There are three wild rice species distributed in modern China: O. rufipogon Griff., O. officinalis Wall. ex Watt and O. meyeriana (Zoll. & Moritzi) Baill. Only O. rufipogon is discussed here as it is the ancestor of O. sativa ssp. japonica, which was domesticated in China (Fuller and Qin, 2009; Fuller et al., 2010; Lv, 2013; Gross and Zhao, 2014; Choi et al., 2017; Ishikawa et al., 2020). O. rufipogon is a perennial that flourishes in wet environments (Pang and Chen, 2002). Sano et al. (1980) found that, when the water level was stable, wild rice populations were predominantly perennials, depending on vegetative propagation from their roots; in temporary swamp habitats, plants relied instead on grains for sexual propagation, following an annual habit. Fuller and Qin (2009) suggested that wild rice stands under more water–stressed conditions produced more grains. Consequently, water availability is a critical factor in wild rice reproductive strategies. The domestication of rice fundamentally changed the life cycle of this plant as domesticated rice is an annual dependent on seed production, while its ancestor is perennial and relies on vegetative or sexual propagation depending on water availability. Overall, assessing the water status of early rice informs our understanding of how human management shaped rice domestication.

The Lower Yangtze River basin (LYRB) (Figure 1) is a center of early rice domestication (Zheng and Jiang, 2007; Fuller et al., 2009). Abundant rice remains have been found at multiple Neolithic sites in this region, ranging from the Shangshan culture (10,000–8,000 BP) to the Guangfulin culture (4,000–3,800 BP) (e.g., Liu et al., 2007; Guo, 2014; Wu et al., 2014; Ma et al., 2016; Zhao, 2018; Zhang et al., 2024). The earliest evidence of irrigation can be traced back to the Early Hemudu culture (around 6,700 BP to 6,500 BP) at the Shi'ao site, with ridges and water inlets/outlets forming paddy fields (Wang et al., 2023) (Supplementary Table 1).

Although early rice remains from the LYRB are generally presumed to represent paddy rice, there is evidence of variation in wetness between and within sites. Phytoliths of rice and associated weeds in rice fields can be classified as fixed types or sensitive types; fixed types appear regardless of the wetness of fields, while sensitive-type phytoliths decrease in proportion when the environment is dry. Thus, the ratio of fixed: sensitive types of phytoliths should reflect the degree of rice field wetness (Weisskopf et al., 2015). Phytolith analysis on Neolithic sites suggested that rice was grown in paddy habitats similar to wild rice at Tianluoshan (6,800–6,500 BP), while drier conditions similar to modern rainfed fields were inferred at Caoxieshan (5,950–5,700 BP), suggesting water control to encourage rice grain production. The drier conditions (drainage to control water level) at Caoxieshan might be realized through small size fields (usually < 10 m²) combined with facilities such as deeply dug reservoirs and channels (Fuller and Qin, 2009). This pattern of paddy fields was commonly found in sites around the Taihu Lake (Ding, 2004, 2010; Qiu et al., 2013, 2014; Gao, 2017; Liu, 2021). At the later site of Maoshan (5,000–4,300 BP), rice again grew under wet conditions in large-scale paddy field systems, which were framed by regular intersecting channels and ridges (Zheng et al., 2014). Alternating wet-dry conditions have been inferred from soil micromorphological analysis at Maoshan (Zhuang et al., 2014).

In addition to watering, manuring regimes of archaeological rice fields have also been considered in previous studies, primarily using geoarchaeological methods. Cao (2008) found that the surface soils of paddy fields from the site of Chuodun had similar values for SOM (soil organic matter), TN (total nitrogen), and TK (total potassium) but higher TP (total phosphorus) and AP (available phosphorus) compared to modern paddy fields, suggesting a long-term application of animal residues as manure. Soil micromorphological analysis provides evidence of animal dung and human feces as fertilizers, as shown at Maoshan (Zhuang et al., 2014). In addition, fungal spores have been interpreted to indicate soil disturbance, erosion and dung decomposition (Zong et al., 2007). Growth experiments can provide novel insights into the impact of factors like watering and manuring on the δ¹³C and δ¹⁵N values of different rice accessions, which directly show the growing conditions of rice plants.

Carbon has two stable isotopes, ¹²C and ¹³C. The ratio between the two stable isotopes is expressed as δ¹³C values (Equation 1), where R_sample and R_standard are the abundance ratios (¹³C/¹²C) of the sample and the standard (PDB), respectively (Farquhar et al., 1982). Carbon isotope discrimination (Δ¹³C) directly expresses the consequence of photosynthesis and is widely used as an indicator for water use efficiency (WUE) of plants (Farquhar and Richards, 1984). Δ¹³C can be expressed using δ¹³C values of atmospheric CO₂ and of plant (Equation 2) (Farquhar et al., 1982).

