The U.S. Midwest and High Plains Aquifer-fed croplands are previously unrealized hotspots of extreme evaporative demand exposure

Total evaporative demand or atmospheric thirst is a primary determinant of agroecosystems’ water use and an indispensable input to scientifically based irrigation design and management. However, despite its extensive use to represent agricultural environments, it has not been assessed for its extreme behavior. Prolonged exposure to extreme evaporative demand conditions a.k.a thirstwaves can be especially stressful for agricultural output, water use, and management, but remain uninvestigated owing to lack of meaningful metrics for quantifying and reporting ‘extreme thirst exposure’. In this letter, I present spatial (county-level) and temporal (1981–2021) changes in exposure to extreme thirst during the agricultural growing season across the conterminous U.S. (CONUS). Using a fully physical metric of evaporative demand, i.e., standardized short crop reference evapotranspiration (ETo), I define two novel measures: cumulative extreme thirst exposure (thirstcum) and average extreme thirst anomaly (thirstanom) to represent the seasonal-level severity of thirstwaves. Both metrics showed significant spatiotemporal variation with long-term averages of 12 mm (thirstcum) and 0.66 mm d−1 (thirstanom) for CONUS. Distinct spatial patterns were revealed for extreme thirst exposure that had little in common with those observed for total ETo. Spatially, hotspots of high extreme thirst exposure were co-located with the Midwest and High Plains aquifer regions, that account for 64% of total acreage and 28% of irrigated acreage nationally, respectively. Critical for food and water security, these regions have experienced the highest extreme thirst exposure nationally, hence necessitating reevaluation of regional disparities in water stress. While thirstcum and thirstanom have increased by 5.6 mm and 0.21 mm d−1 on an average in CONUS, worsening of extreme thirst exposure is especially concerning for the High Plains aquifer region (12.6 mm and 0.54 mm d−1, respectively). The emergence of previously unrealized hotspots in regions critical for water security uncover potential pitfalls for planning and adaptation that may result from overlooking extreme measures of evaporative demand.


Introduction
Atmospheric evaporative demand (AED) or 'atmospheric thirst' describes loss of water from the surface driven combinedly by temperature, humidity, wind speed, and radiation.Encapsulating these atmospheric factors, AED is an effective determinant of aridity, vegetation water use, and water stress (Grossiord et al 2020, Vicente-Serrano et al 2020).Although its wide use and application extend to many functions, agricultural water management and stress assessment has probably benefitted the most from quantifying AED.Among a large range of physical and mathematical representations of AED, the combination-based ASCE standardized Penman-Monteith equation is the most widely used formulation tailored for agricultural applications (Allen et al 1998, Walter et al 2005, Jensen and Allen 2016).Within the United States, daily to annual mean AED represented by grass or alfalfa-reference evapotranspiration (ET o or ET r , respectively) are understood to be smoothly increasing from their minima in the Northeast to their maxima in the Southwest, except for impacts from local elevation (Hobbins 2016, Kukal et al 2023).Understanding mean spatial changes in ET r and ET o has benefitted many applications, including but not limited to estimating large scale crop water use ( However, like many climate variables, smooth spatial patterns observed for mean/total AED may not necessarily apply to their variability measures, especially extremes (Jones et al 2014).A conterminous U.S. (CONUS)-wide decomposition of AED showed that spatial features observed for ET r mean versus variance were distinct and had little similarity (Hobbins 2016).AED variance over shorter time scales (days to weeks) can be equally, if not more meaningful, than seasonal/annual mean behavior for sectoral impacts.For instance, prolonged periods of extremely high ET o can expose vulnerabilities in irrigation water management infrastructure and practice (Hobbins et al 2019, Parker et al 2021, Noguera et al 2022).Due to its ability to measure drought onset and persistence successfully and independently, extreme ET o is the basis for effectively predicting and describing droughts using evaporative demand drought index (Hobbins et al 2016, McEvoy et al 2016).
Despite the importance of extreme ET o for agricultural stress and water management, long-term climatology and change in extreme ET o exposure are currently unknown globally.A primary bottleneck to this status quo is lack of metrics that appropriately reflect seasonal-level extreme ET o conditions in a meaningful manner, while being comparable across space and time.Consequently, the spatial variation of extreme ET o exposure as well as its temporal evolution has not been mapped, posing important questions: What does the climatology of extreme ET o exposure look like within the CONUS and are certain regions more vulnerable than others?Furthermore, has exposure to extreme AED changed over time (how much, in what direction, and where?), and how are these changes different than those observed in mean ET o ?How do spatial differences in extreme ET o exposure and its change relate to spatial footprint of U.S. agriculture?This letter addresses these knowledge gaps by characterizing measures of cumulative extreme thirst of the atmosphere during 1981-2021 in all CONUS counties.

