Gaps in assessing crop yields by focusing on photosynthetic and nitrogen-use-efficiencies (NUE)

Crop genotypes with high photosynthetic and nitrogen-use-efficiencies (defined as the ratio of the nitrogen taken up by a crop to the total input of fertilizer nitrogen) are the preferred plants to grow since the assumption was that the total amount of carbon fixed by a plant will determine its potential yields, especially when nitrogen fertilizers are applied in sufficient quantities to maintain high photosynthetic rates [17]. Research has supported these assumptions. For example, Li et al. [18] reviewed 130 publications that assessed data on yield, shoot biomass, and nitrogen concentration that suggested the genetic transformation of crops (rice, maize, wheat) impact their nitrogen-use-efficiency (NUE) in different soil fertilization conditions. The latter study showed that genetic improvement in NUE significantly increased the grain yield of crops. However, this study also reported that potted experiments have a higher yield variance than field-grown crops [18], which suggests that other factors in field experiments limit plant growth and carbon and nitrogen assimilation efficiencies. Thus, the standard metrics to estimate potential yields are not sensitive predictors of growth rate changes under variable field conditions and climates.

Today, focusing on applying nitrogen fertilizers to increase photosynthetic efficiencies is seldom attainable by itself since ecophysiological and site factors limit a crop’s NUE. Simkin et al. [12] described the challenge of feeding the world by needing to increase yields by 40% through improvements in photosynthetic efficiency since “… as much as 50% of fixed carbon is lost to photorespiration…” Gutschick [19] suggested that respiration should become a focus of increasing yields since two-thirds of the original photosynthate goes into maintenance and operational costs during a grain crop’s entire growing season. Managing respiration to increase growth rates by reducing photorespiration; however, ignores its other essential role in plant physiological functions. For example, plants must continue to produce antioxidants to protect against reactive oxygen species when excitation energy cannot dissipate during drought as stomata are closed to conserve water [20]. This supports that there are limits to how much science can manipulate a crop’s genotypic potential without resulting in unintended consequences on yields. Further, these studies support what Boyer [21] wrote in 1982 that a crop only reaches 30% of its genetic potential.

Since the genetic potential of a crop results in a plant specialized to grow under specific site growth conditions, plants are less able to adapt to growth conditions that change due to unpredictable temperature and precipitation regimes. Farmers need to select crop plants capable of growing in specific soil and micro-climate niches but also to possess the traits to adapt to short-term changes in environmental conditions, e.g., phenotypically adapting by allocating energy to plant parts such as roots to acquire a growth-limiting resource such as water during short-term droughts. Plants that are specialists in adapting to site-level conditions must also be generalists adapting to unpredictable temperature and precipitation regimes.

Rizzo et al. [22] used an extensive database from 2005 and 2018 to show a plant’s genetic potential explained the smallest fraction (13%) of crop yields and management practices explained 29% of the yields in Nebraska, United States (13% of the increased yield was associated with addressing a crop’s genetic potential, 48% were correlated to decadal-level climate change, while agronomic improvements in management explained 29% of the yield increases). In industrial farming, a crop’s genetic potential allows it to efficiently grow in a specific site-level condition, while a crop’s phenotypic potential is met by farm managers, not the plant, who become the adaptive agents mitigating a plant’s inability to adapt to a changing growth environment by fertilizing, irrigating or applying pesticides [22–28]. The Rizzo et al. [22] study suggested that farmers can manage about 42% of the factors explaining yield levels, while climate change accounted for almost half of the yields reached by crops and has been the most problematic to manage.

Since multiple variables may explain significant changes in yields, it is challenging to identify which variables or combinations of variables describe changes in crop yields without introducing other unintended limitations to growth or environmental pollution [29–30]. Others recommended studying NUE more holistically, including the process of the crop acquisition of soil nitrogen. For example, Govindasamy et al. [26] suggested the initial increases in NUE were due to indigenous soil fertility levels increasing N uptake by crops. Improving NUE continues to be a focus to increase crop yields due to high N fertilization applications, which frequently results in water eutrophication and soil pollution [26,28,31–32]. Congreves et al. [33] reviewed NUE definitions and indices and what is currently ignored in the traditional index, such as “accounting for a wider range of soil N forms, considering how plants mediate their response to the soil N status, including the below-ground/root N pools, capturing the synchrony between available N and plant N demand, blending agronomic performance with ecosystem functioning, and affirming the biological meaning of NUE.” Therefore, focusing on only one or two variables to help increase yields across large heterogenous landscapes will probably not be efficient since other edaphic and environmental factors will become important controlling factors of crop growth within a large landscape, affecting the yields disproportionately.

It would be essential to understand and include a plant’s C allocation to growth and maintenance functions to assess the efficacy of an assessment protocol to measure changes in a plant’s NUE. For example, Asibi et al. [32] wrote how the overuse of N fertilizers resulted in low NUE when there was no simultaneous accounting for increased water-use efficiency. Cassman et al. [24] wrote, “Trends in NUE and the cultivated area will ultimately determine global N fertilizer requirements and the risk of N losses to the environment.” Currently, the dominant factors that increase potential crop yields are selecting cultivars through genetic improvements and improving NUE and water-use efficiencies while decreasing the negative impacts of high fertilizer application rates [24,34]. And importantly, over the last five decades, forest research has provided insights into how plants phenotypically adapt to N fertilizer additions in natural environments [35], which could help frame crop research.

It follows then that modifying NUE by adding N fertilizer to the soil to increase crop yields should involve holistically monitoring soil health for any degradation, subsequently leading to decreases in crop yields. For example, Fan et al. [14] and Jiao et al. [16] wrote how fertilizer applications in China increased from 1 t ha^-1 in 1961 to 6 t ha^-1 in 2015, resulting in increased grain productivity and grain yields in about half of the countries’ arable land area. But, to reach these yield increases, China consumed 35% of the global fertilizer and increased arable land irrigation by 32% while becoming the second-largest producer and consumer of pesticides, accounting for 14% of global use [14]. Despite fertilizer applications, irrigating crops, and controlling pests and pathogens, the yields of cereal crops decreased in China from 4% in the 1970s to 1.9% in the 1990s [14]. They wrote that the amount of nitrogen recovered in the aboveground crop biomass was 35% in the 1990s but declined to 28.3% for rice, 28.2% for wheat, and 26.1% for maize, all these values are lower than world averages of 40–60%. China represents an example of the inevitable holistic trade-offs that need to be anticipated. That is, yield optimization means that there will probably be interactive effects on other, potentially unknown, agricultural economics and ecological processes that would affect future yields.

The introduction of synthetic N fertilizer applications by the Green Revolution were essential to achieve high crop yields; however, understanding how N fertilizer applications impact crop growth holistically is incomplete because crop management and efficiency are based mostly on monitoring harvested product yields and not total plant productivity or edaphic conditions. For example, plant growth rates can significantly increase when soil N levels are low [36] because plants phenotypically adapt by growing more roots.

Further, N-fertilized crops need a higher application rate of pesticides to protect the yield [37] since C allocation to defensive plant chemicals decreases. Fürstenberg-Hägg et al. [38] reported that plant defenses and insect herbivory pressure have metabolic costs. The plant must produce physiologically expensive defensive chemicals using photosynthetically fixed C, reducing its growth and development. (see Martinez et al. [37] for a review of all the links between N fertilizer and pesticide applications and their unintended impacts on ecosystems, wildlife, and people). The C allocation trade-offs that a plant experiences are highlighted by addressing a plant’s response to herbivory. These are not reflected in the total photosynthate produced but are part of the within-plant C allocation patterns directly impacted by N fertilizer applications. When a plant needs to defend itself, it does not grow more roots to adapt to a drought or acquire more nutrients. These trade-offs, therefore, support understanding and managing the source-sink relationships of a total crop and not just the harvested product, which better reflects the relationship of a plant’s phenotypic plasticity to its soil and environmental conditions.

