The experiment was conducted at the Experimental Farm of the Soil and Water Engineering Department of Hohai University, Nanjing, China (longitude 118°83′E and latitude 31°95′N) during July and October, 2016. The area has a typical humid subtropical monsoon climate with an annual precipitation of 1062 mm. The mean temperature is 15.5°C. The used cylindrical pots were 51 cm in height and had a 16 cm inner diameter. Each pot was firstly filled with 1.2 kg gravel-sand soil at the bottom and then covered with 8 kg of dry soil. A drainage hole at the bottom of each pot and a movable basin under the pot were used to collect percolation water. Detailed information about the experimental pots was reported in Wang et al. (2022).

The experiment had nine treatments, consisting of three water regimes and three soil clay contents. Each treatment was replicated four times. The pots were placed under a plastic shelter on a randomized complete block design. For the three water treatments, the pots maintained 25 mm of water over 7 d after transplanting to ensure plant establishment. After that, the upper limit in all the treatments was set as 30 mm flooding water, and the lower limits were 100%, 90%, and 70% of saturated soil water content, respectively (denoted as I₁₀₀, I₉₀, and I₇₀, respectively). The specific irrigation process in this study is shown in Wang et al. (2022). The soil treatment was controlled using three different soil clay contents, i.e. 40%, 50%, and 60% (denoted as S₄₀, S₅₀ and S₆₀, respectively). The original soil (i.e. S₄₀) had a sand, silt, and clay fraction of 20.81%, 38.94%, and 40.25%, respectively. The S₅₀ and S₆₀ treatments were formed by mixing with respective amounts of pure clay. The selected physicochemical properties and corresponding measurements of soil are same as reported in Wang et al. (2022).

Two seedlings of rice (Oryza sativa L cv. Nanjing44) were transplanted in each pot on 20 July, 2016. Potassium phosphate (0.10 g P kg⁻¹ soil) and potassium sulfate (0.13 g K kg⁻¹ soil) were applied and incorporated before transplanting. In addition, all pots were fertilized with urea (0.15 g N kg⁻¹ soil) in a four-split-application during vegetative and reproductive growth stages.

The original pot weight with dry soil was recorded. The gravimetrical soil moisture content was measured by weighing the pots based on weight loss every day:

The length, depth, and width of soil cracks were recorded by a steel rule with a 2mm diameter steel rod when the soil water content reached the lower limit of irrigation (100%, 90%, 70% saturated moisture respectively) before each irrigation event. The soil crack volume was calculated by assuming triangular shape of the cracks (Bandyopadhyay et al., 2003):

where d, w and l are the depth, width, and length of the crack (cm), respectively.

The data of average crack volume was shown in Supplementary Figure 1 .

All data were analyzed with SPSS software (version 13.0, SPSS Inc., Chicago, IL, USA). The data were firstly tested for normality and homogeneity using the Shapiro-Wilks test and the Cochran’s C-test, respectively. Then, differences between either irrigation regimes or soil clay content for the variables measured were tested using two-ways analysis of variance. When significant differences were detected, multiple comparisons of means were carried out with Duncan’s test at a 5% confidence level. In addition, a linear regression analysis was carried out to determine the relationship between (i) WUE and δ¹³C_organs, (ii) WUE and δ¹⁵N_organs, and (iii) δ¹³C_organs and δ¹³C_whole-plant. Pearson correlation analysis was performed to test for correlations among δ¹⁵N_organs, δ¹³C_organs, [C] _organs, C allocation rate to diverse organs, SPAD of flag leaves, and crack volume at a 5% confidence level.

There were significant differences (p< 0.01) in the SPAD values under different water regimes and soil clay contents ( Figure 1A ). The AWD application (I₉₀ and I₇₀) decreased the SPAD values compared with flooding regime, whereas the elevated clay content significantly increased the SPAD. Compared to I₁₀₀, the SPAD values under the I₉₀ and I₇₀ regimes decreased by 6.07% and 14.01%, respectively, across soil clay contents. The SPAD values under S₅₀ and S₆₀ increased by 8.42% and 16.08%, respectively, compared to S₄₀, across water regimes. There was no significant (p > 0.05) interaction between the water regime and clay content on SPAD values. In addition, as shown in Figure 1B , the panicle length was significantly influenced by irrigation as well as soil clay content. Across soil clay content, the I₇₀ and I₉₀ regimes significantly decreased the panicle length by 31.60% and 6.49%, respectively, compared to I₁₀₀. Across irrigation regimes, the panicle length notably increased with the increased clay content (p<0.01).