In plant physiology, plant growth (the accumulation of dry matter) mainly depends on photosynthesis. The majority of carbon in plants comes from immobilized CO₂ from the air (Farquhar et al., 1989; Xiao, 2017). When there is sufficient water supply, ¹²CO₂ is preferentially fixed through photosynthesis over ¹³CO₂, resulting in a low abundance of ¹³C in the fixed CO₂. However, when water availability is limited, the fractionation of ¹³CO₂ is less significant and the δ¹³C value increases (Farquhar and Richards, 1984; Farquhar et al., 1989; Ehleringer et al., 1993; Wallace et al., 2013; Styring et al., 2016). This is because in a dry environment, stomata in leaves will close to reduce evaporation. Plants then preferentially take up CO₂ from inter–cellular space, leading to the remaining dissolved CO₂ in cells to become enriched in ¹³C. In addition, lower stomatal conductance is disadvantageous for external CO₂ to enter leaf cells. Thus, paddy rice grown in waterlogged conditions is expected to have lower δ¹³C values than dry rice, as previously observed in west Asian cereals like wheats and barleys (e.g., Araus et al., 1997, 1999; Wallace et al., 2013, 2015; Bogaard et al., 2016; Styring et al., 2016).

Previous studies have observed a relationship between the stable carbon isotopic composition of rice and water availability (see citations in Supplementary Table 2). For example, Gao et al. (2018) reported pot experiments showing that, under water stress, rice leaves of both japonica landrace Qiuguang and heritage indica landrace Qishanzhan had higher δ¹³C values than leaves of better watered plants. Furthermore, high water stress was associated with a higher δ¹³C value compared with low and medium water stress conditions.

Kaushal et al. (2016) analyzed rice grains in India and proposed that, for the same genotype of rice, higher δ¹³C values were found in dry season rice compared to wet season rice. This could be a result of both a drop in relative humidity and increase in water salinity (as sea water incursion led to the enrichment of the heavier isotope, ¹³C). Kaushal and Ghosh (2018) also found that δ¹³C values reflect relative humidity. They tested 105 rice genotypes grown in distinct topographic settings and climatic zones in the southwest monsoon season in India. A significant negative trend was found between δ¹³C values in rice grains and relative humidity using linear regression. Similar results were observed by Nayak et al. (2022), such that rice grains from the South Deccan (a semi-arid region) had higher δ¹³C values than those from East India (a monsoon region).

Different rice genotypes have different water use efficiency (WUE), shown by Δ¹³C values of rice leaves (Impa et al., 2005) or various rice plant organs (such as panicles, though grains were not tested) (Kano-Nakata et al., 2014). Centritto et al. (2009) studied leaves of five upland-adapted cultivars and three lowland cultivars and found greater variations in Δ¹³C values across genotypes than across water conditions. These findings showed that variations between rice genotypes/accessions also require consideration when interpreting carbon isotopic values of archaeological rice samples. The most direct inferences on water status are accessible from variation within local rice populations in archaeobotanical assemblages.

Nitrogen has two stable isotopes, ¹⁴N and ¹⁵N. The ratio between the two stable isotopes is expressed as δ¹⁵N values (Equation 3), where R_sample and R_standard are the abundance ratios (¹⁵N/¹⁴N) of the sample and the standard (AIR), respectively (Robinson, 2001).

Nitrogen is the most important mineral nutrient for rice. It participates in the production of a series of important substances, such as nucleic acids, amino acids, phospholipids, chlorophyll and endogenous hormones (Hu et al., 2017). Inorganic nitrogen is the main nitrogen source for plants but only accounts for 1%−2% of the total nitrogen in soils. Apart from a small proportion of organic compounds like amino acids, most organic nitrogen is transformed into inorganic nitrogen by microbes in soils before uptake by plants (Hu et al., 2017). Plants mainly take up two forms of inorganic nitrogen: ammonium (NH4+), produced by biological N₂ fixation (nitrogen-fixing microorganisms convert N₂ into ammonia, which reacts with water and forms ammonium) and mineralization/ammonification (breakdown of complex organic matter molecules into monomeric organic nitrogen compounds and ammonium), and nitrate (NO3-), produced by nitrification (conversion of ammonium into nitrate by nitrifying bacteria) (Denk et al., 2017).

The transformations of nitrogen influence nitrogen isotope fractionation in soils and thus the nitrogen isotope composition of plants (Hedges and Reynard, 2007). In a comparative study of upland crops and paddy rice, Lim et al. (2015) presented a model in which upland vegetables (cabbage, pepper, onion, and garlic) take up N mainly through ammonification and nitrification processes, with nitrate (NO3-) as the main N species. Anaerobic waterlogged conditions differentiate the nitrogen cycle of paddy field rice from that of upland crops because they restrict the nitrification process and nitrate production and increase the rate of denitrification. Denitrification converts nitrate to N₂, causing the remaining nitrate to be enriched in ¹⁵N. Consequently, ammonium becomes the dominant N species in paddy rice, which is directly absorbed by plants. ¹⁵N-enriched nitrate can also be utilized by paddy rice as a secondary N source. Lim et al. (2015) suggested that these sources together make the δ¹⁵N values of paddy rice (c. +5.6–15.5‰) much higher than those of upland plants (c. +0.3–6.4‰), based on isotopic analysis of homogenized aboveground plant parts. Similarly, Yoneda et al. (2019) found that paddy rice from experimental fields in wetlands had around 2.5‰ higher δ¹⁵N values than pot-grown dry rice. The contribution of biological nitrogen fixation (BNF) by diazotrophs such as cyanobacteria and Azolla to contrasting δ¹⁵N values in dryland crops and paddy rice requires further investigation, but it is clear that BNF contributes to the productivity of paddy fields (De, 1939; Ley et al., 1959; Yoshida and Ancajas, 1973; Ladha et al., 1983; Ladha and Reddy, 2003).

