Elucidating the impact of temperature variability and extremes on cereal croplands through remote sensing

Introduction

The wheat crop is crucial to support global food security with the largest global cropland coverage (Thenkabail et al., 2012). Yet, similar to other cereal crops its level of production and productivity is vulnerable to a changing climate. Increases in mean growing season temperature, an increase in the frequency of extreme heat stress events at key phenological stages (e.g., the thermo-sensitive anthesis and grain-filling periods), uncertain precipitation futures and increased risk of droughts and subsequent moisture shortages all have a potentially negative impact on wheat crop yield (Ortiz et al., 2008; Asseng et al., 2012; Lobell & Gourdji, 2012; Lobell et al., 2012; Gourdji et al., 2013a; Koehler et al., 2013; Teixeira et al., 2013). Across the world's major wheat croplands, the thermo-sensitive periods (TSP) of crop development coincide with the timing of highest average maximum temperatures annually (Asseng et al., 2010), and growing season temperature and extreme heat events during the TSP are projected to increase in the next century (Gourdji et al., 2013a; Mathison et al., 2013; Teixeira et al., 2013). Climate change impacts are already limiting wheat crop yields globally: models indicate that warming trends since 1980 led to a 5.5% reduction in wheat production (Lobell et al., 2011). Model projections of increased exposure to heat stress during the TSP up to 2100 suggest that suitable adaptations need to be implemented urgently to secure climate-resilient wheat production (Teixeira et al., 2013; Deryng et al., 2014).

The underlying agricultural system (e.g., access to irrigation, cultivar type, soil type and ecosystem services), which varies within and between cropping landscapes, can increase or decrease the sensitivity of wheat crops to harmful climate impacts (Luers et al., 2003; Luers, 2005; Asseng et al., 2010; Gourdji et al., 2013b; Teixeira et al., 2013). For example, access to sufficient irrigation can enable transpiration which cools canopy temperatures relative to atmospheric temperatures, reducing the potential negative impact of warming during the TSP on the crop (Wassmann et al., 2009; Asseng et al., 2010; Gourdji et al., 2013a; Teixeira et al., 2013). Therefore, to understand the impacts of climatic variation, and specifically heat stress during the TSP, on wheat crop yields as it occurs in real world cropping landscapes requires the ability to observe and test a variety of temperature variables whilst capturing the spatial variation in the underlying agricultural system. Studies exploring the impacts of climatic variables on wheat crop yield are more informative to climate-resilient adaptation when they include local or regional detail. This is because potential adaptations often include shifting dates of cropping systems, implementing zero-tillage to avoid periods of heat stress co-occurring with the TSP, and the need to be sensitive to location-specific double/triple cropping rotations (Lobell et al., 2012, 2013; Teixeira et al., 2013).

Crop yield-climate interactions are usually explored using either (i) crop simulation models which aim to replicate crop physiological responses to climatic variation (Challinor et al., 2005; Asseng et al., 2010; Koehler et al., 2013) or (ii) regression models trained with crop yield and climate data aggregated within administrative boundaries (Lobell & Burke, 2010; Schlenker & Lobell, 2010; Rowhani et al., 2011; Urban et al., 2012). Crop simulation models are often complex and require large amounts of input data to represent the underlying complexity of the agricultural system (Welch et al., 2010; White et al., 2011); they are, therefore, limited in their application over large spatial extents. For example, a recent global crop model inter-comparison project simulated climate change impacts on crop yield at a 0.5° spatial resolution (Rosenzweig et al., 2014); this will aggregate farm and field level crop yield variability. This issue is particularly pertinent in low-latitude regions where there is the largest concentration of small and marginal farmers with a heterogeneous landscape and crop yields are projected to be most vulnerable to climate change impacts (Challinor et al., 2014; Rosenzweig et al., 2014). Alongside aggregating spatial variability in agronomic conditions, crop simulation models applied at a global level require simplification of agronomic practices; for example, specifying either full irrigation or no irrigation which will introduce further uncertainty in simulated yields (Rosenzweig et al., 2014).

