Enhanced evapotranspiration was observed during extreme drought from Miscanthus, opposite of other crops

The impact of extreme drought and heat stress that occurred in the Midwestern U.S. in 2012 on evapotranspiration (ET), net ecosystem productivity (NEP), and water‐use efficiency (WUE) of three perennial ecosystems (switchgrass, miscanthus, prairie) and a maize/soybean agroecosystem was studied as part of a long‐term experiment. Miscanthus had a slower initial response but an eventually drastic ET as drought intensified, which resulted in the largest water deficit among the crops. The substantially higher ET at peak drought was likely supplied by access to deep soil water, but suggests that stomatal conductance of miscanthus during the drought may respond differently than the other ecosystems, consistent with an anisohydric strategy. While there was a discrepancy in the water consumption of maize and switchgrass/prairie in the early time of drought, all these ecosystems followed a water‐saving strategy when drought intensified. The gross primary production (GPP) of miscanthus dropped, but was reversible, when temperature reached 40 °C and still provided the largest total GPP among the ecosystems. Increased ET for miscanthus during 2012 resulted a large decline in ecosystem WUE compared to what was observed in other years. The biophysical responses of miscanthus measured during an extreme, historic drought suggest that this species can maintain high productivity longer than other ecosystems during a drought at the expense of water use. While miscanthus maintained productivity during drought, recovery lagged associated with depleted soil moisture. The enhanced ET of miscanthus may intensify droughts through increase supply of deep soil moisture to the atmosphere.


Introduction
As a consequence of global climate change, extreme weather events (heat waves, drought, etc.) are predicted to become more frequent and intense (Collins et al., 2013). The increase in extreme events is predicted to adversely affect water availability and plant growth and has already become relevant in several regions worldwide (Ciais et al., 2005;Rahmstorf & Coumou, 2011;Coumou & Rahmstorf, 2012). Concerns over global climate change have generated an effort to understand how environmental changes, such as seen in temperature and precipitation, influence net carbon exchange between terrestrial ecosystems and the atmosphere (Pingintha et al., 2010). In regions where water deficit occurs frequently, plants developed adaptive strategies to cope with drought (Maroco et al., 1997;Borrell et al., 2006;Araus et al., 2008). One adaptation strategy is to conserve water early in the growing season to maintain support for later growth (Sinclair et al., 2005;Zaman-Allah et al., 2011). Stomata closure to prevent water loss can be induced by high atmospheric vapor pressure deficit (VPD) (Sinclair et al., 2005;Fletcher et al., 2007;Kholov a et al., 2010) or by root signaling triggered by the dry soil status (Zhang & Davies, 1990).
Miscanthus 9 giganteus (miscanthus) is a potential bioenergy crop that due to its high yield potential, and water-use efficiency may drive land-use change in the Midwestern United States (Heaton et al., 2010). Miscanthus, however, has shown high sensitivity to water deficit in pot experiments, resulting in leaf senescence, lower biomass production (Clifton-Brown & Lewandowski, 2000;Mann et al., 2013b), and reduction in photosynthetic performance (Ings et al., 2013). Ings et al. (2013) concluded that miscanthus, in artificial growth environments, continues to maintain high rates of transpiration and physiological activity despite increasing water stress, indicating lack of drought adaptation. In pot experiments, miscanthus may extract all available moisture resulting in loss of photosynthetic function and ultimately susceptibility to drought. Another pot-based experiment demonstrated maintained productivity and a drought-avoidance strategy of another potential bioenergy crop, switchgrass, under a wide range of soil moisture conditions explained by its extensive root development (Mann et al., 2013b). However, the dynamics of species responses to extreme environments are likely to vary between potted experiments and natural growth environments, where rooting depths and competition among neighbors differ, and additional impacts of other environmental drivers, such as heat stress may co-occur. It is unclear whether a similar response would occur under field conditions where deeper roots may have the benefit of accessing deeper water. For example, in a recent study on carbon dynamics, three perennial crops maintained greater annual productivity than the annual crops during severe drought (Joo et al., 2016). A high diversity ecosystem, such as a restored tall-grass prairie, has been proposed as a potential bioenergy feedstock ecosystem (Tilman et al., 2006) and is likely to adopt a different drought response due to the role of various species responding independently to environmental conditions. Annual crops that presently dominate the Midwestern landscape, such as maize, have more limited root development and therefore are likely to be more sensitive to drought, generally providing lower yields in dry summers.