Science-practice gaps in assessing crop yields: Source-sink relationships

The harvest index in seed-producing crops is a C-centric approach that dictates that total shoot dry matter determines aboveground “sources” of photoassimilate, and harvested grain represents the “sinks” [27]. The harvest index also has a C-centric view of yield despite the variation in yields arising from the diversity of soil and climatic environments in which the crop grows. As the harvest index varies with differences in crop management [39], selecting a harvest index likely guarantees a high yield potential only under the environment for which it is selected to plant. The success of this approach at the local level will require managerial diligence and high effort while also being capable of flexibility in the face of climate change. Ultimately, crops are planted in diverse soil and environmental conditions, suggesting a crop cultivar may not perform well in many areas where it is grown. The interaction between harvest index and ecological variation in the growth environment is complex and may not scale according to total yield. However, since harvest-index increases are limited by source and sink strengths, these relationships may provide a valuable tool to identify which cultivars would grow best under different soil and climatic conditions.

The harvest index has a theoretical maximum, and there is a level at which a plant needs to grow more shoot biomass to achieve higher yields [40]. To optimize crop yields, each plant must produce leaves and roots to capture light and assimilate water and nutrients to form the stem to support the leaf canopy, especially flowers and grain. For example, leaf photosynthesis strongly correlates with increased foliar and total plant biomass [41–42] but less with how a plant allocates carbon since crops are generally selected for higher yields and lower root biomass [43]. A study conducted in the Midwest United States reported that maize had 8.3% of its total biomass in roots and 52.1% in grain yield at maturity, while soybeans had 16.6% in roots and 41.0% in grain yield [43]. The high percentage of grain yields does not represent a crop with an evolutionary balanced source-sink relationship where crops adapt to their growth environment. It also demonstrates how selecting traits to maximize crop yields reduces the crop’s ability to adapt to its changing growth environment phenotypically. Climate change may disrupt other normal growth conditions besides temperature and precipitation. It includes novel stressors such as high winds, which would drive carbon allocation to stems or root biomass or increase secondary metabolites in response to pests or other complex combinations of unique factors [44].

A greater understanding of the source-and-sink relationships of the plant facilitates understanding how a plant shifts its allocation to source and sink functions following fertilizer application and whether a plant can adapt to climate change at the site level. This is not just the total amount of C that is fixed and allocated to the harvested product but whether sufficient C is allocated to nutrient and water uptake or to defensive compounds if pests or pathogens attack [35,45–46]. Today, meeting the challenge to improve crop productivity requires increasing our understanding of total crop growth (above- and belowground) on different soil types and the impact of environmental stressors such as climate change on crop yields. There will be unique combinations of conditions that exhibit dynamic interactions at the site level that will drive the ecological response (Fig 1). This is especially important to include in a framework assessing how to improve crop yields where, for example, irrigation may increase yield, but N fertilization may decrease yield. Decreased yields resulting from N fertilization may represent changes of the within C allocation fluxes in plants, e.g., between growth and maintenance to above- and below-ground plant parts (e.g., fruit, seeds, leaves, roots, and defensive chemicals). For example, Baslam et al. [29] explained how increasing N levels might reduce crop yields as plant shoot growth increases with less C allocated to roots.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Schematic depiction of photosynthetic pathways and products.

[1] Sources; the dominant factors that influence nutrient availability, [2] Controls; genomic mechanisms of plant adaptation to disturbance in its environment, and [3] Sinks; plant carbon sink strength as a function of metabolism. The major functional processes associated with each listed at bottom. The plant images are from unknown artists; (left) maize plant (Gong F, Wu X, Zhang H, Chen Y and Wang W (2015) Frontiers | Making better maize plants for sustainable grain production in a changing climate (frontiersin.org) and (center) phenotypic plasticity of roots (Calleja-Cabrera J, Boter M, Oñate-Sánchez L and Pernas M (2020) Frontiers | Root Growth Adaptation to Climate Change in Crops (frontiersin.org) are both licensed under CC BY. The chemical structure of the secondary metabolite was recreated by the author.

https://doi.org/10.1371/journal.pstr.0000122.g001

Measuring changes in photosynthesis may not detect changes in plant growth due to drought, low temperature, and nutrient limitations [47]. For example, a drought will inhibit cell growth and tissue formation before C uptake is inhibited during photosynthesis. Also, meristem cell production will stop at temperatures ≤ 5°C, while net photosynthesis continues at 50–70% of the rate when plant growth occurs above 5°C [47]. Körner [47] supports understanding how above- and belowground growth changes in relationship to drought, low temperature, and nutrient limitations since these variables impact plant growth much earlier than photosynthetic efficiencies. Holland et al. [48] further suggested using a theoretical analysis approach to explore how root and leaf respiration can explain C allocation strategies by increasing the timing of C assimilation to leaves and roots. Their study supports using models to understand growth and recognize that maintenance costs can also enhance yields.

These studies support the need for more research on the interplay of physiological processes, such as how allocation shifts between the shoots and roots are involved in increased water and NUE, while also decreasing pollution from high fertilizer applications. Most experimental field studies emphasize taking measures of seedlings under controlled conditions and modeling crop yields that do not include realistic estimates of how crops allocate C to all the C pools and fluxes [49]. For example, selecting a crop to become more drought tolerant may reduce its yields as there is insufficient C fixed during photosynthesis to balance all the source-sink relationships that are the adaptive mechanisms in response to climate change.

Ecosystem translation of the belowground world to illuminate potential crop yields

If improving photosynthetic efficiency and NUE successfully address all the growth-limiting factors in agriculture, crop yields will probably not plateau or decrease in many parts of the world [6]. Photosynthetic efficiency is crucial since it determines the size of C pools available for improving crop yields. However, within plant source-sink relationships will determine whether a crop can grow and maintain its tissues and adapt to increase its assimilation of resources that limit its growth. Thus, it will decide if a crop will reach its growth and yield potential [27]. Roots are essential in determining how much water and indigenous soil nutrients and N applied in fertilizers are assimilated by a crop plant. Also, roots are the interface exposed to moderately-to-severely degraded land and determine whether crops can grow on a site; Iseman and Miralles-Wilhelm [50] reported that 52% of the global agricultural soils are moderately to severely damaged.

There is increased recognition of the importance of roots in crop adaptation to their changing environment and that N fertilizer decreases root growth [34,45,51–54]. Previously, the assumption was that knowledge of aboveground growth is a surrogate for the total plant response to its dynamic environments. However, Liu et al. [54] synthesized 88 published studies to show the existence of a phenological mismatch in the timing of above- and below-ground growth in response to climate variability. Therefore, we must implement a holistic view of the plant adaptive response at multiple levels of biological organization, from resource allocation to phenotypic growth patterns, and more fully relate the emergent properties that arise under different environmental and ecological conditions to the individual agricultural system components.

Further, the increased frequency of droughts is impacting yields. It highlights the role of roots in a crop adapting to short-term changes in available water supplies and nutrient acquisition. When less photosynthate allocation to roots occurs with higher N applications [35,55–56], a plant may produce less biomass, i.e., lower productivity levels, since plants with smaller root systems have less capacity to acquire soil nutrients and water during a drought. Maslard et al. [53] described how selecting for a diversity of optimal root systems in new soybean varieties is important to address edaphic and climatic limits to crop growth, e.g., cultivars with deep root systems to take up N and water from lower soil layers and cultivars with shallow root systems that can readily acquire nutrients from surface soils such as when phosphorus availability is limited. The total biomass of nine different soybean genotypes by Maslard et al. [53] had statistically significant differences in total plant, shoot, and root biomass, showing how selecting a crop by its root biomass is helpful information for managers.

