Authors: Jose L. Rotundo, Chris Zinselmeier, Nick Hoffman, Caner Ferhatoglu, Esteban Jobbagy, Paige Oliver, Lucas Borras
Categories: Article, Genetics, Plant sciences, Environmental sciences
Source: Scientific Reports
Authors: Jose L. Rotundo, Chris Zinselmeier, Nick Hoffman, Caner Ferhatoglu, Esteban Jobbagy, Paige Oliver, Lucas Borras
Agricultural water productivity, defined as the amount of grain produced per unit of available water, is an important sustainability criterion in modern crop production. Although genetic progress in maize has increased crop yields, there are limited studies on the role of plant breeding on this sustainability metric. In the center of the US corn-belt, maize rainfed yields have more than tripled since 1950 while relying on the same amount of water inputs. Our analysis shows that this shift in grain production resulted in a water productivity increase of 0.191 kg ha^−1^ mm^−1^ year^−1^, corresponding to a relative rate of 4.2% year^−1^, and is the result of both higher biomass productivity and increased harvest index (i.e., the proportion of grain in total crop biomass). The comparison of 61 historical maize genotypes commercially released for farmers in the region since 1934, throughout 28 individual experiments conducted over a period of nine years in which the effects of genetic shifts were individualized, shows that the genetic component of this increase was 1.9% year^−1^, representing 45% of the total gain. This positions plant breeding as a major technological contributor in the generation of more water productive rainfed cropping systems, reducing the need for additional agricultural freshwater to meet increasing societal demands.
Agricultural irrigation consumes 42% of all freshwater withdrawals in the US^1^, and producing more grain per unit of available water both under irrigation and rainfed conditions, is commonly proposed as a sustainability target^2^. Improvements in this sustainability dimension would require less freshwater to produce the same amount of product. Urban and business demand for water is increasing^3^, escalating the competition between water consumers. Simultaneously, aquifers are being depleted throughout the world^4^, surface waters are shrinking^1^, rainfall patterns are changing^5^, freshwater quality is on decline^6^, and interstate or international water compacts restrict the water use for certain groups^7^. Given these challenges, understanding the relevance of different technologies that help increase agricultural production efficiency of freshwater resources is needed for developing more sustainable food production systems.
Water productivity under rainfed conditions is particularly critical for maize. Over one-third of the world’s production of this crop takes place in the US, primarily concentrated in the Midwest corn-belt. The US increased production from 70 to 390 million metric tons of maize grain from 1950 until 2023^8^. While a part of the expansion took place on irrigated land, the majority of this production increase was explained by growing yields under rainfed conditions, which currently represent 85% of total US maize outputs. Hence, over the last decades maize has played a secondary role in total agricultural water use and groundwater depletion across the US (9.9 and 10.8% of total national figures^9^). The Midwest corn-belt rainfed production system uses farm mechanization, crop protection products, fertilizers, genetic hybridization, drainage systems, and digital technologies that have synergistically evolved over time. While the combination of all these technologies allowed farmers to increase their yields (as the outcome of genotype, management, and environmental interacting effects), the relative weight of these components explaining yield increases and their possible interaction with water availability has not been explored.
The description of important commercial genotypes released to the market on different times, commonly referred to as ERA experiments because of their representation of different breeding history timepoints, has identified many direct and indirect effects of breeder selection for improved grain yield, providing insights into potential mechanisms of yield improvements^10,11^. Despite these advances, there are still many gaps in our understanding on the impact of breeder selection on yield. Particularly critical for water resource sustainability is our limited understanding of the impact of long-term breeding on regional US maize water productivity (kg of grain mm^−1^ of available water). Reyes et al.^12^ demonstrated US maize grain yield under drought stress has improved associated with increased water productivity, with no changes in total water use during the crop cycle. However, Reyes et al.^12^ did not report maize water productivity explicitly, and trials were conducted outside representative commercial production environments. Other studies showing improved water productivity with newer genotypes have since been reported with genotypes representing a range of year-of-release in South America^13,14^ and China^15^. However, there are no reports describing genetic gain for maize water productivity in the US corn-belt tested in environments where genotypes were selected and intended to be grown, or reports comparing the relative contribution of breeding to total water productivity gains realized at the farm level in the region. Investigating genetic gains for water productivity is important to understand the overall contribution of plant breeding to the cropping system changes that improve efficiency and reduce pressure on water resources.