In this study, we present the combined effect of drought and heat stress on evapotranspiration (ET), net ecosystem productivity (NEP), and water-use efficiency (WUE) of three perennial and one annual agroecosystem. The objectives of this research were to (i) understand the variation behind the responses to extreme drought for one annual and three perennial ecosystems exposed to historic drought conditions and (ii) to assess whether the efficiency of water use at the ecosystem and at the harvest scales vary in response to extreme drought conditions. These objectives are addressed using four ecosystems representing three perennial and one annual ecosystem and build upon a previous study of the same ecosystem (Joo et al., 2016). There we reported that the perennials maintain higher productivity during a drought than the annual row crops, and now we provide mechanistic insights into the dynamics of water use, productivity, and water-use efficiency before, during, and immediately following the drought.

Climate and site management
Measurements were carried out at the Energy Farm of the University of Illinois at Urbana-Champaign, IL, USA, between 2009 and 2013. The growing season typically starts in April, and summer is generally characterized as warm and relatively wet with temperatures above 20°C from June to August and mean annual precipitation of 1042 mm (Illinois State Water Survey, average between 1979 and 2009). The soil, typical to the region, is deep and fertile Flanagan (fine, montmorillonitic, mesic aquic Argiudoll) with low lying blocks of Drummer (typic Haplaquoll).
In May 2008, four replicated plots (200 9 200 m each) were established by planting three perennial species, switchgrass (Panicum virgatum L.), miscanthus (Miscanthus 9 giganteus), a mixture of tall grass prairie [complete species list in (Zeri et al., 2011)], and a maize (Zea mays L.)/soybean (Glycine max L.) crop rotation. The maize/soybean rotation consisted of soybean planted every third year, which corresponded to the 2010 and 2013 growing seasons. The maize and switchgrass fields were fertilized by the addition of 168, 202, and 180 kg ha À1 nitrogen in 2008, 2009, and 2011, respectively, for maize, and by 56 kg ha À1 in 2010-2012 for switchgrass, whereas soybean, miscanthus, and prairie were not fertilized on any years based on present management practices (Tilman et al., 2006;Davis et al., 2010;Zeri et al., 2011Zeri et al., , 2013. Further information on site management practices can be found in (Zeri et al., 2011.

Flux and meteorological measurements
Eddy covariance and micrometeorological stations were situated in the center of each plot. The eddy covariance system consists of a three-dimensional sonic anemometer (model 81000VRE; R.M. Young Company, Traverse City, MI, USA) and an open path infrared gas analyzer (IRGA; model LI-7500 upgraded to model LI-7500A in early 2012; LI-COR Biosciences, Lincoln, NE, USA). Each eddy covariance system was accompanied by a meteorological tower, equipped by a set of sensors to monitor temperature and relative humidity (HMP-45C; Campbell Scientific, Logan, UT, USA), up-and down-welling short-and long-wave radiation (CNR1; Kipp & Zonen, Delft, the Netherlands), canopy surface temperature (SI-121 or SI-111 Infrared radiometers; Apogee Instruments, Logan, UT, USA), up-and down-welling photosynthetically active radiation (LI-190; LI-COR, Biosciences, Lincoln, NE, USA), soil heat flux (HFP01; Hukseflux Thermal Sensors B.V., Delft, the Netherlands); and soil moisture and soil temperature (model Hydra Probe II; Stevens Water Monitoring Systems, Inc., Portland, OR, USA). A full description of the eddy covariance system has been published previously (Zeri et al., 2011. Ecosystem fluxes were calculated from the 10 Hz data using Alteddy (http://www.climatexchange.nl/projects/alteddy/) from 2008 until 2011 and EddyPro (http://www.licor.com/ env/products/eddy_covariance/software.html) in 2012-2013. Both software packages employed similar methods for correcting the high-frequency data, including coordinate alignment, correction of the sonic temperature due to the influence of humidity, and compensation of density fluctuations by the WPL term (Webb et al., 1980). The obtained data were filtered for periods of no turbulent mixing during nighttime (Foken et al., 2005), and for cases when more than 30% footprint of the data originated from outside of the plots (Hsieh et al., 2000). Missing data were gap-filled, and the fluxes were partitioned from net ecosystem exchange into ecosystem respiration (R eco ) and gross primary production (GPP) as in Reichstein et al. (2005) and Zeri et al. (2011). Ecosystem water-use efficiency was calculated by dividing daily (e.g., Fig. 3) and yearly (e.g., Table 1) integrated net ecosystem productivity (WUEeco), gross primary productivity (WUE GPP ), and harvest (WUE H ) by daily and yearly integrated ET, respectively. It should be noted

Results
The 2012 growing season was among the worst droughts to affect the Midwestern U.S. in decades Among the 5 years of this experiment, the mean maximum summer temperatures of~30°C were typically observed, except for 2011 and 2012, when average daily maximum temperatures reached 35°C in July (Fig. 1). In particular, the 2012 growing season was considerably drier and warmer relative to long-term mean conditions with prolonged unusually warm temperatures. The daily maximum air temperature often exceeded 35°C, occasionally reaching 40°C, between early July and middle September. Moreover, the spring of 2012 was exceptionally warm, with average maximum temperature of 20°C in March, compared to the 5-12°C range typically observed in other years.