Olagunju et al. [52] reported how tropical upland rice adapts to dynamic climatic conditions such as unpredictable rainfall and the resulting droughts by reducing biomass allocation to shoots but not to roots. They further showed that upland rice adapted to these periodic drought events depending upon how soil texture impacted root and shoot growth, i.e., less root biomass is produced as the amount of clay increases in the soil. Olagunju et al. [52] wrote that focusing on the plant’s reproductive parts would result in “more reliable estimates for identifying rice cultivars with higher yield potential at harvest.” They also wrote, “Soil environment that promotes greater allocation of biomass to reproductive structure through a restriction in the expansion of vegetative organs is well suited for upland rice cultivation.” A focus on selecting cultivars for their reproductive yields makes the crop susceptible to the unpredictable rainfall periods that a balance of root biomass and crop yields would not produce.

In addition to observed responses, climate variability may also induce unpredictable responses that will vary by the local milieu of growth conditions. For example, higher soil temperatures are associated with less allocation to root biomass as feedback to changes in soil condition, including loss of moisture, aeration, and nutrients [56]. Calleja-Cabrera et al. [45] wrote about the need to develop an efficient root system that can make a plant better adapted to its site to increase crop productivity. They wrote how increasing temperatures with climate change increases the stresses that a crop will experience and will have to adapt to, e.g., “drought, salinity, nutrient deficiencies, and pathogen infections” [45]. In that case, a crop plant that can adapt to its growth environment should be selected, which means data needs to be collected on total plant biomass and carbon allocation to roots and defensive compounds [35,46]. This selection process must account for different types of variability, the magnitude of change among the most important growth parameters, and how well the cultivar is expected to respond across a diversity of conditions. These carbon allocation shifts in response to site edaphic and micro-climatic factors ultimately determine how well a plant grows at a local site and how much crop yields can increase when growing under dynamic environmental conditions [35,57].

Cakmak et al. [58] wrote how a balanced fertilizer application is needed to maintain growth since mineral deficiencies of phosphorus, potassium, and magnesium impact C partitioning differently between roots and shoots in bean plants. Their study showed the roles of magnesium and potassium in allocating C from shoots to the roots. Woo et al. [34] showed how managing wheat yields to achieve a root radius of 0.1 and 0.3 mm resulted in optimal wheat yields. Since many other factors impact roots, managing fertilizer application rates may not support root growth architecture in the field [34].

Instead of focusing on the product’s increased yield, a farmer can select a crop based on variable rooting depths to address site scale limitations to growth during dynamic climatic conditions. The importance of selecting genotypes based on their root architecture and drought tolerance, as well as the role of roots in increasing soil organic matter levels, is stimulated by the recognition that “allocation pattern indicates environmental plasticity to soil properties, temperature and soil water availability” [51]. Mathew et al. [51] reported how drought stresses reduced total biomass production by 35% and root-to-shoot ratios by 14% and how soil C is mainly derived from root activity and decomposition of root tissues. These are subtle and specific observations of the plant response, and as we learn which responses to measure, we will fine-tune the site-level accommodations required and the selection of varieties that can meet the specific demands of the site.

Selection of plants based on their root architecture still needs to factor in the continued use of fertilizers in farming. This means that when a plant is bred for its increased allocation of photosynthate to the harvested product, it may be less able to maintain and protect plant tissues or acquire other limiting resources needed to grow [59]. This occurs when a crop is selected to optimize one part of the plant, e.g., genotypes high in seed oil and protein content for human consumption, animal feed, transport fuels, and many other products. Under these circumstances, the plant allocates less to defensive chemicals. The farmer must spray herbicides and pesticides to reduce the growth of other competitive plants and to protect the plant due to the tradeoffs that were considered acceptable to management when those protective traits were discounted for the artificially selected desired traits.

A holistic framework: Thresholds of total productivity and yields due to growth-limiting factors

However, one still needs a framework to determine how much site growth limiting factors reduce a crop’s total productivity to determine the potential growth and yields possible per site. This concept supports using a metric—e.g., a plant’s total productive capacity—to estimate whether a crop is close to a threshold of decreasing productivity with additional growth-limiting factors. This index assesses a plant’s growth in its edaphic-climatic environment and allows cross comparisons of different sites [55]. A module, like Ecosystem Fit (eFit), must be created to assess whether a plant can continue adapting its potential productivity in response to site growth-limiting conditions. Since eFit uses parameters of solar radiation, temperature changes, edaphic and climatic factors, it may be useful to reflect or index growth rates or site productivity and compare it to other sites with different conditions. In fact, indices like this could be used to tease apart specific site factors that affect productivity.

A farm needs to be viewed at an ecosystem-level and based on total productivity measures, especially since crop yields may be a third of the total C fixed during photosynthesis [22]. A plant’s total net primary productivity (tNpp) represents how much C can be allocated to acquire growth-limiting resources, such as nutrients and water, in dynamic soil and climatic conditions. In addition, the crop adapts to grow and store carbohydrates to protect itself during disturbances. A crop plant should not be decoupled from its growth environment since management will not be able to address all the limitations to growth that a crop plant will experience. This suggests monitoring photosynthetic efficiencies, and NUE needs to be replaced with a holistic approach that assesses the entire ecology of the crop. We currently don’t have the frameworks to assess how management can counter yield decreases locally, especially since crop growth is generally viewed through a narrow lens that looks at only part of the aboveground portion of the plant. This narrow focus does not include site-level soil and climatic factors which are crucial for understanding the totality of factors that explain decreases holistically at the local-level. For example, it does not factor in the potential negative effect of high levels of N fertilizer which can decrease the plant’s ability to acquire nutrients from unhealthy soil.

The new framework to measure the phenotypic plasticity of a plant first emerged from Gordon et al. [60], who used the estimated Theoretical Maximum Productive Capacity (TNpp_max) using the Loomis and Williams [61] method. Gordon et al. [60] developed a conceptual approach to calculate the eFit or the potential total productivity for a site based on the ratio of field-collected data of tNpp to the TNpp_max calculated from external factors. Gordon et al. [60] described eFit, and Klock et al. [35,62] provided the methodology to calculate theoretical maximum productivity and eFit. In global forests, eFit and total productivity were invariant to leaf behavior traits but were “strongly dependent on temperature, precipitation, elevation, Oxisol, Entisol, and Ultisol soils, and silty loam soil texture” [35].

Gordon et al. [63] demonstrated the utility of using tNpp to calculate eFit on two crop plants grown in Japan. They calculated an eFit of 19% for maize in Iwate, Japan (actual tNpp of 18 Mg ha^-1 yr^-1) and 14% for rice (not displayed) in Akita, Japan (actual tNpp of 15 Mg ha^-1 yr^-1)(Fig 2). In the eFit calculations made by Gordon et al. [63] and the yield data of ~5 Mg ha^-1 yr^-1 dry weight provided in Ritchie et al. [6], about 25% of the tNpp is allocated to the harvested product, and the remaining to plant maintenance and growth. Ecosystem fit showed that both maize and rice reached less than 19% of their theoretical maximum productivity threshold (TNpp_max). Gordon et al. [60] also showed how eFit is sensitive in describing how edaphic and climatic site conditions limit crop growth and yields, which are factors that can be managed or mitigated.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Diagrammatic representation of an example calculation of actual productivity and theoretical maximum productivity.

Maize in Japan using data published in Gordon et al. [60]. The potential productivity shows how much tNpp needs to increase to achieve a higher fit when ecological management alleviates the growth-limiting factors constrained by soil health and climatic factors. [Calculate eFit = tNpp observed/TNpp_max * 100] [Globally, dry weight yields of maize are about 5 Mg ha^-1 yr^-1 [6] which is ~ 25% of total actual net primary productivity (tNpp).]

https://doi.org/10.1371/journal.pstr.0000122.g002

Definitions of yield and productivity

There is confusing terminology in the literature on what is included in the terms “yield” and “productivity” of a crop. Yields are generally related to agricultural production but are not synonymous with productivity that ecologists would use. For example, yields are the usable measured biomass or weight or also volume of the crop harvested, such as the grain, cereal, and fruit produced per unit of land [64]. Further, if yields are given in weight terms, the weights are moist weight values and are reported as a standardized moisture content, which is important for proper grain storage and comparison across other research trials [65]. It does not include the weight of any other part of the plant. The actual yield on a farm varies depending on the amount of sunlight or radiation reaching the plant, the plant’s water and nutrient uptake efficiencies, the crop’s genetic potential, and how much pests and pathogens decrease the crop product harvested.

