southwestern EWB (figure 3(b)).
Land surface anomalies in 2010 marked the continuation of a longer-term decline in early season primary production in the EWB. From 2000 to 2010, regionally-averaged NDVI dropped more than 12% (figure 4(a)). This decline coincided with a comparable drop in regionally-averaged values of the self-calibrated Palmer drought severity index (scPDSI). However, note that mean scPDSI increased slightly in 2010. In this case, regional averaging included higher scPDSI values in the Western EWB in 2010. Similarly, the modest drop in regionally-averaged NDVI from 2009 to 2010 reflects positive NDVI anomalies in the Western EWB ( figure 3(b)). In contrast with regional drying during the 2000 s, the longerterm scPDSI time series from 1951 to 2010 has no overall trend (figure 4(b)); suggesting that recent drought and its effects on regional primary productiopm may simply represent normal multi-decadal variability.
We also assessed NDVI trends from 2000 to 2010 on a spatially-explicit basis. Statistically significant (p < 0.05) declines in early season primary production occurred across most of the region (figure 3(c)) with a spatial pattern closely resembling 2010 anomalies ( figure 3(b)). Only in the far Western EWB were trends significantly positive (figure 3(c)).
Next, we considered whether the NAO could have played a role in the cold winter/warm early season pattern observed in 2010 (figure 1). The winter NAO has a well-documented effect on surface air temperatures across Western Eurasia through its influence on the Atlantic storm track and warm air advection from the North Atlantic into Northern Europe (Hurrell 1995). Figure 5(a) shows that in the EWB, positive NAOs are associated with warmer winters, negative NAOs with colder winters. This teleconnection is reversed during the early season ( figure 5(b)). Thus, negative (positive) NAOs are associated with warmer (cooler) AMJ temperatures in the EWB-consistent with a flip in temperature regime when the NAO is persistently negative from winter through the early season.
During the winter NAO's negative phase, the Atlantic storm track shifts Southward (Hurrell 1995). This influence extends into the Western EWB, where negative NAOs are associated with increased DJF precipitation, positive NAOs with drier winters ( figure 6(a)). This teleconnection is thus consistent with the positive precipitation anomalies observed in DJF 2010 ( figure 6(b)). During the early season, figure 6(c) shows that in central portions of the EWB, negative (positive) NAOs are associated with decreased (increased) AMJ precipitation. While this teleconnection is mostly consistent with negative precipitation anomalies in AMJ 2010, it does not explain negative anomalies West of approximately 50°E longitude (figure 6(d)).
Lastly, we focused on the NAO as a driver of early season primary production in the EWB. Here we used multivariate linear regression with the DJF and AMJ phases of the NAO as independent variables and mean early season NDVI as the response (figure 7). This model structure was justified by an absence of correlation between the DJF and AMJ NAO over the 60 year NOAA record (r = 0.08, p = 0.57; figure 2).
Slope coefficients for the DJF NAO were predominately positive across the region ( figure 7(a)). In the EWB, negative DJF NAOs are associated with reduced primary production during the early season, positive DJF NAOs with increased productivity. This pattern likely reflects a lagged, negative effect of colder winters when the DJF NAO is negative versus a positive effect of warmer winters during positive NAOs (e.g., figure 5(a)).
Slopes coefficients for the AMJ NAO were less uniform across the region. In the Western and Northern EWB, slope values were primarily negative ( figure 7(b)). These are the more mesic portions of the EWB, as illustrated by regional productivity gradients (figure 3(a)), where primary production is likely more limited by early season air temperatures than moisture availability. Thus, given the AMJ temperature teleconnection ( figure 5(b)), a positive (negative) effect of warmer (cooler) air temperatures is expected when the AMJ NAO is negative (positive). By contrast, across much of the Southern and Eastern two-thirds of the EWB, slope coefficients were generally positive (figure 7(b)); indicating an opposite effect of the AMJ NAO. Here, in the more arid portions of the EWB ( figure 3(a)), negative NAOs are associated with reduced AMJ primary production, positive NAOs with increased productivity. This relationship likely reflects the influence of AMJ precipitation and temperature teleconnections (figures 5(b) and 6(c)) with respect to drought, e.g., dry, warmer-than-normal early seasons during negative NAOs. This is not to say that warmer air temperatures are not expected to positively affect primary production during the initial phase of plant growth when spring soil-moisture is likely adequate. However, note that peak NDVI (leaf area) in the EWB typically occurs in mid-June (result not shown); indicating that the AMJ period encompasses a substantial proportion (if not the majority) of primary production over the entire growing season. Integrated over these three months, we expect vegetation in the Southern EWB to be more waterlimited than temperature-limited; with warmer-than-normal In each year, early season NDVI was averaged across the EWB over the 10 MODIS composites from mid-April to the end of June. Early season scPDSI was averaged regionally from monthly scPDSI values for the months of April-May-June. (b) Regionally-averaged scPDSI times series from 1951 to 2010. By Mann-Kendall trend test, this time series has no overall trend (p = 0.342).  (Kalnay et al 1996). Larger circles indicate highly significant regressions at the p < 0.01 level; smaller circles indicate significance at the p < 0.05 level. Results at 2.5°spatial resolution smoothed for display by bilinear interpolation. (b) As in (a), but for the AMJ NAO and mean AMJ surface air temperature. air temperatures during negative NAOs likely exacerbating plant moisture-stress.