Another agronomic factor that potentially increases the δ¹⁵N values of rice is the application of organic manure. Chung et al. (2017) observed that organic rice had δ¹⁵N values above +5‰, while rice grown with synthetic fertilizers had lower values. Yun et al. (2011) found that pig manure significantly increased the δ¹⁵N values of paddy rice compared to those grown with synthetic fertilizer and hairy vetch (Vicia villosa Roth.), though the differences in δ¹⁵N values decreased over time.

Fabaceae forms an important part of green manure applied to modern rice cultivation, such as Astragalus sinicus L., Vicia faba L., and Vicia sativa L. (e.g., Xiao, 1963; Huang and Shi, 1964; Yuan et al., 2011; Xi, 2014). Legumes usually have δ¹⁵N values around 0‰ and the application of such green manure to crops like wheat, barley and potatoes was reported to reduce crops' δ¹⁵N values compared to those of the conventional group and the group receiving pig manure (Laursen et al., 2013). Long-term fertilization would not change the δ¹⁵N values of legumes unless there was rather high level of organic manure input over long time (Fraser et al., 2011). Another factor that may dramatically increase the δ¹⁵N values of legumes is using organic manure with high δ¹⁵N values, such as seabird guano (Szpak et al., 2014). In the LYRB during the Neolithic, only a few Fabaceae remains were found at some of the archaeological sites (Qiu, 2015; Gao, 2017) and whether they were used as green manure in rice cultivation is little discussed. The organic manured paddy rice samples in this study include rice grown both with and without green manure of Vicia faba (see below), and can be used as a reference when comparing with archaeological samples in the future.

There have been various recent applications of stable carbon and nitrogen isotope analysis to modern rice in agronomic research (see citations in Supplementary Table 2). These isotopic values have been measured to assess the growing environment (Liu et al., 2012; Zhang et al., 2014), to quantify root interaction between rice and weeds (Gealy and Fischer, 2010; Gealy and Gealy, 2011), to evaluate planting methods (Cai and Fu, 1985; Li, 2012; Zhao, 2017), to estimate yields (Laza et al., 2006; Wang, 2016), to select featured landraces (Xiong et al., 2017), to authenticate organic crops (Chung et al., 2017), and/or to differentiate their geographical origin (Korenaga et al., 2010; Chen et al., 2016).

There have as yet been few archaeobotanical studies using stable carbon and nitrogen isotope analysis of rice to infer management (e.g., Kaushal et al., 2019; Shoda et al., 2021; Nayak et al., 2022) or to inform palaeodietary reconstruction (e.g., Wu, 2017). Research in China has mainly focused on the LYRB (Gupta et al., 2013; Wu, 2017) and the Central Plains (Tao et al., 2022a,b; Wang et al., 2022; Tao et al., 2023), two important regions where changes in farming practices and social complexity occurred. In the LYRB, studies have been conducted at the site of Tianluoshan, where the process of rice domestication was ongoing and foraged resources played a major role in human diet (Fuller et al., 2009). Gupta et al. (2013) reported higher δ¹⁵N values of bulk samples of rice husks (+7.4‰, n = 1, ca. 4,900 BC) and one charred grain (+4.1‰, 4,700–4,500 BC) at Tianluoshan compared to modern chemical fertilized materials collected from the same province (−3.8‰ and +0.8‰, respectively) and suggested the possible use of human waste as manure. However, a baseline of unmanured modern rice from the local environment was lacking, and the reduction in δ¹⁵N values of modern grains due to mineral fertilizer (Yoneyama et al., 1990) was not taken into account. The inference of manuring at Neolithic Tianluoshan therefore requires further investigation.

Wu (2017) analyzed charred rice grains from layer 8 (4,800 BC) to layer 4 (4,500 BC) at Tianluoshan, with three to five single grains analyzed for each layer to complementing the previous approach based mostly on husk remains. It was observed that more than half of the rice grains yielded δ¹⁵N values higher than +4‰, as was the case for other plant remains, such as water chestnut, south sour jujube, foxnut, acorn, plum and persimmon. This finding suggested that the commonly used δ¹⁵N value of +0–4‰ for plant foods as a baseline in dietary analysis was too low for archaebotanical materials from this region.