We first estimated gains for yield and water productivity since 1950 at the regional level using historical weather and USDA yields, where this total gain estimation encompasses the joint contribution of genotype (G), management (M), and environment (E) factors. We focused on the US state of Iowa because of its relevance in US maize production and because it is a rainfed production system (Fig. 1a). For estimating the genetic gain (G component only) for yield and water productivity out of the total gain (outcome of G x M x E), we used 28 trials conducted in Iowa from 2015 to 2023 having 61 ERA genotypes released for commercial sale from 1934 to 2020, all well sold and commercially targeted for this region.Fig. 1 Maize production in the US and its increase in grain yield and water productivity. (a), Average maize production acreage in the US at the county level from 2019 to 2023, with the state of Iowa highlighted in red. (b), Description of the average Iowa yield progress per year from 1950 until 2023. (c), Description of summer water availability over time for the state of Iowa from 1950 until 2023. (d) Description of the average water productivity (kg of grain ha^−1^ mm^−1^ of water) for the state of Iowa. Water availability was determined as the rainfall from 60 days prior to planting until physiological maturity).
Maize grain yield across the state of Iowa increased at an average rate of 126 kg ha^−1^ year^−1^ from 1950 to 2023 (Fig. 1b). In absolute terms, the average Iowa yield has increased > fourfold over this timeframe, from 2951 kg ha^−1^ in 1950 to 12,149 kg ha^−1^ in 2023 (312% increase). This corresponds to a relative annual rate of total gain for grain yield of 4.27% year^−1^ over 74 years of maize grain production. During the same time, water availability estimated with rainfall from two months before planting to crop maturity remained constant, increasing at a non-significant rate of 0.709 mm year^−1^ from 1950 until 2023 (Fig. 1c).
Maize water productivity in the state of Iowa increased at an average rate of 0.192 kg ha^−1^ mm^−1^ year^−1^ from 1950 to 2023 (Fig. 1d). In absolute terms, Iowa water productivity has increased from 4.5 kg ha^−1^ mm^−1^ in 1950 to 18.9 kg ha^−1^ mm^−1^ in 2023 (307% increase). This corresponds to a 4.20% year^−1^ relative annual rate of water productivity total gain.
The ERA trials distributed across the study region (Fig. 2a) showed that maize grain yield increased by year of commercial release at a rate of 128 kg ha^−1^ year^−1^ from 1934 to 2020 (Fig. 2b), similar to previous reports^10,16,17^. In absolute terms, the average genotype yield has increased > threefold, as genotypes released in 1934 are today yielding 4699 kg ha^−1^ while the most modern genotypes released in 2020 are yielding 16,091 kg ha^−1^ (242% increase). This increase corresponds to a relative annual rate of genetic gain for grain yield of 2.7% year^−1^ over 87 years of release of the tested genotypes.Fig. 2 Increase in yield and water productivity for 61 genotypes released to the market since 1934. (a) Location of experiments (red dots) across the state of Iowa conducted from 2015 to 2023 (n = 28). b, Average yield progress, where each point is a genotype released commercially since 1934. (c) Water productivity changes for genotypes released to the market since 1934 using available water from seasonal rainfall (from 60 days prior to planting until physiological maturity). Each point is a genotype, black dots and black line represent the average of all tested sites, dark grey dots and dark grey line represent the sites with 20% lower than average water availability, and the light grey dots and line represent the WUE in sites with 20% higher than average water availability. In (b) and (c) the open-pollinated landrace, Reid Yellow Dent, is included only as a baseline reference point prior to Pioneer hybrid maize breeding (first genotype of the ERA set used in our study was released for commercial sale in 1934). Reid Yellow Dent is shown with a release of 1920 (although it was available and used widely by farmers from the 1890’s through the 1930’s) and is described in figures with parenthesis since it is not included in the regression analysis. (b) provides a generalized visual representation of genotype corn morphology from the 1930’s and 2020’s.