Yield data does not provide as much information as the tNpp of the whole plant based on dry weight measures of biomass. For example, the productive capacity of a plant can indicate whether it has the potential to adapt to its dynamic environment and grow biomass at the site level. Additionally, since fertilizer applications generally decrease [55] the C allocated to the root system, it would follow that just being attentive to increasing yields would not allow managers to recognize that potential decreases in root biomass could significantly increase the risk of the crop’s ability to respond to lower soil moisture during droughty conditions and decreasing its yields. This would be even more important to consider for perennial crops. The question is whether long-term yields can be improved by selecting crop varieties based on their tNpp while balancing its allocation of C to the harvested product and allowing a plant to adapt to site-level edaphic and micro-climatic constraints and continue to grow.

In this paper, we will consider agricultural yields in terms of product biomass, but they are also known in terms of economic returns and produced on a per-unit-of-land basis [64]. It differs from total crop productivity (e.g., tNpp) measured by ecosystem ecologists as Mg ha^-1 yr^-1, which represents a plant’s annual total biomass productivity. The ecological definition includes the plant’s above- and belowground parts (e.g., leaves, branches, stems, bole, coarse roots, fine roots, mycorrhizas), and biomass loss due to herbivory, autotrophic respiration, and root turnover and litterfall (e.g., carbon fluxes) added back to biomass carbon pools (e.g., [66–67]).

Knowledge of a crop’s actual tNpp and TNpp_max will provide a framework to correlate yields to changing environmental and edaphic conditions represented across a heterogeneous landscape. Actual tNpp, in dry-weight biomass produced during a year, provides an index of growth that can be compared across a diversity of sites. In this paper, we present different productivity terms that are used in our framework. Actual Productivity of a crop is the total Net Primary Productivity (tNpp observed or measured) reached at a site as constrained by soil health, pests and pathogens, drought, temperature, and salinity, to name a few variables. The Potential Productivity (tNpp potential) is the amount that tNpp can be increased using management, i.e., environmental and/or ecological tools, such as those that can enhance or create healthier soils or use more adaptable plants for site-specific conditions. This is determined from field-based research that monitors tNpp changes under different management conditions. Calculating the theoretical maximum productivity (TNpp_max) provides the maximum productivity attainable at a given site, but it is not environmentally or ecologically possible to manage since it would involve managing limiting-growth variables that cannot be economically manipulated. Examples of the variables used to calculate TNpp_max for a given site include solar radiation, temperature, and the growing season length, which are factors we cannot manage.

Characterization and definitions of the dominant soils in India (UN FAO)

1. Cambisols—are young soils, medium and fine-textured, shallow topsoil depth; moderate fertility but commonly deficient in phosphorous (P) and calcium (Ca); high erosion rates, generally good water-holding capacity, good internal drainage, but the dried soil surface becomes extremely hard when dry, hindering root growth and favoring erosion.
2. Fluvisols—are young soils forming in alluvial deposits with little or no profile development, mineral soils conditioned by topography; in coastal areas, they have high levels of salts and aluminum (Al) ions—therefore, low soil pH (i.e., high soil acidity) and high Al toxicity both help create phosphorus deficiency and also N deficiency is common; found in floodplains so they periodically flood and therefore need flood control, drainage or at times even irrigation.
3. Luvisols—are moderately weathered soils, and if they have clay-enriched subsoils, they have high cation exchange capacities and high base saturation; steep slopes are not uncommon and need erosion control; high nutrient content, therefore are fertile and widely used for agriculture; well-drained but soil may become saturated with water for extended periods potentially needing drainage.
4. Nitosols—are strongly weathered soils but are more productive than most red tropical soils and are deep soils with favorable physical properties with deep rooting so they are resistant to erosion; contain low-activity clays (i.e., have a lower capacity to retain and supply nutrients), high P fixation enhanced by iron/aluminum (Fe/Al) chemistry; generally fertile soils despite low available P and low base status but need to add P fertilizer; plant available nutrients are fairly deep (~150 cm), they are well-drained but total moisture storage is good because they are deep; however they are hard when the soils are dry. They are exploited widely for plantation agriculture.
5. Vertisols—have high content of shrink and swell clays that are strongly impacted by wet and dry conditions (i.e., harden when dry and become sticky when wet), little textural differences by depth; good for mechanized farming if the rainfall is high or they are irrigated; many areas are not farmed because they would need to be irrigated; low soil permeability, so irrigation may cause waterlogging. They are best suited for pastureland use and cultivating plants, such as rice, that thrive in standing surface water.
6. Xerosols—are desert soils that have mostly sandy soil and are in low rainfall areas; so, they have low N and organic matter (OM) but high concentrations of calcium carbonate and soluble salts and phosphate, therefore they are frequently infertile requiring substantial management. Generally, they have soil moisture deficits and are susceptible to wind erosion, so they are unsuitable for rain-fed agriculture. But if irrigated, these soils may be among the best soils for farming. Generally, these soils are of little or no value for agriculture due to the lack of rainfall.

Source: https://www.britannica.com/science/soil/FAO-soil-groups, https://www.isric.org.

Comparison of the dominant soil groups, mean annual precipitation, mean grow temperature, and Theoretical Maximum Productive Capacity (TNpp_max)

The seasonal grow temperature was strongly associated with latitude (r = –0.67, t = –16.874, df = 354, p < 0.001) whereas TNpp_max was independent of latitude (r < 0.001, t = 0.002, df = 354, p = 0.99), even though the latitudinal range of this study was ~25° (8.31° N– 32.80° N), nor was TNpp_max associated with longitude (r < -0.075, t = -1.777, df = 354, p = 0.16) where the range was ~19° (69.47° E– 88.59° E).

Unsupervised cluster analyses identified a single threshold for grow temperature of 26° C and two relatively homogenous clusters of TNpp_max, one on either side of 289.1 Mg ha^-1 yr^-1, thus data summaries were made for two TNpp_max groups, classified as low (≤ 289.1 Mg ha^-1 yr^-1) and high (> 289.1 Mg ha^-1 yr^-1). We then utilized a series of parametric and non-parametric approaches to understand the association between TNpp_max, tNpp, and eFit and dominant soil groups, and climatic variables.

The statistical power of the low and high TNpp_max groups to soil type was low for a meaningful comparison (power < 0.24), therefore we aggregated TNpp_max. An omnibus Welch’s heteroscedastic F Test for the data (n = 356) with post hoc Bonferroni correction for multiple pairwise comparisons found differences of at least one soil group with an effect size (ω² = 0.35 CI_95% [0.18, 1.00]) considered moderate per Cohen’s 1988 convention [68] (Fig 3). As variability of TNpp_max among the dominant UN FAO soil types was not correlated with geography, we fitted a generalized linear model (estimated using ML) to predict TNpp_max with dominant soil type to identify which soil types were the drivers of the association (Table 1). The model’s total explanatory power was low (R² = 0.14) with an intercept corresponding to Cambisols of 291.91 (95% CI [290.2, 293.7], t(350) = 324.09, p < .001). Standardized parameters were obtained by fitting the model on a standardized version of the dataset. 95% Confidence Intervals (CIs) and p-values were computed using a Wald t-distribution approximation.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Distribution of TNpp_max to the dominant UN FAO soil groups (n > 8).