Moreover, often crop simulation models do not capture the differential impacts of heating events during key phenological stages well. In a review of 221 peer-reviewed climate-crop simulation model studies only 14 partially or fully addressed the issue of heat stress (White et al., 2011). Deryng et al. (2014) use the PEGASUS 1.1 crop simulation model to test the impact of projected changes in heat stress during the TSP on crop production. Whilst this study suggests heat stress during the TSP will limit crop yields it is limited by only using one crop simulation model and the ability to accurately capture the dynamics in the underlying agricultural system such as rates of irrigation application, farmer decisions on planting dates and using fertilizer application data aggregated to national levels (Deryng et al., 2014). A multi-model comparison of different crop model projections of temperature impacts on wheat crop yield suggested that differences in crop model structure, parameterization and representation of temperature impacts was a larger source of uncertainty than uncertainty in climate projections (Asseng et al., 2013). This issue was echoed in another multi-model study assessing climate change impacts on crop yield where differences in the processes the models simulated and how the models were parameterized led to uncertainty in projected climate impacts (Rosenzweig et al., 2014). In this global gridded crop model inter-comparison study, it was also noted that few models simulated heat stress at critical crop development stages such as the TSP. Crop model uncertainty was also deemed a larger or equivalent source of uncertainty than that introduced through climate models when simulating wheat yield under future climates using the Global Large Area Model (GLAM) (Koehler et al., 2013). Studies have shown that CERES and APSIM crop simulation models underestimated the shortening of the wheat growing season when exposed to increased heat stress events (Lobell et al., 2012). This suggests that climate-crop simulation models may underestimate the true negative impacts of climate change on crop yield; this is pertinent given projected future warming and increases in extreme heat days (Gourdji et al., 2013a; Mathison et al., 2013).

In contrast, crop yield-climate models trained at the administrative boundary level aggregate the complexity of the underlying agricultural system which can be problematic in heterogeneous agricultural landscapes. Also, crop yield-climate models trained at the administrative boundary level cannot capture the differential impact of climatic variables at varying phenological stages such as heat stress during the TSP, thus, missing information to inform optimum climate-resilient adaptations.

In this paper, we demonstrate how remote sensing data can be used to quantify the impacts of temperature during the TSP on wheat crop yields in real world cropping landscapes, thus, overcoming the limitations of the two, previously discussed approaches. The local detail captured in remote sensing data enables a more appropriate representation of spatially heterogeneous agricultural systems. Remote sensing estimates of crop production incorporate measures of underlying system factors within a pixel (e.g., farmer decisions, access to irrigation, sowing date). The repeat coverage of remote sensing enables monitoring of crop phenology at a fine spatial resolution across a large spatial extent. This, therefore, enables the discrimination of temperature events and warming which occur at different phenological stages, such as the TSP, and an assessment of their impacts on final crop yield.

These research questions are tested in the wheat croplands of Punjab and Haryana in north-west India, a major global wheat breadbasket, which was heavily developed using Green Revolution advances and favourable government policy since the 1960s and 1970s (Murgai et al., 2001; Aggarwal et al., 2004; Perveen et al., 2012; Ojha et al., 2013). The wheat crop in the region supports, directly, the livelihoods of tens of millions of people and also supports national government procurement for buffer stocks and national food security and welfare schemes (Aggarwal et al., 2004; DES, 2012; Perveen et al., 2012). Key to the Green Revolution in Punjab and Haryana was the development of an extensive irrigation infrastructure to support wheat cropping during the dry rabi season (November–April) (Aggarwal et al., 2004; Ojha et al., 2013). In the 2010–2011 growing season 98.8% and 99.4% of the wheat crop were irrigated, and 80% and 84% of annual precipitation in a normal year falls during the monsoon outside the wheat growing season, in Punjab and Haryana, respectively (DES, 2012) (Fig. 1). Therefore, water availability and precipitation variability are unlikely to be limiting factors for wheat crop yield; however, there is a marked rise in temperature through the latter half of the wheat growing season when the TSP occurs (Fig. 1). Also, the supplementary material in Lobell et al. (2012) highlights the increased frequency of extreme heat events (daily maximum temperature >34 °C) during the latter stages of the wheat crop growing season in this region (March and April). This makes this region a suitable site to test for monitoring temperature impacts during the TSP on final wheat crop yield.