Similar challenges of identifying relevant baselines to interpret isotopic data were encountered in studies of rice grains from northern China. These studies (Tao et al., 2022b; Wang et al., 2022; Tao et al., 2023) argued that rice was grown in well-watered, fertile conditions in the Central Plains from around 2,600 BC to 221 BC based on δ¹³C and δ¹⁵N values, but their baselines for watering and manuring levels were derived from experiments with wheat, barley and lentil in western Eurasia (Bogaard et al., 2013; Wallace et al., 2013). Whether or not it is appropriate to transfer thresholds of dryland crops in western Eurasia to rice in East Asia remains open to question. Tao et al. (2023) cited values of rice grains from wetter eastern India and drier South Deccan in Nayak et al. (2022) but used baselines for wheats in Wallace et al. (2013) to interpret δ¹³C values of rice remains from northern China as well-watered plants. Furthermore, discussion of manuring regimes in these studies needs reconsideration. For paddy rice, the δ¹⁵N values cannot be simply explained as the result of manuring effects if paddy fields naturally have relatively high δ¹⁵N values caused by denitrification, which may obscure any additional effect of manuring.

Recent work was also conducted on charred rice grains from six sites in the Korean peninsula dating from ca. 1,100 BC to 400 AD (Shoda et al., 2021). The consistency of δ¹³C and δ¹⁵N values through time led the authors to question the prevailing hypothesis that there was a shift from dry to wet cultivation. The δ¹⁵N values of rice were consistently higher than those of dry crops during this 1,500–year period. The authors suggested that the elevated δ¹⁵N values of rice were due to denitrification in paddy field conditions and therefore that paddy rice was grown throughout this period.

Kaushal et al. (2019) applied their observation that modern dry season rice grains have higher δ¹³C values than wet season rice grains to interpret the δ¹³C values of archaeological charred rice grains from seven sites in India belonging to the Harappan and other regional cultures, and inferred changes in climatic (dry/wet) conditions. The authors argued that high δ¹³C values under low precipitation conditions may also reflect low crop productivity as lower stomatal conductance in dry conditions can be related to a decrease in the influx of CO₂ required for photosynthesis. The lower yield of rice could lead to reduced population density and a deurbanized phase in the late Harappan culture.

Nayak et al. (2022) analyzed archaeological charred rice grains from monsoonal East India and the western arid Deccan Plateau (where irrigation channels have been discovered at the site of Paithan) dated from around 1,250 BC to 550 AD. Using modern experimental charred rice grains from both regions as a reference, it was suggested that rice grew in a well-watered environments with varying intensity of manure throughout the study period at both the eastern and the western sites.

Whereas rice remains preserved archaeologically are usually charred grains and spikelet bases, or uncharred/waterlogged spikelet bases (e.g., Zheng and Jiang, 2007; Fuller et al., 2009), previous agronomic studies of modern rice isotope values have focused on other plant parts such as roots, shoots and leaves (see citations in Supplementary Table 2). As a result, these modern rice isotope values cannot be used for direct comparison with archaeobotanical values. This study presents a controlled growth experiment to investigate the relationships among rice δ¹³C and δ¹⁵N values and water and manuring treatments. In addition, samples from modern experimental fields complement those from pot experiments. Field experiments ensure realistic growth conditions and access to physiologically mature grains (Christensen et al., 2022) and scope to explore variability of stable isotope values within and between plots under the same treatment.

Three hypotheses are evaluated in this paper. First, we hypothesize that watering will reduce the δ¹³C values of rice grains. A second hypothesis is that paddy conditions will increase rice δ¹⁵N values compared with rice grown under dry conditions. Our third hypothesis is that organic manure will increase the δ¹⁵N values of rice compared with unmanured rice.

A pot experiment was conducted in a glasshouse at the Department of Biology (previously Department of Plant Sciences), University of Oxford, UK. The environmental conditions of the glasshouse were 16 h of daylight and 8 h dark photoperiod, with a 30°C daytime temperature and a 28°C night-time temperature. The irradiance was 800 μmol of photons/m² per second and the relative humidity was 80%.

The rice grains for this experiment were partly obtained from the collections of the International Rice Research Institute (IRRI), which provided three heritage domestic varieties (O. sativa japonica) from Southeastern Asia (referred to as HER001~HER003, see Table 1). Additional rice grains in this experiment were provided by Dr Xiangdong Fu of the Chinese Academy of Sciences; these will be referred to below as “modern” accessions. Modern rice accessions designated with a “W” refers to the “WuYunjing7” variety (MOD001 and MOD002) (Table 1) and “S” refers to the “Shao314” variety (MOD003 and MOD004) (Table 1). DEP1 (“dense and erect panicle1”) is a genetic locus responsible for determining panicle architecture (density of panicle, grain number per panicle and erectness of panicle), which in turn affects rice grain yields. The mutant DEP1 allele (dep1) (referred to as “dep1” lines), i.e., absence of the DEP1 allele, is a semidwarf form that exhibits increased rice grain yields compared with plants having the DEP1 allele (referred to as “DEP1” lines) (Table 1) (Huang et al., 2009). An initial trial experiment was conducted to determine suitable soil media, growing methods and organic manuring rates (see Supplementary material 1 “Initial trial experiment in 2018” for details).