The large dot is the mean and reported adjacent to the plot. The thick horizontal line is the median. The box in the center represents the interquartile range, and the vertical thin gray lines above and below the box represent the rest of the distribution. The grey dashed line represents the grand mean. The totality of the curved shape represents a kernel density estimation where wider sections represent a higher probability of a given value and thinner sections represent a lower probability. Results of an omnibus one-way ANOVA, partial effect size, and number of observations reported at the top.

https://doi.org/10.1371/journal.pstr.0000122.g003

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Summary of general linear model of TNpp_max to dominant soil types.

https://doi.org/10.1371/journal.pstr.0000122.t001

Another approach was to determine whether TNpp_max varies among the dominant soil types by grow temperature. The estimation of TNpp_max is based on incoming solar radiation and the mean growing temperatures for the total month-days with temperatures above 0°C. Therefore, the maximum theoretical potential for growth, i.e., TNpp_max may be affected by environmental factors such as elevation or azimuth of a given site, indicating some amount of top-down control by topography.

There was a significant negative relationship between TNpp_max and the grow temperature suggesting reduced crop productivity as temperatures increase among Luvisols, Nitosols, and Vertisols (Fig 4). The soils that showed no relationship with temperature are also less suitable for agriculture due to nutrient deficiencies (Cambisols, Fluvisols) and low precipitation (Xerosols). The variability of TNpp_max was highest among Luvisols (var = 93.9), Fluvisols (var = 121.4) and Nitosols (var = 132.1), while it was lowest for Cambisols (var = 29.4), Xerosols (var = 35.7) and Vertisols (var = 51.1) (Fig 4).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. Relationship of theoretical maximum productive capacity (TNpp_max), temperature and dominant soil types (UN FAO).

Association between TNpp_max to the average grow season temperature among each dominant soil type. A temperature threshold of 26° C is indicated by the dashed red line, and the Pearson’s Product Moment correlation (r) with p-value is reported at the top for each soil type.

https://doi.org/10.1371/journal.pstr.0000122.g004

Comparison of TNpp_max to mean annual precipitation indicated how the dominant crop growing areas varied between precipitation levels (i.e., dry to wet precipitation groups). Half of the dominant soil types indicate non-significant relationships, therefore maintaining a focus on selection for crops with higher photosynthetic efficiency is not going to provide additional productivity benefits (Fig 5). Most notably, TNpp_max was primarily concentrated in a range of ~280 to 300 Mg ha^-1 yr^-1, but above a threshold of ~1,200 mm of annual precipitation, there was increased variability of TNpp_max among Fluvisols and Nitosol dominated sites probably due to specific climatic niches. In contrast, crops growing in Cambisols, Luvisols, Vertisols, and Xerosols are located in areas that have high TNpp_max, >300 Mg ha^-1 yr^-1. Xerosols and Cambisols were the only soil groups identified in the dry precipitation group (Fig 5).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 5. Relationship of theoretical maximum productive capacity (TNpp_max) to precipitation by dominant soil types (UN FAO).

Four precipitation classes; dry, moist, very moist, and wet mean annual precipitation are from Vogt et al. [69]. The optimal range of precipitation for agricultural productivity is indicated by the grey panel (500–1200 mm^-1 yr^-1). Pearson’s Product Moment correlation (r) with p-value is reported at the top for each soil type.

https://doi.org/10.1371/journal.pstr.0000122.g005

This comparison between TNpp_max and annual precipitation further supports the need to characterize the site-level edaphic and micro-climatic conditions since total productivity is related to the available water supplies and the nutrient delivery status of the soil. The highest variability in TNpp_max was found in the very moist group, ~1,200 mm yr^-1 precipitation, where TNpp_max showed a bifurcated response. For example, Ferric Luvisols had a linear decline and linear increase, whereas crops growing on Chromic Luvisols would be limited in available energy, limiting their adaptive capacity. In contrast, areas of agriculture in the moist and lower threshold of the very moist precipitation groups show more consistent TNpp_max values [Note: Comparisons between Ferric Luvisols and Chromic Luvisols are not included in this paper]. This suggests that when agricultural areas receive greater than ~1,200 mm of rainfall annually, these represent extreme sites for agriculture with a higher likelihood of uncertainty in yields.

Comparison of the dominant soil groups, mean annual precipitation, mean grow temperature, and Total Net Primary Productivity (tNpp)

An unsupervised cluster analysis identified three natural clusters of tNpp classified as (low < 5.12, medium > 5.12 & < 9.51, and high > 9.51 Mg ha^-1 yr^-1). An omnibus Welch’s heteroscedastic F Test for the tNpp clusters found significant differences among the six dominant UN FAO soil types in the low and medium tNpp groups, whereas post hoc Bonferroni correction for multiple pairwise comparisons resulted in no evidence of significant differences in the high tNpp group (Table 2).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Summary of Total Net Primary Productivity (tNpp), ranked by tNpp clusters (low, medium, high), and Ecosystem fit (eFit) by dominant soil type (UN FAO).

Soil groups with (n ≥ 8). Letters indicate column-wise comparisons among the soil groups (where the mean of tNpp does not differ among soil groups displaying the same letter). Aggregated crop lands (bottom row).

https://doi.org/10.1371/journal.pstr.0000122.t002

A summary comparison between tNpp and the soil types identified Luvisols and Nitosol soil types as having the highest range of tNpp values (there was no low range of tNpp observations among Nitosol soil types) (Table 2).

The lowest tNpp were recorded in Xerosols (tNpp range 0.0–6.4 Mg ha^-1 yr^-1), Vertisols (1.0–13.0 Mg ha^-1 yr^-1) and Cambisols (0.2–9.8 Mg ha^-1 yr^-1). The highest tNpp ranges also had higher eFit values and ranges: Fluvisols (0.9–4.7%), Luvisols (0.5–6.6%), and Nitosols (2.1–8.9%). The lowest eFit means were recorded in Xerosols, Vertisols, and Cambisols soil types (UN FAO classification). The variability of eFit was highest with Luvisols and Nitosols, indicating a wide range of productive capacity where there may be opportunities for increasing yields (Table 2).

We also explored whether productivity varied significantly among the dominant soil types by fitting a linear model (estimated using OLS) to predict tNpp with soil group. The model explains a statistically significant and substantial portion of the variance (R² = 0.38, F(5, 350) = 43.03, p < .001, adj. R² = 0.37). The model’s intercept, corresponding to Cambisols, is at 4.31 (95% CI [–3.64, 4.98], t(350) = 12.68, p < 0.001, AIC = 1844). To identify the influence of climatic variables to the response we then fitted a linear mixed model (estimated using REML and nloptwrap optimizer) with grow temperature and precipitation as fixed effects and dominant soil group as a random effect. The model’s total explanatory power was substantial R² = 0.47 with the fixed effects accounting for R² = 0.14. The model’s intercept corresponding to precipitation and grow temperature = 0 is at –12.06 (95% CI [–18.55, –5.58], t(351) = –3.66, p < .001). Within this model: the effect of temperature is statistically significant and positive (beta = 0.64, 95% CI [0.40, 0.88], t(351) = 5.21, p < .001, Std. beta = 0.21), the effect of precipitation is statistically significant and positive (beta = 2.06⁻⁰³, 95% CI [1.38⁻⁰³, 2.74⁻⁰³], t(346) = 5.97, p < .001, Std. beta = 0.28). The model was slightly improved (AIC = 1818).

A statistical analysis of the high and low TNpp_max groups to the six dominant UN FAO soil types by one-way ANOVA found significant differences between at least two dominant soil groups (Fig 6). The Welch’s F-test assumes that data groups are sampled from populations that follow a normal distribution but does not assume that those two populations have the same variance. Pairwise comparisons with correction for multiple comparisons indicated significant differences between varying combinations of all soil types. There were no significant differences in actual tNpp between Cambisols and Fluvisols, nor between Cambisols and Xerosols in either TNpp_max group.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 6. Summary analysis of total net primary productivity (tNpp).