Maize water productivity increased by year of genotype commercial release at a rate of 0.305 kg ha^−1^ mm^−1^ year^−1^ from 1934 to 2020 (Fig. 2c). In absolute terms, maize water productivity increased > threefold, from 11.5 kg ha^−1^ mm^−1^ in 1934 to 38.7 kg ha^−1^ mm^−1^ in 2020 (236%), corresponding to a relative annual rate of genetic gain for water productivity of 2.7% year^−1^ over 87 years of plant breeding.
Estimating water productivity using only seasonal rainfall (Fig. 2c) may be less accurate than other available methods because seasonal rainfall does not directly account for seasonal crop water use. To assess this possibility, and its potential impact on the results, we also conducted a water balance approach to estimate seasonal water productivity for all ERA field trials by quantifying planting water availability, physiological maturity water availability, and seasonal rainfall to enable a direct comparison of two distinct methods on water productivity outcomes. Average genotype water productivity shown by this water balance method increased by year of commercial release at a rate of 0.307 kg ha^−1^ mm^−1^ year^−1^ from 1934 to 2020 (Suppl. Inf. Figure 1), a value very similar to water productivity calculated with rainfall of 0.305 kg ha^−1^ mm^−1^ year^−1^. Likewise, a water productivity estimate with this alternative approach to calculate water availability showed little difference for the average genotype water productivity increase from 1934 to 2020 (both > threefold).
Genetic gains in yield and water productivity interacted with water availability (Supplementary Tables 3 and 4; Fig. 2c). At higher water availabilities the yield gain was higher and water productivity genetic gains were lower than average (Fig. 2c). The opposite happened whenever water availability was lower than average, reducing crop yields but increasing water productivity genetic gain estimates. This result shows water productivity gains from breeding programs were greater in years with limited water availability, helping to mitigate the impact of drought stress on grain yield.
Calculating the ratio between genetic gain over total gains realized at the farm level provides an estimate of the importance of breeding (G) relative to the other plausible yield increase factors, such as environment (E) and/or management (M). Using the full ERA set of genotypes provides the best estimate of genetic gain for maize water productivity. However, to make direct comparisons between genetic gains and progress realized at the farm level (USDA data) for the period 1950–2023 (Fig. 1c), we limited the genetic gain calculation for water productivity to the same time-period. The y-intercept for water productivity of a 1950 commercial release was 16.4 kg ha^−1^ mm^−1^ (Fig. 2c), corresponding to a 136% increase in water productivity from 1950 to 2020 and a relative annual rate of genetic gain of 1.86% year^−1^. Given a total gain realized at the farm level of 4.20% year^−1^ and a genetic gain of 1.86% year^−1^, genetic improvement contributed 45% of the total gain for water productivity.
A limitation in our water productivity analysis is that we did not measure water use at the individual genotype level, assuming year of release has not changed water use at the individual genotype and has remained constant. To test this assumption, we compiled all the studies using ERA maize genotypes where water use and yield were measured in field situations.
Figure 3 describes yield and water use of individual maize genotypes with different year of release from our breeding program^12,18^ and from other independent global breeding programs^13,14,19^. Results show long-term breeding has not increased water use, but very slightly reduced it (Fig. 3a). At the same time, the observed yield trend from these studies (around 2% year^−1^) is like those described in the current study (Fig. 2b). As such, independent studies using different methods to calculate water use at the individual genotype level support the concept that maize plant breeding is consistently increasing the crop yield without increasing the water use by the crop.Fig. 3Yield and water use at the individual genotype level. (a) Relative yield changes as a function of the time difference in genotype release to the market. (b) Relative water use changes as a function of the time difference in genotype release to the market.