Materials and methods

MODIS MOD09A1 8-day composites of land surface reflectance covering Punjab and Haryana for the years 2002–2007 were assessed for quality (i.e., not cloud contaminated) with ‘bad pixels’ as defined by the MODIS Quality Assurance data removed and gap-filled following (Peng et al., 2011). Pixels corresponding to wheat cropping were extracted per-year using a variety of land cover masks (Xiao et al., 2005, 2006), and a spectral matching technique (SMT) classification procedure (Thenkabail et al., 2007). The SMT provided a statistical match between the shape and amplitude of wheat crop phenology and pixel phenology via a spectral similarity value (SSV) (Thenkabail et al., 2007). R² values for the relationship between remote sensing-derived wheat crop extent and district-wise land use statistics (http://lus.dacnet.nic.in/) ranged between 0.86 and 0.92 between 2002–2003 and 2006–2007. Phenology was generated from the MODIS data using the enhanced vegetation index (EVI) which provides a measure of vegetation vigour, photosynthetic activity and biomass (Pettorelli et al., 2005). The gap-filled EVI values were smoothed using a Savitzky-Golay filter which included a fitting bias towards maximum values. This removes noise and fluctuation in temporal EVI profiles due to cloud cover and atmospheric contamination and accounts for the negative bias in reflectance measured at the sensor (Chen et al., 2004). Extracting only wheat pixels minimized error from non-wheat land covers propagating up to the crop yield-climate models (Atzberger, 2013), ensuring that the models captured the interaction between climatic variables and crop yield accurately.

Crop yield was estimated per-wheat pixel for the 2002–2003 to the 2006–2007 growing seasons using a cumulative sum of EVI values over an approximation of the TSP [CUM-EVI_(TSP)]. The period of maximum EVI has been shown to correspond to heading date in cereal crops (Sakamoto et al., 2005). Teixeira et al. (2013) found that a 30 day period around the reproductive crop development phase represented the TSP and captured extreme heat impacts on crop yield. A 30 day period post maximum EVI was taken to represent the TSP. A cumulative sum [or integration of vegetation index (VI) values] and maximum VI values are used commonly as surrogate measures of vegetation productivity and crop yield (Pettorelli et al., 2005; Funk & Budde, 2009; Vrieling et al., 2011; Rembold et al., 2013). VI values post-peak growing season often provide more accurate predictions of crop yield as they correspond to the reproductive and grain-filling development stages of cereal crops (Funk & Budde, 2009; Rojas et al., 2011). CUM-EVI_(TSP) was significantly (P < 0.01) correlated with district-wise wheat crop yield and production between 2002–2003 and 2006–2007 (Fig. 2; http://apy.dacnet.nic.in/). The R² values for the regression relationship between CUM-EVI_(TSP) and district-wise crop yield are stronger than those obtained using Landsat data to estimate district-wise wheat crop yield in north-west India using the Monteith method (Lobell et al., 2010). The R² value for the CUM-EVI_(TSP) crop yield model used here was 0.6 (Fig. 2a) whereas the R² value in the Lobell et al. (2010) study was 0.28. The correlation for the regression relationship between CUM-EVI_(TSP) and district-wise crop yield is also comparable to the correlations obtained with statistical models estimating crop yield from MODIS EVI data in the USA (Bolton & Friedl, 2013). This suggests that in relatively homogenous cropping landscapes (e.g., the USA or Punjab and Haryana) the temporal detail in MODIS data enabling monitoring during the yield sensitive TSP is of greater importance to yield prediction than the spatial detail of Landsat data (MODIS: daily imagery, 8-day composite, 500 m spatial resolution; Landsat: 16-day imagery, 30 m spatial resolution). However, this trade-off should be evaluated on a crop-by-crop, location-by-location and sensor-by-sensor basis. The larger correlation between CUM-EVI_(TSP) and district-wise crop production compared to crop yield is likely due to greater between district variation in cropped area and, thus, production (Fig. 2).

The per-pixel predictions of crop yield were integrated with gridded daily precipitation and maximum and minimum temperature datasets to train crop yield-climate panel-regression models. The APHRODITE (V1003R1) dataset provided precipitation data (Xie et al., 2007; Yatagai et al., 2009, 2012). The APHRODITE daily temperature product (V1204R1) provides daily mean temperature only, which inhibits exploring the differential impact of minimum and maximum temperature, and extreme temperatures, on crop yield. Therefore, Global Summary of the Day (GSOD) stations in Punjab and Haryana with a near complete record of daily minimum and maximum temperatures were extracted from: (http://www.ncdc.noaa.gov/). Weather stations were included on the basis of spatial coverage over Punjab and Haryana and completeness in temporal coverage with minimal missing data. Conservatively selecting stations with reliable and comprehensive temperature records was appropriate over Punjab and Haryana as there is minimal orographic variability, especially over cultivated lands, which would cause dramatic shifts in temperatures over short distances. The data from these stations were used to generate gridded fields, at the same spatial resolution as the MODIS data, using an inverse-distance weighting algorithm. GSOD weather stations have been used as inputs in the generation of gridded climate products (Yasutomi et al., 2011) and to assess climate impacts on crops in north India (Lobell et al., 2012).