Comparison between (a) high TNpp_max group (> 289.1 Mg ha^-1 yr^-1) and (b) low TNpp_max group (≤ 289.1 Mg ha^-1 yr^-1) to the dominant UN FAO soil groups (n > 8). Test results reported at top are for the omnibus heteroscedastic F Test with Bonferroni correction. Statistically significant pairwise comparisons are delineated by brackets with the p value reported above the line (* ≤.05, *** < .001). The large dot is the mean and is reported adjacent to the plot. The thick horizontal line is the median. The box in the center represents the interquartile range and the vertical thin gray lines above and below the box represent the rest of the distribution. The grey dashed horizontal line represents the grand mean. The totality of the curved shape represents a kernel density estimation where wider sections represent a higher probability of a given value, and thinner sections represent a lower probability.

https://doi.org/10.1371/journal.pstr.0000122.g006

The effect size is calculated by the partial omega-squared [ in the figures] and represents an estimate of how much variance in the response variable is accounted for by the explanatory variable. Standardized effect sizes such as omega-squared do not use the original data units but are unitless, allowing comparisons of results between variables that use different units. The effect size is like R², ranging from 0 to 100%, representing the percentage of the variance that the variables in the model collectively explain. Each categorical variable has a value indicating the percentage of the variance it explains. Like R², omega-squared is an intuitive measure that you can use to compare variable effect sizes between models, but it also adjusts for bias, particularly for small samples, and thus is an unbiased estimator.

The analysis identified statistical differences among varied combinations of soil types, where the effect size of the high TNpp_max group was (g = 0.82) (Fig 6a) and the effect size of the low TNpp_max group was (g = 0.86) (Fig 6b). These values are considered high per convention, indicating they are more probable under the alternative hypothesis.

We can infer that the overall relationship between soil types and tNpp are more strongly supported by the evidence than TNpp_max. In this instance (Fig 6), the effect size indicates that the low TNpp_max group (panel b) is more probable under the alternative hypothesis.

Similar to the TNpp_max comparison to grow temperature (Fig 4), a temperature threshold was found above 26°C when comparing tNpp to grow temperature (Fig 7). In contrast to TNpp_max, tNpp did not support a decrease in productivity as temperatures increased. In fact, tNpp was maintained and increased at temperatures higher than 27.5°C. In this comparison, crops growing on Vertisol soil type maintained a similar range of tNpp (5 to 10 Mg ha^-1 yr^-1) in the temperature range between 25.0 to 27.5°C. The highest tNpp levels were found in the Nitosol soil type, where tNpp exceeded 20 Mg ha^-1 yr^-1 (Fig 7). Luvisols had the highest variation in tNpp compared to the other five dominant soil types (UN FAO).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 7. Relationship of total net primary productivity (tNpp) to temperature by dominant soil types (UN FAO).

A temperature threshold of 26° C is indicated by the dashed red line, and the Pearson’s Product Moment correlation (r) with p-value is reported at the top for each soil type.

https://doi.org/10.1371/journal.pstr.0000122.g007

In contrast to the relationship between TNpp_max and mean annual precipitation (Fig 5), there was no upper threshold of tNpp at 1,200 mm of mean annual precipitation. The precipitation threshold reached 1,600 mm (Fig 8). Total net primary productivity varied over a range from 1 to 22 Mg ha^-1 yr^-1 in the Moist and Very Moist precipitation classes. This suggested that tNpp levels were not limited by precipitation levels. The highest tNpp levels were found in the Nitosols (6.0–25.4 Mg ha^-1 yr^-1) and Luvisols (1.4–21.1 Mg ha^-1 yr^-1) soil types (UN FAO), while the lowest range of tNpp values (1.0–13.0 Mg ha^-1 yr^-1) were produced in the Vertisols, Fluvisols (2.6–13.5 Mg ha^-1 yr^-1) and Cambisols (0.2–9.8 Mg ha^-1 yr^-1) soil types (UN FAO). Nitosols show optimal productivity with increasing temperature and precipitation conditions, whereas crops on Cambisols do not show higher productivity across the entire range of precipitation and temperature. Therefore, under limiting growth conditions Nitosol soils are still productive.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 8. Relationship of total net primary productivity (tNpp) to precipitation by dominant soil types (UN FAO).

Four precipitation classes; dry, moist, very moist, and wet mean annual precipitation thresholds are from Vogt et al. [69]. The optimal range of precipitation for agricultural productivity is indicated by the grey panel (500–1200 mm^-1 yr^-1). Pearson’s Product Moment correlation (r) with p-value is reported at the top for each soil type.

https://doi.org/10.1371/journal.pstr.0000122.g008

The National Bureau of Soil Survey and the Land Use Planning under the control of the Indian Council of Agricultural Research (ICAR) have conducted extensive studies on Indian soils [70]. The ICAR classifies soils based on their characteristics as per the Soil Taxonomy of the United States Department of Agriculture (USDA). Chief characteristics are based on genesis, color, composition, and location. The three categories relevant to our reference sites were as follows:

(i) Alluvial soils, comprising ~43% of Indian soils dominate the northern plains and river valleys, and found extensively in deltas and estuaries. Alluvial soils are highly fertile, (ii) Red / Yellow soils are found largely in drier areas and are generally nutrient deficient with sandy to clayey and loamy texture, and (iii) Black soils are best for the cultivation of cotton and cover most of the Deccan plateau, which is characterized by deserts, xeric shrublands, and dry tropical forests. Although these areas receive less rainfall, Black soils have high water retaining capacity and are rich in minerals but deficient in N, P, and organic matter, with a clayey texture.

An unsupervised cluster analysis identified three natural clusters of tNpp among ICAR soil classifications as: (i) low, 0.0–5.1 Mg ha^-1 yr^-1, (ii) medium, 5.2–9.5, and (iii) high, 9.7–25.4 Mg ha^-1 yr^-1 (Table 3). The analysis found tNpp was significantly lower in the Alluvial soils (mean = 2.93, n = 50) compared to the Red / Yellow soil types (mean = 3.69, n = 26) in the low productivity group (W_Mann-Whitney = 996.00, p < 0.001). Whereas, these two groups were similar in the medium productivity group (X²_{Kruskal-Wallis}(2) = 8.88, p = 0.01), and high productivity group (X²_{Kruskal-Wallis}(2) = 8.47, p = 0.01). These comparisons did not identify the range of tNpp that resulted from using the dominant UN FAO soil types. Table 3 shows the clusters of tNpp (low, medium, high) ranked by the three major soil type groupings.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Summary of Total Net Primary Productivity (tNpp) by cluster level (low, medium, high) to the major soil type used by the Indian Council of Agricultural Research (ICAR) [70].

Omnibus tests by tNpp group indicated significantly higher tNpp in red/yellow soil types in the low tNpp group (W_Mann-Whitney = 996.00, p < 0.001), and differences of tNpp among the soil types of the medium tNpp group (X²_{Kruskal-Wallis}(2) = 8.88, p = 0.01), and the high tNpp group (X²_{Kruskal-Wallis}(2) = 8.47, p = 0.01).

https://doi.org/10.1371/journal.pstr.0000122.t003

These results indicate that a soil type may be more important at lower productivity levels. But, as the three India soil types include several UN FAO soil types, it was less successful in identifying low, medium, and high clusters. Each soil type was found in all clusters except for Black which is not present in the low tNpp cluster. This suggests that volcanic soils produce crops in the medium and high tNpp groupings, and these groupings were less able to identify productive agricultural fields that are using the UN FAO soil type produced (Table 3). When comparing the Nitosols soil type (UN FAO) and the Black soil type, they had very similar medium and high tNpp clusters, but the other soil types did not produce similar tNpp clusters with the UN FAO soil types.

Comparison of the dominant soil groups, mean annual precipitation, mean grow temperature, and Ecosystem Fit (eFit)

All soil types had a large range of tNpp, suggesting no significant differences in eFit by dominant soil types (Fig 3). Significant differences were found in eFit by the dominant soil types (UN FAO) grouped by the low and high TNpp groups (Fig 9). Significantly higher eFits were found in the Nitosols soil type (UN FAO), suggesting that this soil type could potentially support higher tNpp through management. The second highest potential growth rates increases were found in the Luvisols soil type (UN FAO), also suggesting management can increase tNpp. All the dominant soil types in India had low eFits (Fig 9) compared to the soils in Japan where eFit of 19% was reached, that was twice as high as eFits calculated for India (8.9%) (see Fig 2) [60]. The variability in eFit was the highest for Luvisols and Nitosols in India, suggesting these soil types have a greater potential for increased management to succeed in increasing growth rates (Table 2).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 9. Summary analysis of Ecosystem Fit (eFit).