Figure 2 shows the effects of water regimes and soil clay contents on the dry biomass of rice organs, evapotranspiration (ET) and WUE_ET. The two-way analysis of variance revealed significant (p< 0.01) differences between the total dry biomass of rice due to the application of different water regimes and soil clay contents ( Figure 2A ). The total biomass notably increased with elevating clay content, but decreased with a reduction in lower-limit of AWD. Compared to S₄₀, the soil treatment S₅₀ and S₆₀ increased the total biomass by 21.72% and 46.65%, respectively, across irrigation regime. The ET was only significantly affected by the irrigation regimes (p<0.01). The ET under I₁₀₀ was significantly higher than that under I₉₀ and I₇₀ regimes, across soil clay content. There were also significant (p< 0.01) differences for total transpiration under different water regimes and soil clay contents ( Supplementary Figure 2C ). With reduction in the lower-limit from 100% to 70% of saturated water content, the transpiration values decreased by 16.39% across soil clay contents. In addition, the transpiration value under S₄₀ was 15.62% greater than that under S₆₀, across water regimes. There was no significant interaction between the water regime and soil clay content on transpiration (p > 0.05).

Figure 2C showed significant (p< 0.01) differences in WUE_ET under different water regimes and soil clay contents. I₁₀₀ resulted in the highest WUE_ET while the lowest WUE_ET value was observed under I₇₀, when analyzed across the soil clay contents. With the increase in soil clay content, the WUE_ET significantly increased. WUE_ET under S₅₀ and S₆₀ increased by 22.97% and 56.08%, respectively, compared to S₄₀, across the water regimes. As shown in Figure 2C , no interaction (p > 0.05) on WUE_ET was found for the water regime and soil clay content treatments. The effect of water regimes and soil clay contents on the WUE_T and WUE_I is shown in Supplementary Figure 3 . Increased clay content significantly enhanced both WUE_T and WUE_I. For WUE_T, across soil clay contents, the highest value and the lowest value were was found in I₉₀ and I₇₀ application, respectively.

For all organs of rice, leaf C concentration ([C]) was significantly affected by soil clay content application (p<0.05, Table 1 ). The C allocation in the grain and leaf was both affected (p< 0.01) by water regimes and soil clay contents ( Table 1 ). Differences (p< 0.01) in stem C allocation were also found under the different water regimes. The root C allocation varied with the soil clay contents (p< 0.05). Specifically, C allocation in the leaf was higher (p< 0.01) under I₇₀ than that under I₁₀₀, across the soil clay content treatments, whereas a contrary trend was observed for grain. The stem C allocation was highest in I₇₀ and lowest in I₉₀. With increased clay content, across the irrigation treatments, the C allocation in the grain markedly decreased, but the C allocation in the root significantly increased. For example, the highest value of root C allocation, as an average, was 11.72% under S₆₀ while the lowest value was 9.30% in S₄₀.

Under the I₁₀₀ and I₇₀ regimes, significant differences were found among the three soil clay contents for C allocation of several organs. However, under the I₉₀ regime, it was similar among the three soil clay contents for all organs. For instance, when increasing the soil clay content from 40% to 60%, the grain C allocation decreased by 9.03% in I₁₀₀ and 10.48% in I₇₀. The interaction between the water regimes and clay contents was not significant (p > 0.05) for any [C] and C allocation in the different organs.

The δ¹³C_grain was affected (p< 0.05) by the soil clay content and the interaction between water regime and soil clay content ( Figure 3A ). With the increasing soil clay content, the δ¹³C_grain decreased under I₇₀ and I₁₀₀ regime, but showed a different trend under the I₉₀ regime. Across water regimes, the δ¹³C_grain under S₄₀ (-25.70‰) was significantly higher than that under S₆₀ (-26.97‰). The lowest and highest δ¹³C_grain (-27.49‰–24.94‰) were found under the I₇₀-S₆₀ and I₁₀₀-S₄₀ treatments, respectively.