Data
The analysis is based on daily short crop (grass) reference evapotranspiration (ET o ) data for 1981-2021 derived from GRIDMET (Abatzoglou 2013) dataset.The original data is available at 4 km spatial resolution and was rescaled to county-level for this analysis using spatial reducer in Google Earth Engine.ET o represents standardized reference ET from a wellwatered and clipped cool-season grass such as fescue or perennial ryegrass having a range in height from 0.08 to 0.15 m, averaging 0.12 m, a leaf area index of approximately 3.0 and albedo of 0.23.The GRIDMET dataset is a hybrid of NLDAS-2 and the parameterelevation regressions on independent slopes model (Daly et al 2008) and is widely employed for applications in water resources management (Albano et al 2022, Melton et al 2022, Kukal et al 2023).
Total cropland acreage and irrigated acreage for each county reported in the US Agricultural Census of 2022 (USDA 2022) were obtained using the USDA NASS Quick Stats Database (USDA 2024).Boundary datasets for U.S. regions and the High Plains aquifer were also obtained for calculating zonal statistics.

Defining and detecting thirstwaves
Thirstwaves were defined following the widely used definition of a heatwave that is based on air temperatures (Perkins and Alexander 2013), but employing ET o as the central metric.A thirstwave was deemed to have occurred when during atleast three consecutive days, daily ET o was greater than the 90th percentile of ET o corresponding to each Julian day.The 90th percentile used in this definition was based on a 15 day moving window over a baseline period, i.e. 1981-2000.The 15 day moving window was centered on the Julian day being evaluated.ET o data during 1981-2000 served as baseline records for establishing climatologically extreme thresholds using which thirstwaves were detected through the study period.The analysis was limited to the warm season (April until October) representing an extended agricultural growing season in the CONUS.

Measuring extreme thirst
Two indicators were used to assess extreme evaporative demand exposure: cumulative thirst exposure (thirst cum ) and average thirst anomaly (thirst anom ).The indicator thirst cum describes the extra evaporative demand that is received during all thirstwaves occurring during the season.It is calculated as ET o anomaly on each thirstwave day relative to the threshold (defined in 2.2) cumulated across all thirstwave days identified during that season (equation ( 1)).
where thirst cum is the cumulative thirst exposure, n the number of thirstwave days in a season, and ET o anom is the ET o anomaly on a day during a thirstwave relative to the 90th percentile threshold.This indicator focuses on the seasonal excess thirst experienced beyond the ET o threshold.The second indicator, thirst anom represents the average contribution of each thirstwave day to thirst cum and was determined using equation ( 2), where thirst anom is the average thirst anomaly across all thirstwave days in a given season, and frequency of thirstwave days is the total number of thirstwave days detected in that season.Both measures were calculated for each growing season during 1981-2021 at the county-level.

Trend calculations
Trends in both measures were calculated for all counties following the nonparametric Theil-Sen slope estimator (Sen 1968).The resulting trends were reported as total change in the characteristic (over 41 years).The statistical significance for each of the monotonic trends was tested using the Mann-Kendall trend test (Mann 1945, Kendall 1975).

CONUS-mean exposure to extreme thirst
Long-term (1981-2021) mean thirst cum was 12 mm for the CONUS, although substantial variability was encountered in both space and time (coefficient of variation of 1.70).Long-term mean thirst cum values ranged from 3.7 to 27.8 mm across counties, signifying spatial heterogeneity.Considering individual counties and seasons, thirst cum values as high as 309 mm were observed, however, most (5th to 95th percentile) observations were limited between 0.9-40.5 mm.While thirst cum is a cumulative measure of extra ET o received beyond the extreme threshold, it may be received over widely different number of days across regions and seasons.Thus, it becomes necessary to normalize thirst cum using thirstwave frequency for a given season and crocounty.The resulting quantity, thirst anom measures the mean contribution of each thirstwave day observed to thirst cum received.On a long-term mean basis, thirst anom averaged 0.66 mm d −1 for the CONUS and ranged between 0.27 mm d −1 and 1.27 mm d −1 across counties.Across individual counties and seasons, thirst anom was as high as 2.79 mm d −1 , but most (5th to 95th percentile) values were between 0.24-1.32mm d −1 .It had a relatively lower coefficient of variation of 0.52 (sd = 0.34 mm d −1 ) than thirst cum , a consequence of adjusting for thirstwave frequency (mean of 14.9 d).