On 25th May 2018, rice grains of all accessions (Table 1) were placed on moistened paper in petri dishes and moved to a germination cupboard. Germinated grains were planted in pots after 3 days (three grains per pot to allow for unsuccessful growth) and thinned out to one on the 18^th of June. Three levels of watering treatment (low, medium/med, high; see Table 2) and two levels of manuring treatment (1% and 5% manuring level, i.e., 1% and 5% organic matter content in the mixed soil, respectively) were set in this experiment (Table 1). The 5% manuring level was only applied to heritage landraces due to the limited amount of grains of modern accessions. For each variety, 3 pots were grown under each watering treatment. Pots were randomized in terms of position within the glasshouse every 2 or 3 days to reduce potential edge effects and to decrease the influence of uneven artificial lighting in the glasshouse. A nutrient solution (see Supplementary material 1 “Initial trial experiment in 2018”) was applied every 2 weeks.

For all of the pots, a 2–3 cm depth water level above the soil was maintained in each tray until the 24^th of June, when plants reached the 3-leaf stage. The 3-leaf stage is reached when the rice develops three true leaves (the first–developed leaf has a sheath but no blade, so is not counted), which indicates the end of the seedling stage and the start of the tillering stage (Zhejiang Agronomy University, 1980). Before the 3-leaf stage, rice seedlings grow by consuming the substances stored in the endosperm. Subsequently seedlings became autotrophy through photosynthesis (Shao et al., 2008).

Pots of medium and high watering level were moved to deep plastic trays (each containing 3 liters of water) to replicate paddy conditions (2–3 cm of water above soil). Low watering level pots remained in shallow trays and were kept at 100% soil saturation by observing daily whether there was remaining water in trays; if there was no water left, 170 g of water (per pot) was added to achieve saturation. This amount was the capacity of the growing medium (clay granules) in one pot.

A short-day treatment was introduced on the 25^th of June (daylight from 9 am to 6.30 pm) to trigger flowering. Rice plants started flowering in August. By this time pots of medium watering level were moved back to shallow trays for saturated conditions, while high watering level pots remained in deep trays with paddy condition until harvest. The medium watering level was designed with reference to the phytolith analysis by Weisskopf et al. (2015), which suggested that drainage was applied to increase the production of grains in Neolithic fields of the LYRB. Ideally rice plants grown in paddy fields cycle through wet and dry conditions several times during the growth period to provide intermittent air ventilation for roots (Zhejiang Agronomy University, 1980). Rice plants (medium watering level) in this pot experiment were moved from wet to dry conditions once, at the flowering stage.

Rice was harvested from the 17^th to 21^st of September, when the grains were fully ripe. For heritage landrace HER001 (117264), one of three plants under the 5% manuring level, medium and high watering levels did not survive (Table 1). Growth information such as the length of the growing period, the height of rice replicates, water input and so forth is shown in the Supplementary Figures 1–3. Different from other heritage landraces, HER002 (128438) behaved similarly to modern varieties in terms of growth rate (Supplementary Figures 2A–C) while for plant height, heritage landraces were usually taller than modern varieties (Supplementary Figure 3). In addition, for all accessions, high watering level plants had shortest height while low watering level ones were tallest. Similar findings have been observed by Zhu et al. (2013).

Rice samples were also collected from modern, mostly organic fields in South Korea and China from 2016 to 2020 (Table 3, Figure 1), to complement the glasshouse experiment and further test the hypotheses of this paper. These samples cover different water regimes (both paddy rice and dry rice) as well as various manuring regimes (different levels of organic manure and/or chemical fertilizer).

The South Korean paddy rice samples all derived from the same watering and manuring treatment at the same farm in Buyeo (126°50'29.6”E, 36°10'24.1”N, c. 40–60 m asl). Green manure crops (wheat, barley and hairy vetch) were grown every year from October to May before transplanting the rice seedlings. After rice ears fully emerged, water was drained from the fields and organic manure (“ear manure”) was applied using a tractor. The farm was divided into multiple fields and ten fields were randomly selected for sampling. For each field, several rice panicles were randomly collected by a sickle, which formed one sample.

Paddy rice samples were also collected in 2017 and 2019 at the Anyway farm, Shanghai (121°29'45.4”E, 31°13'34.6”N, c. 2 m asl), China. The Shanghai rice grew under the same paddy watering condition but were treated with five different manuring regimes, including a control group (no organic manure or chemical fertilizer added) and two types of organic manure applied at varying rates (see Table 3). In the same plot, the treatment remained the same between years. Plots averaged 0.33 ha in area and ranged from 0.23 to 0.63 ha. The layout of sampling plots with size and manuring regimes is shown in Supplementary Table 5. In addition, rice was rotated with Vicia faba as green manure each year after harvesting, starting from the winter of 2017. Thus, rice samples harvested in the autumn of 2017 (referred to as “Shanghai 2017 samples” below) can be viewed as an example of organic rice without green manure. For sampling, plots were each divided into a north, middle and south section and one sample (usually a clump of rice plants growing together) were collected from each section.

Paddy and dry rice samples collected from Linyi City (118°21'28.9”E, 35°6'14.4”N, c. 50–60 m asl), Shandong province of North China provided another important part of the field samples in this study. In 2019, dry rice samples were collected from three locations, with location 1 either nothing (control group) or applying chemical fertilizer and location 2 and 3 applying chemical fertilizer and organic manure (Table 3). In 2020, dry rice treated with chemical fertilizer and paddy rice treated with chemical fertilizer and organic manure were grown at the same farm, with two plots around 50 m away from each other. Both plots received chemical fertilizer in 2018. The Shandong 2020 dry rice suffered from heavy rainfall in the summer that year, and the ridge around the field impeded drainage of water and the maintenance of dry conditions. At harvest time, half of the field did not produce plants. Since the remaining field was limited in size, samples were randomly collected from plants which produced panicles. It is reasonable to suggest that these samples grew under conditions more similar to paddy rice than normal dry rice. With the exception of Shandong 2020 dry rice, the paddy and dry rice samples from Shandong followed the same sampling method as Shanghai rice samples.