Linear regression models with fixed-effects terms [Eqn 1] were fitted to explore whether the impact of temperature and extreme heat events during the TSP on crop yield could be detected using remote sensing data.

(1)

where i refers to pixel i and t refers to time of observation t = 2002….2007. x_it is the predictor temperature variable in pixel i at observation t, c_i is the fixed-effects term for pixel i and ɛ_it is an error term. CUM-EVI_(TSP) was regressed against average minimum, maximum and mean temperature and, extreme degree days (EDD) during the TSP. Average daily minimum and maximum temperature were computed separately as they have been shown to have differing impacts on cereal crop yield (Peng et al., 2004; Welch et al., 2010). EDD was defined as:

(2)

where T is the threshold temperature for which EDD is computed individually for each temperature from 32 to 42 °C at 1 °C increments, t_i is the maximum temperature on day i and N is the number of days in the TSP (N = 30). Computing EDD above a threshold maximum temperature, increasing with 1 °C intervals enabled assessment of whether declines in yield due to exceedance of critical temperatures during the TSP can be detected empirically from remote sensing data. On the basis of performing a Hausman test, fixed-effects terms were preferred over random-effects terms. The same sign (i.e., positive or negative slope coefficient) in the relationship between temperature variables and crop yield was obtained when using random-effects terms indicating that the results were not artefacts of including fixed-effects terms. Also, we further validated the regression models using a leave-one-out approach; the fixed-effects regression models were performed leaving out temperature and wheat crop yield data for each year in turn.

Including a fixed-effects term allows the regression intercept to vary spatially. It can be thought of as accounting for time-invariant effects of the underlying agricultural system unique to each location (these effects are not specified, but could include omitted variables such as soil condition, fertilizer application and access to irrigation). Through this remote sensing-based approach key climatic stresses in a particular cropping landscape can be elucidated accounting for the differential sensitivity caused by underlying system factors.

Using remote sensing data means that a unique phenology can be captured per-pixel, per-year enabling estimation of phenological parameters such as start-of-season (SOS), end-of-season (EOS) and approximation of the TSP. Coarser spatial resolution yield data (e.g., administrative boundary yield estimates) would mask variation in the timing of key phenological stages. Using coarser spatial resolution yield data would therefore prohibit assessment of intra-growing season sensitivity to climatic variables such as warming during the TSP. SOS, EOS, and TSP can be obtained per-pixel to extract climate information relevant only to the exact growing season of the crop or the TSP. This avoids using monthly or seasonal averages of climate data which correspond to normal growing seasons and may retain climate signals not relevant to crop growth and are not sensitive to spatial and temporal changes in the timing and length of growing seasons. Monitoring phenology using remote sensing means that regression models could be fitted for varying SOS dates. This enables assessment of how varying SOS influenced crop exposure to temperature variation and accounts for the fact that varying day length over a season can influence crop development rates (Lobell et al., 2012). This is important as shifting sowing dates and cropping calendars are a potential adaptation to reduce exposure to extreme heat events (Teixeira et al., 2013). Regression models were fitted for SOS on day-of-year 329, 337, 345, 353, and, 361.

Panel-datasets are multi-dimensional in time and space and, thus, can capture a wider range of temperatures which exhibit greater spatial variation relative to temporal variation at one location (Lobell & Burke, 2010). Panel-regression models have shown improved performance (in terms of strength of association) for capturing temperature-crop yield relationships relative to time-series models (Lobell & Burke, 2010). The temporal extent of the panel-data was restricted to 5 years to mitigate the impact of shifting agricultural practices or technological advances on crop yield. Several studies have incorporated linear, quadratic or cubic regression spline time terms into crop yield-climate regression models to account for such technological development (Schlenker & Lobell, 2010; Rowhani et al., 2011; Hawkins et al., 2013). However, this requires the assumption that the impact of developmental change on crop yield is universal across the spatial extent of the panel. Given the spatial variability in cropping-systems (Panigrahy et al., 2010), and variation in levels of natural, physical, social, and financial capital, which influence uptake of agricultural practices (Erenstein et al., 2007), across the panel, utilizing a time term to capture development and assuming its spatial universality was not justified. Crop yield-climate regression models trained using panel-data are less sensitive to the temporal extent over which observations were taken (Lobell & Burke, 2010). For example, Welch et al. (2010) detected the signal of minimum and maximum temperature impacts on rice yield using panel data with observations taken over 5 years.