Spatial patterns of extreme versus total thirst
Mapping of long-term mean thirst cum (figure 1(a)) and thirst anom (figure 1(b)) revealed that spatial patterns of extreme evaporative demand are quite dissimilar to that of total ET o received during the same period.Rather than following a smooth northeastsouthwest increasing trend as seen in total ET o (figure S1(a)), both metrics of extreme ET o exposure were generally lower in the much of the western and eastern U.S. and were the highest in the High Plains and the U.S. Midwest (figure 1).These spatial dissimilarities underscore the importance of assessing ET o extremes in addition to mean behavior.For instance, while these regions have experienced the most consequential extreme AED exposure CONUS-wide, the region otherwise is assessed as a moderate AED environment (figure S1(a)), when assessed using total ET o .

Hotspots of extreme thirst and agricultural footprint
Areas of predominantly large cropland acreage as well as irrigated acreage are co-located with hotspots of extreme evaporative demand exposure.All counties were represented in a bivariate space between either of the two extreme thirst exposure metrics and cropland acreage (total and irrigated), each classified into three equal quantiles (figure 2).It was found that Midwestern U.S. cropland (black boundary in figures 2(a) and (c)) fell into the third quantile (dark green) for both variables, with the highest total acreage as well as highest long-term mean thirst cum and thirst anom .Additionally, High Plains aquifer region (black boundary in figures 2(b) and (d)) fell into the third quantile (dark green) for both variables.Extreme thirst exposure in these hotspots remain relatively higher than that in other U.S. regions during most years, as demonstrated by region-specific thirst anom time series (figure 3).The Midwestern region accounts for 64% of total cropland, and the High Plains aquifer region accounts for 28% of irrigated acres nationally, implying that significant portions of regions critical for U.S. food and water security are impacted by extreme thirst exposure.

Trends in extreme thirst exposure
Both thirst cum and thirst anom have predominantly increased across CONUS during 1981-2021.On an average, thirst cum and thirst anom in the CONUS have increased by 5.6 mm and 0.21 mm d −1 , respectively during the study period.The worsening of     even at wetter sites (Zhang et al 2021).As a result, they suggested that an irrigation scheme combining both soil moisture and VPD can significantly lower irrigation water use without any yield penalty, underscoring the direct importance of evaporative demand for water management.It can be argued that VPD alone cannot represent evaporative demand entirely as it does not account for radiative energy, and thus, ET o should be the preferred metric for comprehensively reporting extreme atmospheric thirst.

Discussion
The observance of severe thirstwaves in irrigated croplands motivates investigating the impact of prolonged (multiple days) periods of extreme ET o on water stress and use.Are the impacts from cumulated exposure to extreme ET o (thirstwaves) equal to or greater than the sum of its parts, i.e. daily-level extreme ET o ?If so, how do these impacts vary under absence or presence of soil moisture stress?It is likely that repeated exposure extreme thirst over multiple consecutive days (thirstwaves) impedes crop function to a greater degree relative than short-duration extreme exposure (Jackson et al 2021, Lesk et al 2022).Currently, models rely on a daily crop water stress factor (K s ) to calculate ET c by suppressing unstressed ET c , which only applies in suboptimal soil water conditions (Allen et al 1998) and not under extreme thirst exposure.Crop physiological response to contrasting thirstwaves and their mechanisms need to be investigated at appropriate scales (leaf, canopy, and ecosystem-levels) and incorporated into modeling frameworks similar to simulations of heatwaves (Ingvordsen et al 2018, Bernacchi et al 2023).Increasing exposure to extreme ET o may also impact irrigated agriculture by exposing vulnerabilities in irrigation infrastructure, and demanding producerlevel adaptation (Daccache and Lamaddalena 2010).Irrigation water withdrawal and application systems are designed based on peak water demand computed using mean ET o (Zaccaria and Neale 2014)  Extreme ET o also holds particular significance for water management in specialty crops, where irrigation is also a mitigation tool for extreme heat (Parker et al 2020).Our findings show large (and increasing) magnitudes of extreme thirst for top specialty crops producing regions such as California (figure 1).The Midwest also has a smaller but significant specialty crop footprint (Kistner et al 2018).In these systems, irrigation management intended to meet two separate goals: meeting crop water needs and buffering heat stress, while also managing for improved crop quality via regulated deficit irrigation (Stewart et al 2011).Since extreme ET o encapsulates heat stress and other environmental stresses into an integrative measure, revisiting these goals from an extreme thirst standpoint may aid in achieving greater synergy among managing soil moisture deficits and extreme thirst stress.