To summarize, within those parts of the EWB exhibiting both declining primary production from 2000 to 2010 (figure 3(c)) and highly negative land surface anomalies in 2010 ( figure 3(b)), the combined influence of the DJF and AMJ phases of the NAO is reflected in coefficients of determination generally exceeding 50% (figure 7(c)).

Discussion
We found land surface anomalies preceding the Russian heat wave that were empirically consistent with a persistently negative NAO in 2010. Given that the NDVI is an effective proxy for leaf area in semi-arid ecosystems (Fan et al 2009, Li andGuo 2012), these anomalies point to the importance of an accurate representation of leaf area dynamics (vegetation phenology) before and during the Russian heat wave, as has been recently shown in simulations of the 2003 European heat wave (Stéfanon et al 2012a). Notably, reductions in early season NDVI prior to the European heat wave were also concentrated in agricultural landscapes (Zaitchik et al 2006). Miralles et al (2014) recently demonstrated the importance of soil drying and resulting increases in sensible heat flux as drivers of air temperature extremes during the Russian heat wave. Interestingly, they found soil moisture anomalies at the heat wave's onset closely resembling the pattern of land surface anomalies that we document. Areas with initial soil moisture deficits later exhibited the highest absolute temperatures recorded during the heat wave. However, Miralles et al (2014) also found that such deficits were not a necessary pre-condition for an event as spatially extensive as the Russian heat wave. In areas where soil moisture deficits were not initially present, atmospheric blocking effects including warm air advection and high insolation under clear skies were sufficiently strong enough to force land-atmosphere feedbacks that rapidly dried soils, tipped the surface energy balance toward high sensible heat fluxes, and entrained hot air within the atmospheric boundary layer (Miralles et al 2014). In light of these results, land surface anomalies linked to the NAO might be best treated as playing a contributing role in the Russian heat wave, in that they affected the partitioning of latent and sensible heat at the heat wave's onset, but not a causal one, such that an anomalously negative NAO was not a necessary pre-condition. Similarly, Stéfanon et al (2012b) found that a characteristically Russian-type of heat wave is dominated by synoptic circulation, not spring drought. On the other hand, land surface pre-conditioning has been shown to increase the sensitivity of summer temperatures in Europe (daily maximum air temperature) to atmospheric blocking (Fischer et al 2007, Hirschi et al 2011, Quesada et al 2012.  (Rudolf et al 2010). Results at 0.5°spatial resolution smoothed for display by bilinear interpolation. Larger (more dense) circles indicate regressions statistically significant at the p < 0.01 level. Smaller (less dense) circles indicate significance at the p < 0.05 level. (b) 2010 DJF precipitation anomalies from the NCEP-CAMS (Ropelewski et al 1984). Results at 2.5°s patial resolution smoothed by bilinear interpolation. (c) As in (a), but for the AMJ NAO and AMJ precipitation. (d) As in (b), but for 2010 AMJ precipitation anomalies.
We expect that coupled atmosphere-land surface models taking into account both leaf area and soil-moisture anomalies prior to the Russian heat wave will better resolve the relative importance of synoptic circulation versus land surface preconditioning in 2010.