Results

Temperature during the TSP and crop yield relationship

Average minimum and average maximum temperature during the TSP have greater explanatory power compared to EDD computed with incrementing temperature thresholds and average growing season temperature (Table 1). Increases in average minimum temperature during the TSP and average growing season temperature have a larger negative impact on wheat yield compared to increases in average maximum temperature during the TSP (Fig. 3). However, increases in all temperature variables have a negative impact on wheat crop yield (Figs 3 and 4). Generally, an increase in the temperature threshold above which EDD is computed registers a greater negative influence on wheat crop yield (Fig. 4). However, a temperature threshold of 35 °C had a noticeably larger R² compared to all other temperature thresholds (Table 1), there was also an increase in the size of the slope coefficient at 35 °C compared to lower temperature thresholds (Fig. 4). The negative impact of temperature during the TSP, determined by both the value of the slope coefficient and the fit of the model, increased with later SOS dates (Figs 3 and 4; Table 1). For example, for a SOS on day-of-year 329 a per-unit (i.e., 1 °C) increase in average minimum temperature during the TSP would result in −0.0453 t ha⁻¹ decrease in wheat yield whereas the same warming for SOS on day-of-year 361 would result in a yield decrease of −0.0792 t ha⁻¹. The negative impact of exposure to heating events greater than 35 °C during the TSP also increased with later SOS dates (Fig. 4). The same sign (i.e., positive or negative slope coefficient) and patterns in the magnitude of R² values were observed in the leave-one-out regression models as occurred in the models using the full panel (Supporting Information). This suggests the results from the full panel fixed-effects regression models are robust in capturing temperature impacts on wheat crop yield and are not artefacts of including anomalous years or data points in the regression models.

Table 1. R² values for average minimum and average maximum temperature during the TSP, average growing season temperature (T) and EDD_(T) accumulated above incrementing temperature thresholds when wheat crop yield was regressed on each of them independently, shown for different SOS dates. Only results significant at P < 0.001 are shown

Variable	SOS (day-of-year)
	329	337	345	353	361
T(TSP)(min)	0.25	0.34	0.39	0.54	0.61
T(TSP)(max)	0.24	0.39	0.38	0.38	0.43
T	0.06	0.13	0.14	0.10	0.06
EDD(T): 32	0.08	0.11	0.05	0.06	0.18
EDD(T): 33	0.13	0.19	0.15	0.09	0.24
EDD(T): 34	0.10	0.19	0.14	0.08	0.29
EDD(T): 35	0.14	0.30	0.34	0.37	0.54
EDD(T): 36	0.04	0.15	0.17	0.19	0.23
EDD(T): 37	0.13	0.11	0.04	0.08	0.13

Discussion

Average minimum and average maximum TSP temperature explain a greater amount of yield variation compared to mean growing season temperature (Table 1). The TSP approximates the reproductive and the beginning of the grain-filling development stages. While several studies focus on the impacts of daily maximum temperature or daytime temperatures above a critical threshold during the TSP or the entire growing season (Asseng et al., 2010; Lobell et al., 2012; Gourdji et al., 2013a; Koehler et al., 2013; Teixeira et al., 2013), it was shown here that minimum temperature during the TSP has an equivalent if not greater negative impact on wheat yields (Fig. 3; Table 1). Lobell & Ortiz-Monasterio (2007) used the CERES-wheat model and historical wheat yield data in Mexico and California to explore the differential yield impacts of minimum and maximum temperatures. While this study also showed the negative impact of warming minimum temperatures on wheat crop yield, it noted that the CERES-wheat model mechanism for simulating grain filling did not replicate the impact of minimum or maximum temperatures observed in historical yields (Lobell & Ortiz-Monasterio, 2007). It is important for further research to contribute to greater understanding of the differential impacts of daily minimum and maximum temperature, and associated physiological processes. This is pertinent given an observed (1970–2005) trend of increasing ‘hot nights’ in north-west India during the pre-monsoon wheat growing season (Kothawale et al., 2010) and an observed (1970–2003) increasing trend in winter minimum temperatures (Kothawale & Rupa Kumar, 2005). These results also suggest that the negative influence of average growing season temperature may be due to it capturing heating events during the TSP and, that adaptive efforts to increase wheat crop resilience to warming, should focus on the TSP.