The δ¹³C values of the heritage accessions ranged from −30.2‰ to −27.8‰ at the 1% manuring level (Figure 2A, Table 4, Supplementary material 3). Modern accessions ranged from −30.9‰ to −28.1‰. Modern S-line accessions (MOD003, MOD004) were distributed within the same range of the values as heritage landraces, while W-line accessions (MOD001, MOD002) were around 1‰−1.5‰ lower than heritage accession values. The δ¹³C values of pot experiment samples showed greater variation between accessions than within accessions.

For all accessions at the 1% manuring level, a tendency toward higher δ¹³C values can be observed at the low watering level compared with the medium and high watering levels. However, the δ¹³C values of HER002 (128438) and MOD003 (S-DEP1) increase when moving from the medium to the high watering level. A Spearman's rank-order correlation test was conducted to compare the δ¹³C values of plants grown under three watering regimes. A negative correlation was found in one heritage landrace, HER001 (r_s = −0.738, p = 0.023) and three modern accessions MOD001 (r_s = −0.794, p = 0.019), MOD002 (r_s = −0.738, p = 0.023), and MOD004 (r_s = −0.949, p = 9.585e⁻⁰⁵) (Table 5). The means of the 1% manuring level heritage and modern accessions at three watering levels are shown in Table 4.

The δ¹⁵N values of heritage accessions ranged from +9.6‰ to +16.8‰ at the 1% manuring level and from +10.3‰ to +17.1‰ at 5% manuring level (Figure 2B, Table 4, Supplementary material 3). The δ¹⁵N values of modern accessions ranged from +11.1‰ to +16.9‰. The range of the δ¹⁵N values of HER002 covered the range of all the modern accessions. Overall, higher water input raised the δ¹⁵N values of all accessions from the low to the high watering level and the medium to the high watering level (Figure 2B), with HER001 and MOD004 being exceptions. A negative correlation was observed in one heritage accession HER002 (r_s = 0.843, p = 0.004) and one modern accession MOD001 (r_s = 0.738, p = 0.023) using Spearman's rank-order correlation analysis (Table 5).

Turning to the effect of manuring on δ¹⁵N values, plants grown at the 5% manuring level showed a tendency toward higher δ¹⁵N values compared with those grown at the 1% manuring level, as observed in HER001 and HER003 but not HER002. For heritage landraces the manuring effect on δ¹⁵N values was significant for the medium watering level samples (CI = 95%, p = 0.03595) but not the high watering level samples (CI = 95%, p = 0.1388) (Figure 2C, Table 5).

Based on Interquartile Range (IQR) analysis performed in R, three samples (LY083, SHC30N, and SH325) from fields were identified as outliers and excluded in the following discussion.

Overall, modern field rice δ¹³C values ranged from −28.6‰ to −25.5‰ (Figure 3, Table 6, Supplementary material 3), which were about 2.5‰ higher than samples grown in the pot experiment. For paddy rice, δ¹³C values from South Korea samples ranged from −27.4‰ to −26.3‰. Shanghai samples had a range of −28.6‰ to −26.8‰ and −27.4‰ to −26.7‰ in 2017 and 2019, respectively. Shandong paddy rice in 2020 ranged from −28.3‰ to −27.8‰. For dry rice collected in Shandong in 2019, samples had a range of −26.5‰ to −25.9‰ at Location 1, −26.1‰ to −25.5‰ at Location 2 and −27.2‰ to −25.8‰ at Location 3.

The δ¹³C values of Shandong dry rice 2020 (mean = −27.5 ± 0.1‰, n = 10) separated from most dry rice samples in this study and overlapped with the range of paddy rice, which were in accordance with their actual growing condition due to drainage problems (see Field collections, above). Therefore, these samples cannot be considered as “normal” dry rice and were excluded in the following discussion. The South Korea samples were grown under the same watering and manuring treatment in an organic farm over the long term, with one sample collected from one field, which provides a good reference for the expected range in δ¹³C values of rice grains grown under the same conditions in modern fields. In terms of watering treatment and its potential effect on rice δ¹³C values, the values of paddy field samples (mean = −27.5 ± 0.5‰, n = 92) and dry field samples (mean = −26.3 ± 0.4‰, n = 32) show a significant difference (CI = 95%, p < 2.2e⁻¹⁶) based on the Mann-Whitney U test (Table 7).