Meeting future food demands while minimizing expansion of cultivated area depends on continued intensification, requiring further grain yield increases per hectare^20,21^. Yet, another equally important challenge is to minimize the irrigation needs given the critical depletion of agricultural water resources globally^22^. In our study, conducted at one of the most productive grain-belts of the world, we isolated the contribution of genetics (by testing the same genetics across experiments) from other factors (crop management and environment) on regional improvements in water productivity, an increasingly relevant data-driven sustainability criterion. We demonstrated that 45% of the total water productivity increase in rainfed maize production on this relevant region is explained by commercial plant breeding. The 55% non-genetic residual is accounted for by the many management and environmental changes that occurred during this period that also contributed to increased water productivity. These changes include more efficient crop protection products, improved soil drainage (thereby reducing flooding events and permitting more timely plantings), optimized plant populations, modified tillage systems (such as minimum tillage that reduce direct soil water evaporation and increase water infiltration), and higher fertilization rates, in addition to weather or climate trends that have favored crop yield increases.
Water productivity, as defined in this work, has been widely used by others^23–26^. Despite its prevalence in the literature, this metric has some constraints compared to alternatives that can be applied in smaller trials. Given the scale of the experimental approach reported here, it was not feasible to estimate water productivity at the plot level. The first constraint is that it overlooks differences in water use across genotypes. While we acknowledge this issue, we have provided evidence from the literature showing that in a set of five independent publications, the water use of different ERA maize genotypes did not change (Fig. 3). This dataset also included a subset of genotypes tested in our work. A second constraint is that our metric does not account for water table supply. The areas where the trials were conducted are indeed influenced by the water table^27,28^, and there is evidence that the effect of the water table interacts with rainfall, serving as a buffer^29^. This might have been the case in our studies, where sites with relatively low precipitation achieved high yields. In situations of low precipitation and water table influence, we acknowledge that the estimation of water productivity might be inflated. However, this does not negate the influence of breeding on increasing water productivity.
Most studies describing future climate impact on agricultural production face the problem of their inability to predict the dynamic technological adaptations that will occur at the cropping systems level^30^, generating yield predictions that have shown to contradict observed field outcomes^31^. Advances in agricultural technologies are directional and unavoidable, but mostly unpredictable^32^. Previous studies have shown that modern maize genotypes have increased drought stress tolerance compared to older genotypes^17,33,34^. In the present article we expand this knowledge to show how maize breeding technologies delivered commercial genotypes with consistently higher yield per unit of available water in their target environments and showcase how this genetic improvement was realized at the production systems level. We also showed that higher water productivity is especially evident in years with lower-than-average rainfall. In this context, it should not be surprising that newer maize genotypes are likely more resilient to the future environments because weather projections commonly predict higher probabilities of water deficits^34–37^.
The water productivity gain for Iowa was 0.191 kg ha^−1^ mm^−1^ from 1950 to 2023 (Fig. 1c) and, to our knowledge, it is the first long term quantitative report for water productivity in the Central US corn-belt. Riccetto et al.^23^ reported a mean precipitation-based water productivity of 13.9 kg ha^−1^ mm^−1^ for Iowa for a smaller period (1995 to 2012). Our estimated average water productivity for this same period (1995 to 2012) is 14.8 kg ha^−1^ mm^−1^ for Iowa, close to their reported value.
Evaluating 61 maize genotypes spanning 87 years of breeder selection describe a large water productivity range (Fig. 2c). The oldest genotypes evaluated in the current study, released for commercial sale in 1934 and 1936, showed water productivity values of approximately 10 kg ha^−1^ mm^−1^ whereas genotypes released after 2015 demonstrated water productivity values higher than 35 kg ha^−1^ mm^−1^. Without this genetic gain in water productivity Iowa would require 3.5 times more land under production to reach the same level of grain produced today. Genetic gain for water productivity was 0.305 kg ha^−1^ mm^−1^ year^−1^ since 1934 (Fig. 2c), and similar genetic water productivity gains were obtained by Liu et al.^15^ in China, with gains of 0.31, 0.38, and 0.72 kg ha^−1^ mm^−1^ year^−1^ across full irrigation, moderate-drought, and intense-drought water treatments, respectively, based on nine maize genotypes adapted to their region. Likewise, maize genotypes adapted to South America and spanning 32-years of breeding progress also showed improved water productivity with breeding^13,14^. Collectively, these results demonstrate that while the development of genotypes with greater water productivity has rarely been a direct breeding target, selection for yield has shown clear positive long-term breeding effects on yield per unit of available water globally.