Need for incorporating extreme thirst exposure as a climate service for agricultural audiences
A primary reason for lack of understanding extreme thirst impacts on agriculture and water resources is that measurement, quantification, and reporting of evaporative demand has been limited to mean/aggregate values so far.Many important hydrological and environmental variables are characterized and reported for their extreme behavior to agricultural audiences, including extreme temperatures (heatwaves/freeze risk), heat accumulation (killing degree days), precipitation (meteorological droughts/floods), soil moisture (agricultural droughts/floods) and others.This has not been the case for ET o , where reporting is limited to hourlyto-daily total ET o values at sparsely distributed reference agricultural weather monitoring sites (Palmer 2008, Shulski et al 2018, Marek et al 2020).The argument that weather extremes can be more meaningful for predicting sectoral impacts is well established in the context of temperature, where extreme temperatures have received equal if not greater emphasis than average temperatures (Brown et al 2008, Lee et al 2014, Perkins-Kirkpatrick and Lewis 2020).In the context of climate change, it has been shown that small changes in average temperatures can result in disproportionately greater changes in nature of temperature extremes (Mearns et al 1984, Katz and Brown 1992, Nicholls et al 1996).Changes in mean or total ET o (like mean temperatures) are not necessarily synonymous with trends in thirstwaves (like heatwaves), since ET o is assessed as a continuous variable whereas a thirstwave event is assessed as prolonged exceedance of a defined threshold (90th percentile here).These distinctions necessitate that nearreal time as well as short-term forecasts of thirstwaves and associated characteristics/metrics are available to agricultural/environmental managers as climate services.While their impacts are investigated well, heatwaves only partially represent extreme agricultural environments because they overlook impacts from humidity, wind, and radiation.Audiences in agriculture and water management can improve preparedness against extreme thirst if thirstwave information is disseminated via climate services tailored to agriculture (Haigh et al 2018, Born et al 2021).Recent efforts such as evaporative demand drought index is one such index that leverages the skill offered by ET o for drought monitoring (Hobbins et al 2016, McEvoy et al 2016).CONUS-wide maps and time series for EDDI at different timescale are published by NOAA as an early warning guidance tool (NOAA/ESRL Physical Sciences Laboratory).However, the use of EDDI is limited to classify regions and time periods into standardized drought categories and it is not intended to quantify seasonal-level extreme thirst as a continuous metric (Lukas et al 2017), which can be accomplished by thirst cum and thirst anom .

Caveats and future work
The spatial and temporal characteristics of extreme thirst are dependent on how thirstwaves are defined, i.e. extreme ET o threshold used (90th percentile here), minimum consecutive number of days used to identify a thirstwave (3 here), baseline used (1981-2000 here), moving window used for calculating extreme threshold (15 day), and choice of dataset (GRIDMET here).A caveat is that the potential deviation in extreme thirst characteristics because of change in these choices were not evaluated in this study and are an area of future work.Second, given the co-location of extreme thirst and critical agricultural production and water resources regions revealed here, the impacts of extreme thirst metrics on countylevel crop productivity should be investigated in the future, possibly parsing their impacts from heatwaves and soil moisture stress.Lastly, resources should be invested into building interactive platforms for dissemination of extreme thirst tailored to agricultural and water managers (section 4.2).

Conclusion
This research addresses extreme atmospheric thirst by introducing the concept of thirstwaves and their measurement using appropriate seasonal-level metrics.Upon comprehensively analyzing historical extreme thirst in the CONUS, I find that exposure was the highest in the Midwest and High Plains aquifer regions, and that the spatial patterns were quite dissimilar to those demonstrated by total or mean evaporative demand.Although extreme thirst has worsened on an average in CONUS during 1981-2021, I find that trends in High Plains aquifer region are particularly alarming.The fact that these regions are extremely important for national food and water security make them especially vulnerable to thirstwaves, given their co-located nature.Prolonged periods of extreme thirst are equally if not more important for agricultural water stress and agrohydrological outcomes than total evaporative demand.In the past, extremes of evaporative demand have been virtually ignored in global agroecosystems, preventing prioritization of regions that may be disproportionately impacted by thirstwaves.With this CONUS-wide observational evidence, it is argued that extreme thirst is likely a critical determinant of water stress in U.S. agriculture and should be emphasized for better comprehension and delivery via public-facing climate services to relevant audiences.