The δ¹³C values in both leaf and the whole-plant level were significantly influenced by the interaction of water regimes and soil clay contents ( Figure 3B, E , p<0.01). For I₇₀, the δ¹³C_leaf was highest in S₄₀, followed by S₅₀ and S₆₀. For I₉₀, the δ¹³C_leaf showed an increasing trend with the elevated soil clay content. For I₁₀₀, the δ¹³C_leaf decreased with clay content in the order of S₅₀, S_40, and S₆₀ applications. The δ¹³C_whole-plant showed similar variations as the δ¹³C_leaf with varied soil clay contents. The lowest δ¹³C_leaf and δ¹³C_whole-plant existed under I₉₀-S₄₀, while the lowest values of δ¹³C_leaf and the δ¹³C_whole-plant existed under the I₉₀-S₆₀ and I₇₀-S₄₀, respectively. In addition, δ¹³C increased slightly from shoot to root. Especially, under I₁₀₀-S₄₀, the δ¹³C in grain was significantly higher than that in other organs.

There was no significant relationship found between WUE_ET and δ ¹³C in rice organs based on the pooled data of nine treatments ( Supplementary Figure 4 ). Nevertheless, across the soil clay content treatments, there was a significantly positive relationship between WUE_ET and δ¹³C_leaf (R²=0.73, p<0.01) under I₉₀ irrigation regime ( Figure 4B ), whereas there was no significant relationship between δ ¹³C_organs and WUE_ET under I₁₀₀ or I₇₀ AWD regimes ( Figure 4A, C ). The WUE_I and WUE_T presented a similar trend for their relationships with rice δ ¹³C_organs ( Supplementary Figure 5 ). Significant relationships were found between (i) WUE_I and δ¹³C_leaf (R²=0.78, p<0.01), (ii) WUE_I and δ¹³C_stem (R²=0.39, p<0.05), and (iii) WUE_T and δ¹³C_leaf (R²=0.71, p<0.01) under I₉₀ regime. Across the irrigation regimes, no significant relationship was observed between δ¹³C_organs and WUE_s under any soil treatments (p>0.05, data not shown).

Furthermore, a linear regression was carried out to reveal the variation tendency of the δ¹³C_whole-plant with the δ¹³C_organs under three water regimes ( Figure 4D-F ). The δ¹³C_whole-plant could be expressed as a function of the δ¹³C_grain under I₁₀₀ (R²=0.38, p< 0.05) and I₇₀ regimes (R²=0.73, p< 0.01), respectively. For I₉₀ regime, there were significantly positive relationships between (i) δ¹³C_whole-plant and δ¹³C_grain (R²=0.82, p<0.01), (ii) δ¹³C _whole-plant and δ¹³C_stem (R²=0.86, p<0.01), and (iii) δ¹³C_whole-plant and δ¹³C_leaf (R²=0.59, p<0.01).

The SPAD values are positively correlated with the WUE_ET across the irrigation regimes (R²=0.72, p<0.01, Figure 5 ). Additionally, there was a significantly negative correlation between WUE_I and δ¹⁵N_grain (R²=0.14, p<0.05), and between WUE_I and δ¹⁵N_leaf (R²=0.22, p<0.01) ( Figure 6 ) based on pooled data. The WUE_ET was negatively correlated with δ¹⁵N in grain (R²=0.24, p<0.01), stem (R²=0.15, p<0.05), and leaf (R²=0.21, p<0.01) ( Figure 6 ). Similarly, the WUE_T was significantly and negatively correlated with grain (R²=0.23, p<0.01), stem (R²=0.16, p<0.05), and leaf (R²=0.20, p<0.01) ( Figure 6 ).

Pearson’s correlations among SPAD, δ¹³C and δ¹⁵N of diverse organs are shown in Table 2 . The δ¹⁵N_grain (r= -0.50, p< 0.01), δ¹⁵N_stem (r= -0.42, p< 0.05) and δ¹⁵N_leaf (r= -0.44, p< 0.01) showed a strong negative correlation with SPAD value. In addition, a significantly positive relation between (i) δ ¹³C_grain and δ ¹³C_stem (r= 0.40, p<0.05), (ii) δ ¹³C_grain and δ ¹³C_leaf (r= 0.41, p<0.05), and (iii) δ ¹³C_stem and δ ¹³C_leaf (r= 0.54, p<0.01) were found in the current study.