The monthly photosynthetic photon flux density (PPFD) pattern follows the typical intensity of observed radiation of 1000-1400 mol m À2 month À1 in the summer, with the highest radiation received over the growing season in 2012 (Fig. 1). Generally the region receives an average PPFD of~1100 mol m À2 month À1 in May and June, while in 2012 this was 1400 mol m À2 month À1 . The large number of sunny days in 2012 resulted in the highest total PPFD measured between April and September (7489 mol À1 m À2 ) among the studied years. As a combined effect of temperature and light intensity, the growing season shifted earlier in 2012 compared to the other years, which is reflected by the development of the plants (Joo et al., 2016).
The annual cumulative precipitation measured on site in 2012 (752 mm) and 2013 (768 mm) was approximately 75% of the long-term mean of 1042 mm (http://mrcc.isws.illinois.edu/CLIMATE/), while other years fell within the normal long-term range. The monthly distribution of rainfall events in 2012 was different from the other years, with the lowest amount of precipitation immediately preceding and during the majority of the growing season (306 mm between January and August), which was approximately half typical precipitation (Fig. 1). The 2012 drought appeared together with high temperatures up to 40°C, which suggest a combined drought and heat stress on the plants, however for simplicity in what follows we refer to it as drought. A significant precipitation event occurred late in the 2012 growing season which brought the annual total precipitation higher, and this was followed by a very wet beginning of 2013, when the highest amount of cumulative precipitation (620 mm) among the studied years was observed between January and August. Opposite of 2012, the lower precipitation in 2013 occurred well into the growing season and mean temperatures were much cooler during the dry-down period.
The drought of 2012 coupled with high temperatures resulted in VPD being much larger in 2012 compared to the other, nondrought years. A large peak of 4-5 kPa was observed in the summer of 2012, lasting approximately 1 month (Fig. 2). In other years, VPD stayed below 3 kPa with the exception of 2011, when a short period (few days) of 4 kPa was reached.
Miscanthus maintained high ET during the drought relative to other ecosystems, which led to the largest water deficit Despite similar starts of the growing season and rapid accumulation of LAI (Joo et al., 2016), miscanthus shows a lag in the rate of increase in cumulative ET relative to the other perennial ecosystems (Fig. 2). However, the rate of increase in cumulative ET accelerates around day of year 200 after which cumulative ET for miscanthus meets or exceeds the other ecosystems. The exception to this was in 2013, when at the same time miscanthus already began to reduce its ET (Fig. 2). The maize/soybean ecosystem shows similar early-season responses as miscanthus; however, the planting date for the annual row crops is much later than the emergence date for the perennial ecosystems.