References
Allen et al 2020), crop water productivity (Kukal and Irmak 2017), seasonal irrigation withdrawals (Sharma et al 2016), and drought monitoring (Hobbins et al 2016, McEvoy et al 2016).More recently, gridded surfaces of ET o have been used to quantify and disseminate crop water use across large scales using open data services such as OpenET (Melton et al 2022) and ECOSTRESS (Fisher et al 2020).

Figure 2 .
Figure 2. Bivariate maps showing (a) long-term (1981-2021) mean county-level cumulative thirst exposure (thirstcum) and total agricultural cropland acreage; (b) cumulative thirst exposure (thirstcum) and total irrigated cropland acreage; (c) average thirst anomaly (thirstanom) and total agricultural cropland acreage; (d) average thirst anomaly (thirstanom) and total irrigated cropland acreage.The county-level total and irrigated acreage shown are adopted from the US Agricultural Census of 2022.The black polygons in (a) and (c) show U.S. Midwest boundary and polygons in (b) and (d) shown High Plains aquifer boundary.

Figure 3 .
Figure 3.Time series of average thirst anomaly (thirstanom) averaged for the four United States Regions, High Plains (Ogallala) aquifer, and CONUS during 1981-2021.

Figure 4 .
Figure 4. Total change in county-level (a) cumulative thirst exposure (thirstcum) and (b) average thirst anomaly (thirstanom) over the period 1981-2021.The dots depict that the change is significant at the 95% confidence level for that county.
and not extreme ET o occurring over multiple consecutive days.A substantial acreage in the HPA is irrigated under limited pumping capacities (Ajaz et al 2020, Evett et al 2020), and sustained exposure to ET o extremes may exacerbate these existing limitations.With irrigation moving eastwards in the U.S. (McDonald and Girvetz 2013, Zeng and Ren 2022), many regions in the Midwest are increasingly adopting irrigation to supplement water availability during droughts, which are projected to increase in frequency and intensity (Hoell et al 2021).Not only does this adaptation become more meaningful because of the Midwest emerging as an extreme ET o hotspot, but it may also be beneficial to actively use local extreme ET o climatology for designing irrigation systems in these regions.
Abatzoglou J T 2013 Development of gridded surface meteorological data for ecological applications and modelling Int.J. Climatol.33 121-31 Ajaz A, Datta S and Stoodley S 2020 High plains aquifer-state of affairs of irrigated agriculture and role of irrigation in the sustainability paradigm Sustainability 12 3714 Albano C M, Abatzoglou J T, McEvoy D J, Huntington J L, Morton C G, Dettinger M D and Ott T J 2022 A multidataset assessment of climatic drivers and uncertainties of recent trends in evaporative demand across the Continental United States J. Hydrometeorol.23 505-19 Allen R G, Pereira L S, Raes D and Smith M 1998 Crop evapotranspiration.Guidelines for computing crop water requirements FAO Irrigation and Drainage Paper (FAO) p 56 Allen R G, Robison C W, Huntington J, Wright J L and Kilic A 2020 Applying the FAO-56 dual Kc method for irrigation water requirements over large areas of the Western US Trans.ASABE 63 2059-81 Bernacchi C J, Ruiz-Vera U M, Siebers M H, DeLucia N J and Ort D R 2023 Short-and long-term warming events on photosynthetic physiology, growth, and yields of field grown crops Biochem.J. 480 999-1014 Born L, Prager S, Ramirez-Villegas J and Imbach P 2021 A global meta-analysis of climate services and decision-making in agriculture Clim.Serv.22 100231 Boyer J S 1982 Plant productivity and environment Science 218 443-8 Brown S J, Caesar J and Ferro C A T 2008 Global changes in extreme daily temperature since 1950 J. Geophys.Res.Atmos.113 D05115 Ciais P et al 2005 Europe-wide reduction in primary productivity caused by the heat and drought in 2003 Nature 437 529-33 Daccache A and Lamaddalena N 2010 Climate change impacts on pressurised irrigation systems Proc.Inst.Civ.Eng 163 97-105 Daly C, Halbleib M, Smith J I, Gibson W P, Doggett M K, Taylor G H, Curtis J and Pasteris P P 2008 Physiographically sensitive mapping of climatological temperature and precipitation across the conterminous United States Int.J. Climatol.28 2031-64 Evett S R, Colaizzi P D, Lamm F R, O'Shaughnessy S A, Heeren D M, Trout T J, Kranz W L and Lin X 2020 Past, present, and future of irrigation on the US Great Plains Trans.ASABE 63 703-29