The correlations between the crack volume, [C] and the C allocation of diverse organs are shown in Supplementary Table 1 . With increasing crack volume, the C allocation to grain decreased in varying degrees (r=-0.69, p<0.01) while the C allocation to stem and leaf increased (r=0.47, p<0.01 and r=0.81, p<0.01, respectively). However, the crack volume did not have a significant correlation with any organs [C] ( Table S1 ). Furthermore, the [C]_grain was negatively correlated with the [C]_stem (r=-0.40, p< 0.05). For the C allocation of diverse organs, the C allocation to stem and leaf in rice showed a strong negative correlation with grain (r=-0.66, p<0.01 and r=-0.69, p<0.01, respectively).

Water-saving irrigation regime, such as AWD, can improve WUE while maintaining or even increasing rice yield (Song et al., 2021). Consistent with this, in the current study, an increased WUE_T of rice was observed when reducing the lower-limit of irrigation from 100% to 90% of saturated water content, though the differences were not statistically significant ( Supplementary Figure 3 ). Nevertheless, when the irrigation lower-limit further decreased to 70% of saturated water content, a significantly lower WUE_T was observed compared to I₁₀₀ and I₉₀ regimes ( Supplementary Figure 3 ), which mainly due to the higher degree of water deficit in AWD under the I₇₀ treatment (Carrijo et al., 2017). It has been reported that slight drought stress could induce partially stomatal closure, hereby decreasing transpiration and improving WUE_T (Ma et al., 2021). Whereas, severe drought stress impaired carbon fixation and physiological disorders in plants, such as the reduction of photosynthetic capacity, leading to decreased WUE_T (Wang et al., 2020). A previous study also indicated that SPAD values of leaves were closely related to photosynthetic capacity (Wang et al., 2012). The significantly lower SPAD values of I₇₀ compared to I₉₀ and I₁₀₀ ( Figure 1A ) could indicate the decreased photosynthetic capacity in I₇₀. Consequently, the biomass of I₇₀ was significantly decreased with a reduction in WUE_T compared to I₁₀₀ ( Figure 2A and Supplementary Figure 3 ). Similarly, WUE_ET of I₇₀ was observed significantly lower than those of I₁₀₀ and I₉₀ ( Figure 2C ).

In addition to irrigation regimes, the clay content of soil also affected the WUE_ET and WUE_T of rice. In the present study, the WUE_ET and WUE_T increased with the elevation of soil clay contents ( Figure 2C and Supplementary Figure 3 ) (Fotovat et al., 2007). Similarly, Dou et al. (2016) found that, compared to sand soils, the plants grown in clay soil exhibited higher WUE due to better nitrogen status by increased availability of organic matters and water in clay soil as well as improved nitrogen uptake for plants. Consistent with this, higher SPAD values was observed in the presence of elevated clay content ( Figure 1A ), which could increase plant photosynthesis capacity (Wang et al., 2012), in line with higher biomass production under this treatment ( Figure 2A ). WUE_I showed a similar changing trend under different irrigation regimes with varied clay content as WUE_ET and WUE_T ( Figure 2C and Supplementary Figure 3 ), implying that the three WUEs were mainly regulated by photosynthesis associated with biomass production (Fotovat et al., 2007).

Previous studies reported that carbon allocation in different organs of crops could be induced by irrigation regimes (Arndt and Wanek, 2002; Oliver et al., 2019; Liu et al., 2020). In this study, there were significant differences in C allocation to different organs of rice grown under different irrigation regimes ( Table 1 ). Regarding aboveground organs, C allocation to grain decreased significantly with reduced level of irrigation, but a contrary trend was observed for stem and leaf ( Table 1 ). This result is in contrast to the finding of Song et al. (2021) showing that AWD promoted carbohydrate transfer from stem to grain compared to flooding irrigation. A possible explanation is that moderate drought stress could increase the transfer of C to grain, but this could not occur under severe drought stress due to the drought damage (Oliver et al., 2019). Additionally, Liu et al. (2020) suggested that when plants were exposed to drought stress, the new carbohydrates were preferentially transported from shoot to root, and consequently resulted in an increase of C allocation to root. However, in the current study, the C allocation to belowground organs were not affected by irrigation regimes ( Table 1 ). The phenomenon may be attributed to the larger cracks formed in AWD, which could stretch and tear the roots and might influence the C allocation among organs by inducing root-pruning signal ( Supplementary Table 1 ) (Bordoloi et al., 2020).