The difference between cumulative precipitation and ET (Cum(P-ET)) between 2009 and 2011 was generally positive, meaning that none of the ecosystems experienced water deficit, which is typically resulted in greater rates of ET than precipitation. In 2010 cumulative P-ET values fell (approximately À100 mm for prairie, where negative values indicate water deficit); however, this occurred close to the end of the growing season and late precipitation caused all ecosystems to finish with surplus moisture. On the other hand, in 2012 water deficit reached between À110 mm (maize) and À284 mm (miscanthus). At the same time, cumulative evapotranspiration of miscanthus reached 800 mm by the end of 2012, which was among the highest values for all species and over the duration of this experiment. The 2012 drought showed the largest variation in ET among the four ecosystems. Following the drought, in 2013 cumulative precipitation was generally greater than cumulative ET for all ecosystems. The progression of ET throughout 2013 was similar for all species with the exception of miscanthus, which had the lowest ET. As a result, miscanthus accumulated the largest water surplus among the ecosystems in 2013. The 5-yearcumulative P-ET was nearly the same for all species until the 2012 drought when the miscanthus ecosystem experienced the second largest water deficit, surpassing switchgrass. The 2012 drought had the largest impact on the miscanthus field, miscanthus generally having the best (largest) long-term water content; in 2012, this ecosystem turned to have the second lowest long-term water content, with a slow recharge in 2013. Although available data are limiting and there is large variation in the soil moisture measurements and their sensitivity among the ecosystems (due to methodological limitation), the deep soil moisture content measured at 100 cm belowground still confirmed the lowest water content level at the miscanthus plot, with a minimum value of approximately 0.05 water fraction by volume (wfv) during the drought, followed by switchgrass (0.19 wfv; Fig. S1). Until DOY 190, an intensive decline of soil water content was observed at the miscanthus field. After the drought, all ecosystems showed soil water content recharge (Fig. S1).

Net ecosystem productivity and water-use efficiency
Daily mean ET, net ecosystem productivity (NEP), and ecosystem water-use efficiency (WUE or WUE eco ) for the drought year and the years preceding and following the drought show that miscanthus deviates strongly in all three metrics relative to the other ecosystems (Figs 3 Fig. 2 Vapor pressure deficit over the diurnal time period, the cumulative evapotranspiration for each ecosystem (second row), and cumulative annual (third row) and experiment-long (forth row) water balance over each year. The four species experienced water deficit in 2012, in accordance with a peak of vapor pressure deficit (VPD). The largest effect was observed for miscanthus with the highest evapotranspiration due to the enhanced transpiration. and 4). In 2011 and 2013, miscanthus had relatively similar responses of ET and NEP with seasonal maximum values of~4 mm day À1 and~12 g m À2 day À1 , respectively. During the drought in 2012 ET of miscanthus peaked at 8 mm day À1 , twofold higher than observed in other year and nearly 1.5 times larger than observed for other ecosystems (Fig. 3). At the same time, NEP reached a maximum value at DOY~200, similar to the previous year. Both ET and NEP of miscanthus declined after DOY~200, but followed a typical annual curve after DOY~220 (Figs 3 and 4). Despite the drop in NEP, however, miscanthus reached the highest cumulative NEP (1102 g m À2 carbon) in 2012 (Table 1), as a prolonged carbon uptake was observed (until DOY~265) compared to the other species (until DOY~200) (Joo et al., 2016). The substantial increase in ET coupled with no net gain in NEP for miscanthus during 2012 resulted a large decline in WUE eco compared to what was observed in 2011 and 2013 (maximum values of 3 g m À2 mm À1 day À1 ).
The other three ecosystems maintained relatively consistent WUE eco throughout the drought relative to the previous years (Fig. 3). Switchgrass and prairie had similar ET and NEP values with a maximum of~6 mm and~9 g m À2 day À1 , respectively, in 2012, which were nearly the same as in 2013 (Fig. 3). WUE eco of these species followed a relatively consistent annual progression, even during the drought. ET of maize was nearly the same in 2011 and 2012, while NEP was slightly lower in 2012. A decline of WUE eco was observed for this crop as well in 2012; however, it was not as remarkable as for miscanthus. A shift in timing of maximal productivity was observed in 2012 relative to the other years for all species; however, miscanthus maintained a relatively large NEP throughout the entire 2012 growing season. The timing of peak WUE eco for the maize/soybean ecosystem varied among the 3 years, but this variation is largely attributed to management decisions associated with variation in planting date due to differences between maize and soybean as well as meteorological conditions influencing field access.