Many agroecology concepts linked to higher resource use efficiency at the cropping systems level focus on crop diversity and crop rotations as possible solutions^38^. However, differential long-term research and development investment in some specific crops compared to others need to be considered when optimizing the use of resources at a regional level. It is important to bear in mind that currently, optimal whole-farm water productivity may be found with less diverse farming landscapes dedicated to the few available crop species for which breeding has achieved maximum efficiencies. For example, sorghum has traditionally been considered a drought tolerant crop, but today modern maize, under appropriate agronomic management, can be more drought tolerant than modern sorghum in the US^17^ and evidence from South America also shows maize out-yielding sorghum in the more drought prone environments in recent years^39^. In fact, the observed large maize water productivity changes over time (Fig. 2) is not to be expected in all cultivated species due to the differential investment maize breeding has been receiving compared to other crops. We estimate that in the last 40 years maize breeding at Corteva Agriscience, and its heritage companies, consistently had more than a tenfold investment compared to sorghum breeding across a much larger geography in the US and globally. This contrast indicates that increased breeding efforts across multiple crop species are needed to avoid narrowing the diversity of cultivated landscapes when higher yields per unit of available water becomes a key target.
Because our water productivity estimate is grain per unit of water, it could be hypothesized that changes are not related to the capacity of modern crop genotypes to produce more biomass per unit of available water but simply an artefact of harvest index changes^40^. This is true because harvest index has increased on average from 0.47 to 0.54 from 1960 until today^41^, but this change cannot explain the large differences we observe in water productivity during the same time frame (Fig. 2). Other recent results have shown that long-term breeding has increased crop biomass across a range of water environments^17^, and that yield changes are not only a consequence of harvest index but to crop biomass production to a similar degree.
Our study did not estimate the interaction between genetics and management. Technologies are commonly combinatorial, and farmers identify and profit from positive interactions (e.g., plant population x N fertilization impact on yield, where the combination of both management decisions provide higher yields than adding any of them separately). Long-term plant breeding effects on water productivity were estimated under recent cropping systems management in the US mid-west, assuming yield and water productivity would provide similar values over time. Since modern genotypes respond in yield more to higher input levels than older ones (like with nitrogen^42^) their true effect might be slightly different than described here. Water availability might have also been affected by management, like minimum tillage reducing direct soil water evaporation and increasing water infiltration^43^. Lastly, water table levels close to soil surface affect crop water availability and can be directly affected by drainage changes or more indirectly affected by land use changes^29^. However, we obtained a very similar water productivity estimated when using rainfall alone or using a water balance approach (with our satellite-based water availability estimates on in-season soil moisture differences). While we realize some of these effects could be modifying our reported values, they would not affect the general trends reported here.
To summarize, maize long-term breeding for yield is allowing more efficient production systems in terms of water use. The influence of commercial plant breeding on maize water productivity has been commonly overlooked. The impact of breeding technology has reduced the need for additional agricultural land or freshwater to be diverted to meet modern societal demands, helping to achieve more resilient cropping systems to rainfall shortages or higher evapotranspiration demands, delaying the approach to any water ceiling in maize production globally.
Maize yield survey data, including all production practices, from the 99 counties of the state of Iowa over 74 years (1950–2023) were obtained from the USDA National Agricultural Statistics Service^44^. Data was retrieved using the function nassqs() from package ‘rnassqs’^45^ using R software^46^.
Centroids (latitude and longitude) of each Iowa county were extracted using the function `st_centroids” from “sf” package^47^. Water availability, using precipitation during the growing season as a proxy, was obtained from an internal Corteva weather data pillar connected to IBM weather services for the 1950 to 2023 period. The growing season was defined as 60 days prior to planting until physiological maturity. The 60 days prior to planting were used to account for initial soil water that might be potentially available during the season and that would not have been accounted for if the analysis would have started at planting date.