In addition to irrigation regime, soil clay content also affected the C allocation of different organs in rice ( Table 1 ). Under I₁₀₀ and I₇₀ treatment, C allocation to grain varied with clay contents ( Table 1 ), possibly due to the changed water and nitrogen status of soil as affected by the increased clay contents (Fotovat et al., 2007), as previous studies indicated that C allocation of plants differed significantly with respect to the water and fertilizer conditions (Xu et al., 2007). Nonetheless, under the I₉₀ treatment, similar C allocation was observed among organs under different clay contents ( Table 1 ). The possible reason for this discrepancy is ascribed to the soil clay content-induced cracks. For the I₁₀₀ regime, no soil crack was observed with high soil water potential ( Supplementary Figure 1 ). Therefore, the different clay contents under the same irrigation regime led to the differences in the ability of soils to retain C, water, and nutrient ions, which might impact the C allocation to grain and leaf ( Table 1 ) by affecting plant photosynthetic capacity (Zhao et al., 2021). For I₇₀, the crack volumes were positively correlated with soil clay contents ( Supplementary Figure 1 ), which in turn may significantly influence the soil water and fertilizer contents due to enhanced preferential flow (Cheng et al., 2021). Hence, the C allocation pattern in rice was changed accordingly ( Table 1 ). Regarding I₉₀, the potential increase in N leaching loss associated with enlarged crack volumes (detailed information shown in Wang et al., 2022) was largely offset by the rise in nitrogen retention capacity associated with increased clay content, thereby restricting the variation in the availability of water and nitrogen in soil and C allocation in rice.

It has been widely accepted that variation in allocation patterns in plants could result in δ¹³C changes in plant organs (De Souza et al., 2005). Compared to autotrophic organs (leaves) that supply plants with carbon, heterotrophic organs (stems, grains and roots) tend to be rich in ¹³C (Zhang et al., 2015). The difference of δ¹³C values among organs under I₉₀ was significantly smaller than those under I₁₀₀ and I₇₀, though there was still a tendency for increased δ¹³C from leaf to root ( Figure 3 ). Other potential reasons for organ-specific differences in δ¹³C could be related to the differences in fractionation processes during the enzymatic reactions, and the chemical composition of different organs, such as the amounts of lipids and lignin (Kano-nakata et al., 2014; Zhang et al., 2015). It was found that δ¹³C_whole-plant also showed a strong correlation with δ¹³C_leaf under the I₉₀ regime ( Figure 4E ), which was in agreement with the finding of Gouveia et al. (2019b). Moreover, δ¹³C_grain showed the most consistent and significant correlation with δ¹³C _whole-plant under three irrigation regimes (I₁₀₀, I₉₀ and I₇₀) ( Figure 4D-F ). This result may be attributed to the isotopic fractionation during the allocation and transfer processes of carbon within plants (Sanchez-Bragado et al., 2014). As shown in equation 4, δ¹³C _whole-plant was the integrated δ¹³C values of different organs. Similarly, as indicated by Araus et al. (1993) and Zhu et al. (2021), δ¹³C_grain was the result of the combined δ¹³C values of assimilates produced by different photosynthetic organs, such as the ears and the flag leaves, responsible for grain filling after anthesis, and the remobilization of nonstructural carbohydrate reserves stored in the specific organ, such as the sheaths and culm. Thus, accordingly, δ¹³C_grain showed the most consistent correlation with δ¹³C _whole-plant across different treatments, and might be a priority indicator of δ¹³C _whole-plant. The strong negative correlation between [C]_grain and [C]_stem found in this study (p<0.05, r=-0.40, Supplementary Table 1 ) further supported the aforementioned speculation. However, there were weak correlations between other δ¹³C_organs (δ¹³C_leaf, δ¹³C_stem, δ¹³C_root) and δ¹³C_whole-plant under I₁₀₀ or I₇₀ treatment ( Figures 4D, F ). We speculated that this weak correlation may be related to variations in carbon allocation patterns under the two regimes with varied soil clay contents ( Table 1 ).