Throughout the experiment, WUE calculated using gross primary productivity (WUE GPP ) was relatively consistent for the perennial ecosystems, particularly for the prairie (Fig. 5b). Switchgrass showed a general decline in WUE GPP throughout the experiment, but in all cases, it was greater than or equal to WUE GPP for the prairie. Miscanthus, however, had greater variation in WUE GPP from year-to-year and did not follow any trend as did switchgrass. The maize/soybean ecosystem showed the greatest variability with the maize years (2009,2011,2012) having the highest WUE GPP and the soybean years the lowest. The hot and dry conditions 2012, and to a lesser extent in 2011, resulted in lower WUE GPP for maize than the more typical 2009 growing season. With the exception of 2009, the annual row crops had lower WUE GPP values than the perennial ecosystems, although they were similar to prairie in 2011 and in 2012.
Harvest WUE (WUE H ), calculated from the carbon in harvested biomass, was substantially more variable across the experiment (Fig. 5a). The miscanthus and switchgrass ecosystems showed a gradual increase in WUE H during the first 1-2 years after which the values stabilized, whereas the prairie ecosystem showed highly variable values of WUE H throughout the experiment. In 2009 and 2011 when maize was planted and in 2010 and 2013 when soybean was planted, WUE H was relatively consistent; however, the 2012 maize WUE H was much lower than observed in the other maize-growing years. All species, other than switchgrass, showed a decrease in WUE H during the 2012 drought. Normalizing ET based on VPD (Fig. 5c) shows that within the perennial ecosystems, miscanthus generally has lower values than the other two ecosystems except during the 2011 and 2012 growing seasons. During the two hot, dry years, all ecosystems had lower ET/VPD than the more typical growing seasons and the variation among the perennial ecosystems was smaller. To calculate this parameter, values of ET and VPD were averaged for the entire dataset of the particular year to make conclusions for the entire ecosystem. Using VPD normalized ET to calculate WUE (WUE GPP 9 VPD), the intrinsic water-use efficiency can be obtained. This resulted in more variation among years than WUE GPP itself (Fig. 5b), and with the exception of 2013, it was lowest in the prairie and similar between miscanthus and switchgrass. The two hotter and drier years, overall, had the highest WUE GPP 9 VPD. Water-use efficiency based on harvested biomass (a) and gross primary production (GPP) (b) for each ecosystem over the duration of the experiment. Also shown is the vapor pressure deficit (VPD) normalized evapotranspiration (ET) (c) which is used as a proxy for canopy conductance, and water-use efficiency (WUE) based on GPP using VPD normalized ET (d).

Water usage dynamics of miscanthus opposite to other ecosystems
Generally crops conserve water use by partial stomata closure at high VPD (Ball et al., 1987;Sinclair et al., 2005;Fletcher et al., 2007;Kholov a et al., 2010). Based on the cumulative P-ET curve in 2012 (Fig. 2), miscanthus prevented water loss the longest (until DOY~170) during the drought compared to the other species that experienced a gradually increasing water deficit starting earlier at DOY~150. Despite having the second largest water 'surplus' by the end of 2011 and the early-season water conservation, miscanthus had a rapid water use coupled with the lack of precipitation in 2012, which led miscanthus having the largest cumulative water deficit [ Fig. 2; Cum(P-ET) and 5 years cum(P-ET)], tied with prairie. In comparison with switchgrass and prairie, miscanthus extracted deep soil water, likely due to its long root system, to supply the large ET. The measured belowground biomass density and depth profile in the summer of 2011 confirmed that the perennials had substantive root systems, extending to a depth of at least 100 cm in contrast to maize that was dominant in the top 10 cm of the soil (Anderson-Teixeira et al., 2013). Among the perennials, switchgrass and miscanthus had extensive root systems below 50 cm, suggesting that these two ecosystems had the ability to reach deep soil moisture in case of a severe drought period. The deep soil moisture content (at 100 cm belowground) of the four ecosystems suggests that indeed miscanthus extracted the most water from this deep layer (Fig. S1). After several years of relatively similar responses among the four ecosystems, the drought in 2012 caused a divergence in the water balance (Fig. 2). While these responses suggest surpluses and deficits in the water balance, it is important to consider that periods when surpluses occur do not directly translate into excess moisture in the area, but instead lead to losses from surface and subterranean flows. Our results also demonstrate that during an extreme drought, miscanthus has the largest flux of water transport of deep soil water content toward the atmosphere (refer to Fig. 3, ET values).