Regressing EDD above incrementing temperature thresholds showed the expected negative impact of increasing temperature during the TSP on wheat crop yield (Fig. 4). These results suggest that wheat yields in Punjab and Haryana are more sensitive to daily maximum temperatures greater than 35 °C rather than a crop-specific critical temperature, often reported as 34 °C for wheat (Hatfield et al., 2011; Lobell et al., 2012; Gourdji et al., 2013a). The impact of cumulative exceedance of 35°C during the TSP on wheat yield also varied with SOS date (Fig. 4; Table 1); this is consistent with a range of other observational and crop simulation studies which did not reveal a uniform yield response to heating events greater than 34 °C (Figs 4 and 8 in Asseng et al. (2010)). It is worth noting that average minimum and average maximum TSP temperature cause a greater, negative rate of change in wheat yields than exceedance of the 35 °C threshold. This indicates that the effect of a critical temperature threshold is less pronounced in real world cereal cropping systems compared to more controlled, experimental environments or modelling frameworks where the critical temperature signal can be isolated (Jagadish et al., 2007; Asseng et al., 2010; Teixeira et al., 2013). There is likely spatial variation in crop variety, irrigation and vapour pressure deficit during the TSP across Punjab and Haryana, but also in croplands across the globe, which will alter wheat crop response to extreme heat events masking the effects of a critical temperature threshold (Porter & Gawith, 1999; Asseng et al., 2010; Gourdji et al., 2013b).

Spatial variation in the quality of irrigation water has been shown to influence wheat crop yield in Punjab and Haryana (Tyagi et al., 2005). However, the pixel specific fixed-effects terms included in the regression models should capture the variability in yield due to variation in irrigation water quality (and other omitted variables). As mentioned in the introduction, due to the timing of the rabi wheat growing season outside of the wet monsoon with near complete irrigation [>98% of the wheat crop is irrigated, (DES, 2012)] the wheat crop should not water limited and precipitation should have little impact on final yield. This is confirmed by univariate regression models using growing season total precipitation and TSP total precipitation as predictor variables (Table 2). The slope coefficients for precipitation variables show a very weak negative correlation with wheat crop yield (Table 2); this is contrary to expected theory and is likely an artefact of relatively small amounts of precipitation during wheat growing season and the wheat crop being irrigated. Also, precipitation variables explain a relatively small amount of the variation in wheat crop yield compared to average minimum and maximum temperature during the TSP (Table 1 and 2).

The negative influence of average minimum and average maximum TSP temperature on final wheat yield increased with later SOS dates (Fig. 3). This trend is consistent with observations by Lobell et al. (2012) who observed declines in wheat growing season length associated with later SOS dates in north India; this was attributed to increased exposure to extreme heat days (>34 °C) later in the wheat growing season. These results (Figs 3 and 4; Table 1) suggest that climatic events, namely warming during the TSP, are limiting potential wheat yields in Punjab and Haryana. For wheat crops with a later SOS, earlier sowing may be an escape route reducing exposure to damaging heat events and, thus, closing existing yield gaps. Widespread adoption of zero-tillage represents a suitable adaptive, climate-resilient, management strategy with earlier SOS reducing TSP exposure to extreme heat events with subsequent environmental and socio-economic benefits and, no yield penalty (Erenstein & Laxmi, 2008; Jat et al., 2009). However, even wheat crops with a SOS of day-of-year 329, were negatively impacted by warming during the TSP (Figs 3 and 4; Tables 1). This indicates that alongside reducing later sown wheat crops' exposure to higher temperatures (‘escape’ strategies) adaptations need to be explored which increase wheat crop tolerance to warming during the TSP to reduce temperature-induced yield gaps [e.g., develop wheat varieties tolerant to extreme temperatures (Gourdji et al., 2013b)]. Implementing such adaptations are important given (i) that current temperatures are limiting wheat yield; and (ii) projected future warming trends in the wheat growing season and increased frequency of extreme heat days are likely to further limit wheat yields (Gourdji et al., 2013a; Mathison et al., 2013). Such adaptations could have important future food security implications given coincidental pressures of increased demand for food being placed on these croplands due to population growth (Aggarwal et al., 2004), alongside unfavourable warming (Gourdji et al., 2013a; Mathison et al., 2013).