The linkage between a negative AMJ NAO and reduced primary production in the EWB (figure 7(b)) is consistent with teleconnections related to drought, i.e., with AMJ surface air temperature (figure 5(b)) and precipitation (figure 6(c)). However, the nearly uniform influence of the winter NAO on the EWB (figure 7(a)) is not as intuitive. Previous studies have shown a similarly-lagged statistical relationship between the winter NAO and early season NDVI in central Eurasia You 2004, de Beurs andHenebry 2008); with evidence that negative winter NAOs are associated with a delayed start of season (SOS), positive winter NAOs with an early SOS. However, we found no evidence of a delayed SOS in 2010, or a trend in SOS during the 2000 s (result not shown). In fact, there was an early snowmelt prior to the Russian heat wave (Barriopedro et al 2011). An alternate explanation could be the winter NAO's effect on overwintering survival of winter wheat. In 2010, winterkill totaling nearly two-million hectares of wheat was reported in Western Russia (Vocke et al 2010). However, we suspect that negative winter NAOs impact early season plant growth via deep soil freezing during anomalously cold winters; with subsequent effects on moisture availability in spring. In Southern Russia, Cherenkova (2012) reported reduced infiltration of snowmelt in spring 2010 due to an unusually deep layer of frozen soil. In addition, so-called 'physiological drought' can occur when plant roots restricted to the upper soil profile by frozen sub-soils quickly exhaust available soil moisture (Repo et al 2008). In sum, there is a clear need for further study to explain the winter NAO's lagged influence in the EWB.
Empirical relationships found here point to the need for mechanistic studies of atmospheric circulation into the EWB during the months of April-May-June (AMJ); a 'shoulder season' not typically considered. For example, the AMJ temperature pattern in figure 5(b) is indicative of Northward advection of warm air from North Africa and the Middle East during a negative NAO; suggesting the likely importance of regional blocking patterns during the early season. To the best of our knowledge, this teleconnection has not been previously described. Such an absence is notable given the AMJ pattern's pronounced contrast with effects of the winter NAO (figure 5(a)), where advection from the North Atlantic plays a primary role (Hurrell 1995). Studies of the NAO's influence on the East Asian summer monsoon have shown that an anomalously negative spring NAO can induce persistent sea surface temperature anomalies in the North Atlantic; exciting a Rossby wave train across Eurasia during summer and enhancing atmospheric blocking centered on the Ural Mountains (Wu et al 2009(Wu et al , 2012. Whether Rossby wave propagation during the early season might also explain the AMJ temperature teleconnection is an important question. Declining primary production in the EWB (figure 3(c)) was consistent with a decadal decline in both the DJF and AMJ phases of the NAO (figure 2). Such trends during the 2000 s may simply reflect normal multi-decadal variability (e.g., figure 4(b)). However, a number of studies have now shown that reduced sea ice extent in the Arctic can induce wintertime circulation anomalies resembling a negative NAO (Francis et al 2009, 2012, Honda et al 2009, Seierstad and Bader 2009, Overland et al 2011, Jaiser et al 2012, Liu et al 2012, Screen et al 2013, Tang et al 2014. This raises the possibility that a negative shift in mean Arctic ice extent due to anthropogenic forcing (Comiso 2006, Serreze et al 2007, Wang and Overland 2009) could shift the probability distribution of the winter NAO toward its negative phase (Strong andMagnusdottir 2009, Jaiser et al 2012). Given our results, such a shift could pose a previously unrecognized climatechange risk to wheat production in the EWB. We note, however, that comprehensive GCM studies do not produce an unequivocal, negative forcing of the winter NAO in response to projected sea ice losses (Deser et al 2010, Gillett andFyfe 2013).
Potential negative forcing of the NAO during the AMJ shoulder season has not been investigated. However, others have proposed a link between rapid Arctic climate change (Arctic amplification) and an increase in Rossby wave amplitudes (Francis and Vavrus 2012, Petoukhov et al 2014, Tang et al 2014, i.e., a more meandering atmospheric circulation roughly analogous to a negative NAO. This theory is controversial (Barnes 2013, Screen and Simmonds 2013a, 2013b), but we draw attention to it here because of the EWB's apparent sensitivity to the NAO's negative phase. If Arctic amplification does indeed effect a more meandering circulation at mid-latitudes, the EWB may prove to be an early responder to such a shift.
In sum, our results and those of others (Lobell et al 2011, Rahmstorf and Coumou 2011, Otto et al 2012 clearly show that the wheat growing regions of Russia, Ukraine, and Kazakhstan form a nexus of unusual environmental change. Given projections of an increasing contribution to global food security by the EWB (Fischer et al 2005, World Bank 2009, Lioubimtseva and Henebry 2012, we recommend the region as an important focal area for future climate change studiesemphasizing the EWB's potential vulnerability to Arctic amplification.