Water productivity (kg mm^−1^) was estimated as the ratio between seed yield and water available during the growing season. This metric, while it may present some limitations (see discussion), has been widely used to characterize environments or genotypes^23–26^ when there are logistical constraints in estimating water use at the individual plot level for many genotypes. Its widespread use is largely due to its practicality and the ability to provide a general overview of water productivity across different conditions. Despite its limitations, it serves as a valuable tool in large-scale studies where detailed measurements are not feasible.
Percent progression of maize planting and harvest to estimate season duration were retrieved from USDA-NASS for the state of Iowa. No information was available at the county level. Variables retrieved from USDA-NASS were ‘CORN—PROGRESS, MEASURED IN PCT PLANTED’ and ‘CORN, GRAIN—PROGRESS, MEASURED IN PCT HARVESTED.’ Planting and harvest were defined when they reached 50% progression. The period covered for this estimate was 1981 to 2023. For years before 1980, planting date was estimated as the average planting date for the period.
Iowa grain yield, water availability during the growing season, and water productivity (y) were averaged across county centroids and modeled as a function of year of production (Yi) using a linear regression model (lm()) in R version 4.3.2^46^. Total gain values for yield (kg ha^−1^ year^−1^), water availability (mm year^−1^), and water productivity (kg ha^−1^ mm^−1^ year^−1^) were obtained as the estimate of slope of the linear model.
Sixty-one maize commercial genotypes, released for commercial sale between 1934 and 2020, and one open-pollinated landrace (available in 1920), were selected for field testing (Supplementary Table 1). Twenty-eight individual experiments were conducted across Iowa from 2015 to 2023 (Supplementary Table 2) with a planting population of 84,000 plants ha^−1^. Crop management practices were common to the area, fertilization was applied following local recommendations and plots were kept clean of weeds, pests, and diseases. All experiments were rainfed, none were irrigated. Experiments were planted between 22-Apr and 23-May, with a median planting date of 5-May. Plots were mechanically harvested for yield determination. Experiments were two row plots 5.3 m long with a row spacing of 0.76 m. Genotype entries were randomized within each location. Of the 28 experiments, only three experiments were replicated with 2 blocks per location. The remaining experiments had only one replication.
For each site (defined as the combination of year, location, and planting date), seasonal rainfall data was obtained 60-days prior to the planting date to the date of physiological maturity, defined as 1500 Growth Day Units (GDUs) after planting, using 8 °C as base temperature. Cumulative rainfall was calculated for this period, for each site, and used in further analysis. With this water availability we estimated water productivity, calculated as the ratio between yield and water availability.
For the ERA trials, a second estimate of water productivity was calculated using a water balance approach. The water balance was calculated as soil water availability at planting, plus rainfall from planting to physiological maturity, and subtracted soil water availability at physiological maturity. The soil water availability at planting and physiological maturity was retrieved from satellite downscaled SMAP water availability^48^. This product provides volumetric soil water content for the 1 m upper profile. The upper and lower volumetric soil water limits used to calculate available soil water were obtained from USDA National Resource Conservation Service^49^. Since results were similar using both approaches for calculating water availability, only results using the former will be discussed.
Grain yield and water productivity (y) were modeled as a function of year of release (Yi), water availability (Wj), year of release by water availability (YxWij), genotype nested in year of release (H/Yik), site (Sl), and blocks nested in site (b/Slmn). Parenthesis for model terms in Eq. 2 indicate random 1\documentclass[12pt]{minimal}
\usepackage{amsmath}
\usepackage{wasysym}
\usepackage{amsfonts}
\usepackage{amssymb}
\usepackage{amsbsy}
\usepackage{mathrsfs}
\usepackage{upgreek}
\setlength{\oddsidemargin}{-69pt}
\begin{document}$$
{\text{y}}{{{\text{ijklm}}}} = {\text{Y}}{{\text{i}}} + {\text{W}}{{\text{j}}} + {\text{YxW}}{{{\text{ij}}}} + \left( {{\text{H}}/{\text{Y}}{{{\text{ik}}}} } \right) + \left( {{\text{S}}{{\text{l}}} } \right) + \left( {{\text{b}}/{\text{S}}_{{{\text{lm}}}} } \right)