In 2013, the drought recovery year showed strong differences in cumulative ET between miscanthus and the other ecosystemslikely a response to the excessive water use in 2012 resulting in depleted soil moisture, and thus longer recharge of soil moisture. Indeed, switchgrass and prairie recovered to once again have a cumulative surplus of water [ Fig. 2, Cum(P-ET)] and had a similar evapotranspiration rate to those observed predrought (Fig. 2, Cum.ET). On the other hand, miscanthus showed lower postdrought ET, suggesting possible lag in recovering from the previous year's drought. The difference in the cumulative P-ET between miscanthus and the other perennials in 2013 suggests a greater amount of water needed by miscanthus and for soil recharge. The soil moisture content at 100 cm belowground confirms that by June 2013 the soil at the miscanthus plot reached nearly the same water content (0.3 wfv) as switchgrass (0.35 wfv). It is likely that despite the wet conditions of this year, the very low water availability at the miscanthus field at the start of 2013 limited the ET (and productivity as discussed below), and this suggests that the majority of the precipitation is used to recharge the soil under the miscanthus field in 2013.
In fact, when considering the drought response behavior, the categorization of species as (an)isohydric is based on leaf water potential measurements by definition; however, the observed response of miscanthus relative to the other ecosystems is consistent with anisohydric responses, whereby a species tends to continue evapotranspiring despite experiencing conditions that cause isohydric species to close their stomata. The apparent drought-avoidance water-saving strategy early in the drought allowed for miscanthus to maintain productivity later than for the other ecosystems (e.g., Fig. 4), but at the expense of significant water use later, at the peak of drought, following a drought-avoidance water spender strategy at this time (Table 2). Once the pool of water begins to deplete, the rates of ET will necessarily decline, and likely causing an eventual drop in productivity, as observed later in the drought. The response of the other ecosystems, which tend to follow a response more typical of isohydric species, shows relatively less variation during the drought year and a decline in productivity as soon as the drought conditions intensified. While there was a discrepancy in the water consumption of maize and switchgrass/prairie in the early time of drought, all these ecosystems followed a water-saving strategy when drought intensified, suggesting stomatal regulation in response to increasing VPD. These results highlight that the differences between what are likely isohydric vs. anisohydric species are relatively small during nonstress years but that the differences are amplified during a drought.
The observed stress resistance strategies can be explained by understanding the dynamics of instantaneous water and carbon fluxes of the ecosystems (Fig. 4). At the beginning of drought (Phase 1: DOY <170), VPD reached~3 kPa, which has been reported to trigger stomata closure of several species (Li et al., 2003;Vitale et al., 2007;Aires et al., 2008;Yang et al., 2012). After the water spender behavior (high ET) of switchgrass and prairie at the early stage of drought, the  (Figs 2 and 4) suggests that these species indeed regulated stomata conductance due to the increasing VPD and decrease in soil moisture, consistent with isohydric responses. On the contrary, miscanthus prevented water loss until the drought intensification, following a water saver strategy in the earlier phase of the drought (VPD~3 kPa). Our results are consistent with what has been shown for miscanthus under controlled environmental conditions to have no significant change in stomata conductance during mild drought, but a vigorous response of stomata closure during severe drought conditions (Clifton-Brown & Lewandowski, 2000). At DOY 170, the rapid increase in water deficit observed for miscanthus was due to exceptionally large ET until DOY 220. During this time (Phase 2; DOY 170-200), ET of miscanthus ranked among the highest observed for all ecosystems for all years with a peak of 8 mm at DOY~200. This high ET was driven by the peak of VPD, reaching a maximum of 5.1 kPa, combined with extreme high temperatures up to 40°C between DOY 170 and DOY 220.