Comparison between (a) high TNpp_max group and (b) low TNpp_max group to the dominant UN FAO soil groups (n > 8). Test results reported at top for omnibus heteroscedastic F Test with Bonferroni correction (α = .05). Statistically significant pairwise comparisons are delineated by brackets with the p value reported above the line (* ≤.05, *** < .001). The large dot is the mean and is reported adjacent to the plot. The thick horizontal line is the median. The box in the center represents the interquartile range and the vertical thin gray lines above and below the box represent the rest of the distribution. The totality of the curved shape represents a kernel density estimation where wider sections represent a higher probability of a given value and thinner sections represent a lower probability. The grey dashed line represents the grand mean.

https://doi.org/10.1371/journal.pstr.0000122.g009

The TNpp_max (Fig 4) suggested a negative relationship between TNpp_max and mean grow temperature; however, the opposite relationship was produced between Ecosystem fit and mean grow temperature (Fig 10). A positive linear relationship suggests that as temperature increases, the eFit of a crop will increase due to reaching higher tNpp levels (Fig 10). The significant 26°C threshold produced with TNpp_max (Fig 10), is not a threshold produced with eFit. The tNpp included in the eFit estimates show that the range of variance in eFit does not vary between 24°C and 29°C (Fig 7). The dominant soil types support and maintain a range of eFit values (Fig 10), with Luvisols and Nitosols consistently producing a higher eFit than other soil types. These results suggest that eFit is less impacted by temperature than by the dominant soil type.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 10. Relationship of ecosystem fit (eFit) to temperature by dominant soil types (UN FAO).

An average growing season temperature threshold of 26° C is indicated by the dashed red line, and the Pearson’s Product Moment correlation (r) with p-value is reported at the top for each soil type.

https://doi.org/10.1371/journal.pstr.0000122.g010

In contrast to the relationship between TNpp_max and mean annual precipitation (Fig 5), eFit did not reach an upper threshold with mean annual precipitation at 1,200 mm (Fig 11). For example, eFit varied across a precipitation range of 500 to 1,600 mm of mean annual precipitation while the thresholds of eFit were determined by the dominant soil types (UN FAO). The mean annual precipitation varied from the Moist and Very moist precipitation classes. This suggested that eFit levels were not limited by precipitation levels, and crop adaptation to its site will be less impacted by precipitation compared to other site factors. The highest eFits were found in the Nitosol and Luvisol soil types (UN FAO). The lowest eFit levels were found in the Vertisol, Fluvisols and Cambisols soil types (UN FAO).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 11. Relationship of ecosystem fit (eFit) to precipitation by dominant soil types (UN FAO).

Four precipitation classes; dry, moist, very moist, and wet mean annual precipitation thresholds are from Vogt et al. [69]. The optimal range of precipitation for agricultural productivity is indicated by the grey panel (500–1200 mm^-1 yr^-1). Pearson’s Product Moment correlation (r) with p-value is reported at the top for each soil type.

https://doi.org/10.1371/journal.pstr.0000122.g011

Site description and droughts

India was selected as the case study because it’s where the original Green Revolution emerged to address inadequate agricultural productivity to feed the country’s rapidly growing population. India is an Agrarian society and emerged as the most populous nation in April 2023 [88]. India also has a rich history of research and data on its soil types and climate and is experiencing a loss of crops due to droughts [72]. India has experienced many droughts and an increasing frequency of significant droughts over the last 20 years compared to an average year [89]. A report on the meteorological history of droughts recorded the following pattern [89–90]:

1. During 1871–2015, there were 25 major drought years, defined as years with All India Summer Monsoon Rainfall (AISMR) less than one standard deviation below the mean (i.e., anomaly below percent): 1873, 1877, 1899, 1901, 1904, 1905, 1911, 1918, 1920, 194 1, 1951, 1965, 1966, 1968, 1974, 1979, 1982, 1985, 1986, 1987, 2002, 2009, 2014 and 2015.
2. The frequency of drought has varied over the decades. From 1899 to 1920, there were seven drought years. The incidence of drought came down between 1941 and 1965 when the country witnessed just three drought years.
3. However, during the 21 years, between 1965 and 1987, there were 10 drought years which was attributed to the El Nino Southern Oscillation (ENSO). Among the many drought events since Independence, the one in 1987 was one of the worst, with an overall rainfall deficiency of 19% which affected 59–60% of the normal cropped area and a population of 285 million. This was repeated in 2002 when the overall rainfall deficiency country as a whole was 19%.

These droughts are reducing crop yields but at different rates across its agricultural landscapes. India’s agricultural areas experienced either low or high drought frequency between 2000 and 2019 [89]. Agricultural productivity decreased by 40% where farming is dependent on precipitation. The worst drought year in India was 1987, when rainfall was 75% below normal. Kumar [91] reported rainfall was less than 50% of the average, and food grain output productivity of the yield dropped by 20%. Also, Kumar [91] reported that between 1978 and 1983, lands that were not irrigated had a 30–50% decline in yields, but even irrigated lands experienced decreased yields of 10–20%. The food grain productivity decreased by 29 million tons from an expected 90 million metric tons for this year [91].

Data, study design and analysis

This paper used 356 country-wide sites with administrative-level climatic and soil data to demonstrate eFit and tNpp’s utility in developing an internal site standard. We filtered the data to the land cover classification of croplands west of 90° E longitude, with a grow season temperature above 19° C and a sample size greater than eight (n > 8) in each dominant soil type. We explored how much growth-limiting factors decrease tNpp, identified the potential productivity at the site level, and the potential for managers to increase actual productivity.

This study was designed to compare two assessment frameworks to identify sensitivity to site-level factors and which variables determine the potential yields and productivity of a crop at a site level: (1) the light-use-efficiency model (or carbon-centric approach); and (2) the use of eFit based on total tNpp. The second framework incorporates the first framework and expands the assessment to be sensitive to site-level limitations on crop growth by eco-region and dominant soil types in India. The framework aims to estimate a crop’s TNpp_max and actual productive capacity based on local edaphic and climatic conditions under which plants actually grow.

This research explored whether a plant’s photosynthetic potential will be sufficient to determine its productive potential and whether it can link the site-level limits of soils and climates. This framework uses ‘total’ crop productivity, not just the ‘product’ biomass harvested. It provides methods to estimate the maximum photosynthetic potential of a site and the actual productivity to determine which cultivars should be planted at a given site, especially focusing on the soil and climatic characteristics by knowing how the crop may adapt to its site.

Within a GIS we mapped the primary boundary of India to constrain the spatial extent of our analysis (https://map.igismap.com/gis-data/india/administrative_outline_boundary) and level-two administrative divisions (which includes districts and is part of the Global Administrative Areas 2015 (v2.8) dataset). Reference sites consisting of spatial datapoints were developed by calculating the centroid of each second-level administrative boundary division, including districts, that is part of the Global Administrative Areas 2015 (v2.8) dataset (https://gadm.org/).

Prior to further analysis, the northeast region of India consisting of the eight states of Assam, Arunachal Pradesh, Manipur, Meghalaya, Mizoram, Nagaland, Tripura, and Sikkim were removed from the data set as the landscape is dominated by tropical, subtropical, and temperate broadleaf and mixed forests where there is limited agricultural development. Through a series of steps within a geographic information system (GIS), a diversity of data types and sources were gathered and assigned to each site with overlay and spatial relationship functions.