Modern field samples had δ¹⁵N values ranging from +0.9‰ to +11.3‰ (Figure 4, Table 6, Supplementary material 3), mostly lower than samples grown in the pot experiment. For paddy rice, the δ¹⁵N values of the South Korea samples ranged from +6.4‰ to +11.3‰. Shanghai samples had a range of δ¹⁵N values from +4.5‰ to +8.1‰ and from +1.9‰ to +4.6‰ in 2017 and 2019, respectively. Shandong paddy rice in 2020 ranged from 5.1‰ to 6.5‰. For dry rice, samples collected in Shandong in 2019 had a range of +1.3‰ to +4.7‰ at Location 1, +0.9‰ to +1.8‰ at Location 2 and +3.0‰ to +7.0‰ at Location 3.

In terms of watering effects on δ¹⁵N values, the control group of paddy rice from the Shanghai 2017 collection (mean = +6.1 ± 1.2‰, n = 9) had higher values than the control group of dry rice from Shandong (Location 1) (mean = +3.9 ± 1.0‰, n = 3) (Figure 5, Table 6). A significant difference between the two groups was observed using the Mann-Whitney U test (CI = 95%, p = 0.03636) (Table 7).

Turning to the effect of manuring on δ¹⁵N values, the organic rice samples of various manuring regimes from Shanghai had very similar δ¹⁵N values in the year 2017, with average values ranging from +5.7 ± 0.8‰ (n = 9, Runbao 5 kg) to +6.1 ± 1.1‰ (n = 9, Runbao 40 kg) (Figure 6). The δ¹⁵N values of the control group (mean = +6.2 ± 1.1‰, n = 8) almost covered the whole range of organic manured groups. The Mann-Whitney U test also showed no difference between the control group and the organic manure group (Table 7). For samples collected in Shanghai in 2019, likely due to the application of green manure of Vicia faba, the mean values of the δ¹⁵N values of rice (+2.7 ± 0.7‰ to +4.0 ± 0.5‰) were lower than those of soils from 2018 (+4.2 ± 0.4‰ to +5.3 ± 0.4‰) (Supplementary Table 7). On the other hand, the average δ¹⁵N values of organic groups in 2019 were still similar to each other and did not differ much from the control group (Table 6).

Since both watering and manuring potentially have a positive effect on δ¹⁵N values while watering has a negative effect on δ¹³C values, statistical tests were applied to see whether there was a consistent correlation between the δ¹³C values and the δ¹⁵N values of unmanured/organic manured paddy rice (without legume green manure). The distribution of the data was first checked using the Shapiro-Wilk test. Only the δ¹⁵N values of Bokashi 5 kg group did not follow a normal distribution. The non-parametric Spearman rank correlation test was conducted on this group while the remaining groups were analyzed using the Pearson correlation test. A negative Pearson correlation between the δ¹³C values and the δ¹⁵N values was seen in South Korea samples (CI = 95%, p = 0.02205), Runbao 10 kg (CI = 95%, p = 0.004017), and Runbao 40 kg (CI = 95%, p = 0.05117) samples of the Shanghai 2017 collection (Figure 7, Table 7). In contrast, there was no correlation in the control, Bokashi 5 kg and Runbao 5 kg groups of Shanghai 2017 collection. South Korea samples received long-term intensive organic manure and Runbao 10 kg, 40 kg groups from Shanghai received short-term but more intensive organic manure. The implication is that the negative correlation between δ¹³C values and δ¹⁵N values can possibly be another indicator suggesting the application of intensive manuring.

Under dry conditions, stomata tend to close to decrease evaporation, resulting in a reduced stomatal conductance. C₃ plants such as rice prefer to use ¹²C for photosynthesis, causing the remaining carbon to be enriched in ¹³C. The lower stomatal conductance prevents ¹²C in the air to freely enter stomata, which further raised the concentration of ¹³C (Farquhar and Richards, 1984; Farquhar et al., 1989; Ehleringer et al., 1993; Wallace et al., 2013; Styring et al., 2016). Consequently, rice under a dry condition would have higher δ¹³C values than rice under a wet condition.

For most pot experiment samples, watering does indeed have a negative effect on δ¹³C values. However, the δ¹³C values of HER002 (128438) and MOD003 (S-DEP1) increase from the medium to the high watering level. This might be due to their physiological features since they grew more quickly and thus their medium watering level samples experienced a shorter period under the waterlogged condition. In addition, pot experiment samples showed larger variations in the δ¹³C values between accessions than between watering levels. This indicated that the archaeological application of δ¹³C values to reveal wetness should primarily be used for comparison within local sequences since variations in water use efficiency between populations can be factored out. Paddy rice samples from the pot experiment generally have higher δ¹³C values than those from fields, which was possibly associated with the rather high relative humidity (80%) and high temperature (28°C to 30°C) in the glasshouse where the pot experiment was conducted.

For field samples, paddy rice also exhibited lower δ¹³C values than dry rice as expected. The δ¹³C values of most paddy rice samples were lower than −26.5‰, while most dry rice samples were higher than −26.5‰ (Figure 3). The δ¹³C values of paddy field rice in this study were comparable to those from previous reported values. Supplementary Table 2 summarizes previous studies with δ¹³C values of wet rice grains (or roots and shoots) from pot experiments or fields in East Asia, Southeast Asia, South Asia as well as Australia and United States. The mean values of paddy rice varied from −29.2‰ to −26.3‰, which in general were similar as the range of the field samples in this study (−28.6‰ to −26.3‰).