Although stomata closure to extreme drought is expected and has been measured before (Ings et al., 2013), we believe that our results were additionally influenced by an effect of extreme high temperature (in addition to VPD reaching 5 kPa), which might have triggered the large peak of instantaneous ET for the benefit of a cooling effect, and following a heat stress avoidance strategy at the peak of the drought. In comparison, switchgrass and prairie followed a water saver strategy in response to higher VPD (for prairie refer to the rate of increased water deficit in Fig. 2, or instantaneous ET presented in Fig. 3), or by chemical signaling originating from the roots upon the detection of decreased soil moisture (Schachtman & Goodger, 2008). Crafts-Brandner & Salvucci (2002) reported that leaflevel transpiration rates in maize increased progressively with leaf temperature and peaked above 40°C, which they suggested indicates that stomata closure was not a factor at higher temperatures. This response is inconsistent with our maize genotypes, but similar to what we observed for miscanthus under field conditions and at the canopy scale. However, our results were more likely complicated by the impact of high VPD, heat stress, and water stress together. Furthermore, previous studies are also generally limited to young plants and/or limited soil depth, whereas the deep roots of miscanthus likely play a large role in its observed drought response. Mann et al. (2013a,b) suggested that miscanthus employed a drought tolerance strategy by holding back above-and belowground biomass production, while switchgrass employed a drought-avoidance strategy of growing roots deep into regions of available soil moisture to cope with increasing surface soil moisture deficit. However at our field, where miscanthus likely already reached a mature stage (4 years old), its deep roots access water that other species could not, and with a water spending strategy typical to anisohydric species, led to the highest water deficit of this species. Aires et al. (2008) estimated a potential ET, representing the maximum expected ET from a wet soil-plant surface to be as high as 7 mm for a C3/C4 ecosystem, which would support our results in case of miscanthus lacking the detection of drought by its root system. After DOY 200, during the peak of the drought, the soil moisture that miscanthus was able to reach in deeper soil layers likely became progressively limiting, inducing a lack of water supply for ET (Phase 3; DOY 200-220). This likely resulted the leaves to become heat stressed with reduced NEP. The loss in productivity associated with miscanthus at peak drought in response to combined heat and drought stress is supported by Ghannoum (2009), who described a three-phase response to drought of C4 species; Phase 1 mainly controlled by stomata, which may or may not result a decline in CO 2 assimilation rates, followed by a mixed stomatal and nonstomatal Phase 2, and finally a nonstomatal Phase 3, when reduced enzyme activities, early senescence and nitrate assimilation play a dominant role. After this extreme portion of the drought, the cumulative P-ET reached a stable minimum point (approx. À275 mm in the case of miscanthus; Phase 4, DOY > 220), followed by recovery associated with precipitation events. Note, that while the largest water deficit during the drought for prairie was similar to miscanthus, the dynamic of ET of prairie was more like that of switchgrass, having an increase in the ET starting early in the season, probably compensating for the lack of precipitation and high temperatures.
Enhanced net ecosystem productivity and reduced wateruse efficiency of miscanthus in the drought year The combination of severe drought and heat stress resulted in a decline of NEP for all ecosystems in the second half of the summer in 2012 (Fig. 3), but the annual cumulative NEP and GPP were still large for the perennials, with the highest annual yield for miscanthus (Table 1). Previous studies have concluded that miscanthus (Miscanthus 9 giganteus in particular) is sensitive to limited water availability (Clifton-Brown & Lewandowski, 2000;Ings et al., 2013;Mann et al., 2013b), which has a strong negative effect on the species' yield production (Heaton et al., 2004). In a glasshouse experiment, miscanthus was reported with 56-66% reduction in biomass production due to drought conditions (Mann et al., 2013a). On the contrary, during a severe drought experiment miscanthus was suggested to employ a drought tolerant strategy; that is, it continued to function despite water stress indicating the lack of drought adaptation (Ings et al., 2013). In our case, the unexpectedly high annual cumulative NEP and GPP were due to the long growing season in 2012 and the deep soil water availability, which also explained the contradiction with the declined NEP during the drought stress (Fig. 4). The high temperature (40°C) and depleting soil moisture during our field experiment suggest that higher temperatures may be inhibiting photosynthetic carbon uptake by enzyme breakdown, or stimulating autotrophic respiration. Although C4 plants have a higher temperature optimum than C3 plants, photosynthesis is usually inhibited when leaf temperature exceed about 38°C (Berry & Bjorkman, 1980;Edwards & Walker, 1983), due to the inactivation of Rubisco (Edwards et al., 2001;Ruiz-Vera et al., 2013). As a result of altered ET and NEP during the 2012 drought episode, ecosystem WUE (WUE eco = NEP/ET) of miscanthus was slightly reduced (Fig. 3) relative to other years but was higher than other species. Large intraspecies variation is expected depending on the temperature and drought tolerance of the particular species, supported by our previous observations of these ecosystems (Joo et al., 2016). While miscanthus maintained relatively high productivity until the end of the 2012 growing season, switchgrass and prairie had a clear shift in NEP, with much earlier decline in their productivity in 2012 than in 2011 (Fig. 3). The WUE eco of these two perennial species were much less influenced by the drought, which suggest that the faster response of switchgrass and prairie to drought may provide a more stable performance in the long term. In 2013, the growing season immediately following the drought, all perennials including miscanthus provided similar NEP and WUE eco than in 2011, which was a decline for miscanthus in comparison with the 2012 values ( Fig. 3 and Table 1). Postdrought, miscanthus was assumed to be recovering from the drought based on the reduced ET rates relative to other nondrought years, unlike the other perennials during this relatively cool and wet summer. This could explain the drop in the productivity of miscanthus compared to the previous year, and to the long-term trend of increasing annual yields. While miscanthus still provided the largest annual GPP and NEP among the studied ecosystems in 2013, a larger impact is expected in the case of sequenced drought years. Overall, the perennials, especially miscanthus demonstrated better WUE eco than maize and soybean (Table 1).