Environmental data included ecoregion classification from the Terrestrial Ecoregion GIS data portal (https://www.gislounge.com/terrestrial-ecoregions-gis-data/) which depicts 846 described ecoregions across the planet. Ecoregions are ecosystems of regional extent, color-coded to highlight their distribution and the biological diversity they represent with the goal of E. O. Wilson’s “Nature Needs Half” initiative to protect half of all the land on Earth to save a living terrestrial biosphere (https://ecoregions.appspot.com/).

Major soil types were downloaded from the FAO soil survey portal (https://www.fao.org/soils-portal/soil-survey/), and land cover classifications, and other climatic and environmental variables consisting of vector and raster data were then spatially joined or summarized to the reference sites for further processing [92]. We included only sites where land cover was classified as “cropland,” which served as the locations to calculate TNpp_max, total productivity, and eFit estimates.

Digital Soil Map for India

India is a very diverse landscape composed of 39 level IV ecoregions (21 used in this study) and 10 biomes (5 used in this study). Six dominant soil types were associated with the reference sites, and an additional ten soil types represented the total diversity of soil orders found in India (Fig 12).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 12. Map of dominate soil types and ecological biomes for study area, India.

Country-wide cropland reference sites (dark points) based on the centroid of administrative units (n = 356). Level IV ecoregions are encompassed within the biomes outlined in yellow and listed by number on the map and described in the legend. The colors indicate the dominant soil types used in this study, and the additional diversity of soil types not covered by a reference site are depicted as dark grey. Data sources for the base map as follows: Boundary of India layer is part of the Global Administrative Areas 2015 (v2.8) dataset. Hijmans, R. and University of California, Berkeley, Museum of Vertebrate Zoology. (2015). Boundary, India, 2015. UC Berkeley, Museum of Vertebrate Zoology. Available at: http://purl.stanford.edu/jm149wc6691; Soil vector data based on the FAO-UNESCO Soil Map of the World available at https://data.apps.fao.org/map/catalog/srv/eng/catalog.search#/metadata/446ed430-8383-11db-b9b2-000d939bc5d8; Ecoregion layer is licensed under CC-BY-4.0 and available at Ecoregions 2017 © Resolve https://ecoregions.appspot.com; Administrative boundaries are derived from OpenStreetMap data licensed under Creative Commons Attribution 4.0 International (CC BY 4.0) and available at GIS Data | MAPOG. There were no changes made to the base layers.

https://doi.org/10.1371/journal.pstr.0000122.g012

The characteristics of the three dominant soil types used in this study from the ICAR classification scheme are described in Table 4.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 4. Summary of India’s three dominant soil types using classification scheme by Indian Council of Agricultural Research (ICAR) and descriptions of the soil characteristics that affect crop growth [70].

https://doi.org/10.1371/journal.pstr.0000122.t004

Climatic Variables

The climatic variables were annual mean, minimum and maximum air temperatures, and total yearly precipitation as continuous data to tease out critical environmental thresholds influencing the growth of crops. In addition, average monthly temperature data and average monthly solar radiation data [93] were downloaded and extracted at the geographic coordinates of each reference site [https://www.worldclim.org/data/bioclim.html] at a spatial resolution of 30 seconds (~1 km²). The summary data calculated for each site were: (i) the length of the growing season (i.e., days when the temperature exceeded zero), and (ii) the mean monthly temperature days for the growing season.

We evaluated the underlying structure of TNpp_max and grow temperature with unsupervised clustering analysis to identify threshold effects. We computed all pairwise dissimilarities (distances) between observations based on a Euclidian metric with determination of the medoid by a robust partitioning method (PAM). The optimal number of clusters was selected by evaluating the average silhouette width and the variance explained by the first two principal components with the R package “cluster”. For the data analyses, precipitation thresholds were grouped into four classification levels described in Vogt et al. [69].

Calculation of Maximum Potential Productivity

To estimate crops’ maximum potential productivity (TNpp_max), a modified Loomis-Williams model initially developed for crops in the 1960s was used [61–62]. Their research supports using photosynthesis as each plant’s assimilation framework and explores site limitations on plant productive capacity [61,63]. A crop’s photosynthetic capacity is limited by the resources constraining its ability to fix carbon, such as its ability to acquire growth-limiting resources at the site level, e.g., its edaphic and micro-climatic conditions [69].

The specific theoretical maximum productive capacity potential was estimated using a light-use efficiency model, based on the amount of solar radiation available during a growing season at each site and its plant physiological parameters [35,93]. The light-use-efficiency model is calculated as a product of solar radiation, light-interception efficiency (Ε_i), the efficiency at which canopies absorb solar radiation, and the conversion efficiency (Ε_c), or the rate at which solar radiation is absorbed by C3 plants and is converted into biomass [35,62,94].

Calculation of Ecosystem Fit

Ecosystem Fit (eFit) is the proportion of productive capacity of the site that can be improved upon using as an index the upper site-level maximum threshold of total net primary productivity capacity for the site based on the site’s limitations to growth [63]. In this calculation, Net Primary Productivity (tNpp) integrates green plant functions and includes changes in carbon allocation shifts in the source-sink relationships [46].

The eFit model was developed for forests. It produces an internal site reference productivity level or an index of the site-level potential productive capacity to a plant’s ecophysiological and evolutionary functions, site-level edaphic conditions, and climatic constraints [62–63]. This indexing approach has not been previously used in agriculture to assess the eFit of a crop to local site growth-limiting conditions. Ecosystem fit cannot be currently calculated using field-collected data for crops in this study because of the lack of a robust database populated by data on a diversity of crop source-sink relationships to seeds, fruit, aboveground plant growth, root biomass, and secondary defensive chemicals. The methodology to calculate eFit is described in Klock et al. [62].

Statistical methods and calculations

To understand the underlying structure and threshold effects of the data, we applied an unsupervised (all identifying classifiers removed) clustering method to TNpp_max, tNpp, and the climatic variables of temperature and precipitation. Clustering reveals the natural groupings inherent in the data and provides an empirical estimation of the thresholds that inform how to characterize the vegetation response across the landscape.

We hypothesized that soil conditions would inform primary productivity. Therefore, we tested for a threshold effect by computing all pairwise dissimilarities (distances) between observations based on a Euclidian metric with determining the medoid by a robust partitioning and aggregation around medoids (PAM). The medoid represents the stabilized median of the clustered data. The optimal number of clusters was selected by evaluating the average silhouette width and the variance explained by the first two principal components with the R package “cluster” [95]. The cluster analysis identified two relatively homogenous clusters for TNpp_max and three clusters for tNpp, thus data summaries were classified by these groups, referred to as low, medium, or high (Tables 2 and 3).

A series of parametric and non-parametric approaches were explored to understand the association between productivity metrics (e.g., TNpp_max, tNpp, eFit), and dominant soil groups and climatic variables. All continuous responses were evaluated for meeting the assumptions of normality (if applicable), and to reduce experiment-wise error rates (Type I error, falsely rejecting the NULL), we adopted a more stringent level of significance for all pairwise comparisons with Bonferroni correction to further control for the probability of committing a Type I error [96]. We tested for homogeneity of variance with a Bartlett test. The responses of TNpp_max, tNpp, and eFit were normally distributed and mostly balanced (assumption of Gaussian distribution) and we used Welch F tests to evaluate the relationships to the dominant soil groups. The Welch test is based on the weighted means of each group (sample size and variance) and the grand mean based on the weights mean of each group for the sum of squares. These tests provided the general benefit of high power and low probability of Type I error. We report the grand mean value in figures to represent that value against which the class means were evaluated. We used partial Omega squared as a measure of effect size as it is widely viewed as a lesser biased alternative when sample sizes are small (e.g., a partial effect size [ω2 = 0.25] would indicate that 25% of the variance was explained by the predictor). We used the standardized effect size to represent the practical significance of our results, instead of relying on statistical thresholds, and to make comparisons among very different responses. We also tested for statistical power of each comparison. All statistical tests were based on (α = .05), with analysis conducted in the programming language R ver. 4.2.2 [97] and geospatial data processed using ESRI ArcDesktop ver 10.8 [98].