There was no difference in the δ¹³C values of paddy rice at different locations, in accordance with previous findings that δ¹³C values alone did not distinguish geographic origins (Chung et al., 2017). The distribution of the Shanghai 2019 samples and South Korea samples largely overlapped with the distribution of the Shanghai 2017 samples. Similarly, the range of Shandong 2020 samples was covered by Shanghai 2017 samples. It was the same case for the dry rice field samples as samples from location 3 covered the distribution of samples from location 1 and overlapped with samples from location 2.

The anaerobic conditions prevalent in paddy fields impede nitrification and nitrate production while increasing denitrification. Denitrification transforms nitrate to N₂ and leads to an accumulation of ¹⁵N in the remaining nitrate. Thus, ammonium becomes the dominant N species in paddy rice, which is directly absorbed by plants. ¹⁵N-enriched nitrate can also be utilized by paddy rice as a secondary N resource (Lim et al., 2015; Denk et al., 2017). As a result, paddy rice shows higher δ¹⁵N values than dry rice.

The significantly higher δ¹⁵N values of the paddy rice control group (mean = +6.1 ± 1.2‰, n = 9) compared with the dry rice control group (mean = +3.9 ± 1.0‰, n = 3) suggested that waterlogged conditions elevated the δ¹⁵N values of rice grains. Previous findings analyzed charred rice grains from the Late Yangshao period to Eastern Zhou period (around 3,500 BC to 221 BC) in the Central Plains, China, and suggested that manuring was applied to rice plants throughout the studied periods based on δ¹⁵N values (Wang et al., 2022; Tao et al., 2022a,b, 2023). However, the mean value of the modern paddy rice control group in this study was higher than that of the Er'ligang period (+5.4 ± 2.5‰) and the Yinxu period (+6.0 ± 2.4‰) at the site of Huqiugang (Supplementary Table 8). In addition, the range of the δ¹⁵N values of the modern paddy rice control group (+1.9‰ to +8.1‰) overlapped with the range of the remaining sites (+7.0 ± 1.7‰ to +8.0 ± 1.4‰) as summarized in Supplementary Table 8. Consequently, paddy conditions could also produce δ¹⁵N values similar to at least part of the data reported in these studies, and whether the data can be fully explained by manuring effect requires reconsideration. Whereas light irrigation has no effect on the δ¹⁵N values of dryland crops (Bogaard et al., 2013; Sanborn et al., 2021), paddy conditions have a significant positive effect on the δ¹⁵N values of rice.

In the glasshouse pot experiment, two manuring levels (the 1% and 5%) were applied to three heritage accessions, of which two showed higher δ¹⁵N values of rice grains at the 5% manuring level compared with the 1% manuring level. Significant differences in δ¹⁵N values were found at the medium watering level but not the high watering level as shown by the Mann-Whitney test, suggesting that manuring did have a positive effect on paddy rice grains, though the influence may vary among accessions and needs to be considered together with watering regimes.

Intensive organic manure in initial seasons did not increase the δ¹⁵N values of rice grains from fields in this study. This finding is similar to Fraser et al. (2011), where short-term experiments did not show marked manuring effect on cereal δ¹⁵N values. The δ¹⁵N values of most organic rice from fields were above 5‰, which was an organic rice threshold suggested by Chung et al. (2017). On the other hand, paddy rice from the control group Shanghai 2017 collection (mean = +6.1 ± 1.2‰, n = 9) and from the chemical and organic group Shandong 2020 collection (mean = +5.7 ± 0.4‰, n = 15) were also above this threshold. Consequently, manuring is not the only factor for interpreting δ¹⁵N values of rice grains and currently no solid threshold of manuring levels can be suggested.

In addition, the negative correlation of δ¹³C values and δ¹⁵N values was found in South Korea samples where long-term intensive organic manure had been applied, and in some of the Shanghai 2017 samples which received one season's high input of organic manure. This finding indicates that the effect of intensive manuring on the δ¹⁵N values may be identifiable, but more data are needed where wetness and manuring vary independently.

This study opens a new way for understanding the watering and manuring status of archaeological rice remains. Based on the δ¹³C values, it is possible to differentiate the relative wet/dry status of rice (Figure 3, Table 7) and to infer management practices such as drainage. Combining the results of the δ¹³C values and δ¹⁵N values makes long term application of organic manure discernible for dry rice (Figure 4, Table 8) and future study of dry rice receiving only organic manure will further refine this framework. Turning to paddy rice, since both paddy conditions and organic manure increase the δ¹⁵N values of rice grains, the effect of these two factors may mask each other when the manure input is of short term or at relatively low level. However, long-term application of intensive organic manure has the potential to be diagnosed through a negative correlation between the δ¹³C values and δ¹⁵N values (Figure 7, Table 8). Future studies could focus on paddy and dry rice samples receiving different durations of organic manure application to understand how δ¹⁵N values are affected by long-term manuring. Experimental growth of paddy rice with varying watering and manuring regimes is another important approach to check whether the positive effect of organic manuring on δ¹⁵N values can be differentiated from the effect caused by denitrification in paddy fields.