During the nonstressed years (2009)(2010), there was no clear trend in VPD normalized ET, although miscanthus typically had lower values than switchgrass and prairie (Fig. 5). This relationship is a proxy for integrated canopy conductance (Bernacchi & Van-Loocke, 2015), suggesting that miscanthus may have a lower canopy conductance than the other species. During the 2 years where precipitation was limiting (2011 and 2012), including the drought of 2012, the overall ET/VPD was lowest for all species, pointing at a larger withholding of water these years. An alteration among the ecosystems is also noted, miscanthus having the largest VPD normalized ET among the ecosystems, which is consistent with our results suggesting that miscanthus lacks stomata regulation during the onset of drought. All species recovered in the year following the drought, with values higher than any other year, but miscanthus again showed a lower value than the other ecosystems. Despite this, intrinsic WUE GPP based on VPD normalized ET was higher for miscanthus in the postdrought year than for the other species (Fig. 5d). In the case of prairie and switchgrass, this can be explained by the rainy growing season in 2013 (low temperature, high humidity, low VPD), as their annual ET and GPP values were similar to previous years (Figs 2 and 3). On the other hand, miscanthus had additional effects from declined GPP and ET besides the low VPD in 2013. The higher intrinsic WUE GPP of miscanthus also suggests that the canopy conductance of CO 2 to photosynthesis is less limiting for miscanthus than for the other ecosystems.
The 2012 growing season was exceptionally dry and warm in comparison with long-term averages, with maximum daily temperatures reaching 35-40°C for over a month, which resulted to a peak VPD up to 5 kPa. As a consequence, the drought of 2012 resulted in a large divergence of ET and water deficit among the ecosystems, when all species experienced water deficit, and miscanthus showing divergent responses relative to the other three ecosystems. While switchgrass and prairie showed a gradually enhanced transpiration earlier, miscanthus had a rapid increase leading to a peak of ET later in the drought. At the same time, when temperature reached 40°C, a drop in NEP was observed, which was reversible after the heat stress was over. Miscanthus, with a more extensive root system likely allowed this species to access deep soil moisture during the drought, which in combination with the large ET resulted in the eventual exhaustion of soil water content, a response consistent with anisohydric species. Overall, miscanthus provided the largest annual NEP, which was due to the early start of the growing season and maintained productivity over the drought relative to the other ecosystems. Although WUE of miscanthus declined slightly during the drought, this ecosystem still provided the best water-use efficiency among the studied ecosystems, independent of the WUE metric. While one can argue that the productivity of miscanthus was still the highest among the crops after the drought, this ecosystem was observed to be much more sensitive to severe drought (large water loss and exhaustion of soil moisture during the drought, and reduced productivity in the following year), in agreement with controlled environmental studies. Therefore, a continued decline in both the productivity and the enhanced ET is assumed for miscanthus in sequenced drought episodes, while the other crops might follow a direct (short term) response with stable performance in the long term. While our study showed that a wet year following the drought can recharge soil moisture and recover miscanthus, our results of high ET during the drought suggest that a potential ecosystem-climate feedback with large-scale establishment of miscanthus throughout regions exposed to prolonged drought may intensify drought occurrence through maintained supply of moisture to the atmosphere. However, miscanthus could still provide good ecosystem service at wet regions not exposed to extreme high temperatures.