Farquhar and Richards (1984) reported that C_i/C_a was negatively related to WUE_i, while C_i/C_a was negatively related to organ δ¹³C. Thus, it could be concluded that there was a positive relationship between organ δ¹³C and WUE_i. However, in this study, there was no significant correlation between δ¹³C_organs and WUE_whole-plant based on pooled data ( Supplementary Figure 4 ). This phenomenon could be explained as follows. First, the possible varied leaf boundary layer conductance among different treatments might result in WUE_i independence from stomatal conductance, C_i/C_a and δ¹³C (Cernusak et al., 2009). In this study, although we did not measure leaf boundary layer conductance between the intercellular spaces and the atmosphere, the significantly higher water loss through transpiration as well as the increased leaf biomass for I₁₀₀, compared to I₇₀ ( Figure 2A and Supplementary Figure 2 ), might indirectly indicate the varied microclimate and plant statuses caused by different irrigation regimes. Second, the degree of dark respiration, changed mesophyll conductance (g_m) under different environmental conditions, and varied proportions of uncontrolled water loss to the transpiration might disturb the relationship between C_i/C_a, δ¹³C and A_n/T (photosynthesis rate/transpiration rate) (Farquhar et al., 1989), which consequently result in a poor relationship between organs δ¹³C and WUE_whole-plant. Interestingly, when the data were grouped into different irrigation regimes, relationships between δ¹³C and WUE_whole-plant varied with changed irrigation regimes ( Figure 4 and Supplementary Figure 5 ). For I₉₀, there was a significant positive relationship between δ¹³C and WUE_whole-plant, which was in agreement with the findings by Mininni et al. (2022). However, there were no significant relationships between the WUE and δ¹³C_whole-plant in I₁₀₀ and I₇₀ treatment. For I₁₀₀, the poor relationship between δ¹³C of plant organs and WUE_whole-plant might be related to the relatively higher rate of panicle water loss. As indicated by Scartazza et al. (1998), most water loss through panicle was cuticular, which was independent of stomatal regulation and photosynthesis. Hence, high rates of panicle transpiration might disturb the relationships between δ¹³C and WUE in rice. In this study, increased panicle length ( Figure 1B ) as well as greater grain weight ( Figure 2A ) for I₁₀₀ compared to I₉₀ and I₇₀ might indicate that the relatively higher proportions of panicle water loss to total transpiration would disturb the relationship between δ¹³C and WUE in rice under I₁₀₀. Furthermore, changes of carbon allocation pattern for I₁₀₀ with varied soil clay content ( Table 1 ) might also lead to a breakdown in the relationship between δ¹³C_leaf and WUE_whole-plant (Wen et al., 2022). For I₇₀, the weak relationship between δ¹³C and WUE_whole-plant ( Figure 4 and Supplementary Figure 5 ) might be due to physiological damages under severe water deficit, such as severely reduced photosynthetic enzyme activity and consequently disrupted photosynthetic process (Bogati and Walczak, 2022). In addition, for I₇₀, the bigger cracks formed in this water-limited irrigation regime were associated with the stimulated leaching of water and nitrogen. Meanwhile, the tearing effect of cracks on the root system under the high clay content together with the reduced availability of water and nitrogen in I₇₀ might deteriorate the physiological disorder process, which consequently results in breakdown of the relationship between δ¹³C_leaf and the WUE_whole-plant.

In contrast to the relationship between δ¹³C and WUE_whole-plant, which varies with irrigation regimes, we found a significant negative correlation between δ¹⁵N_leaf and WUE_whole-plant based on pooled data ( Figure 6 ). This result is consistent with the findings of Topbjerg et al. (2014) and Yousfi et al. (2012) but contrasting to the results of Cao et al. (2014). WUE_whole-plant is known to be highly associated with WUE_i, controlled by either A_n or g_s, or a combination of both, and could be improved by enhancing the A_n or by lowering g_s. In this study, with the elevated soil clay content, the WUE_ET significantly increased ( Figure 2C ), mainly due to the increased photosynthetic capacity. SPAD has been extensively used to indicate the [N] and photosynthetic capacity in the leaf in the last few years, and a higher SPAD would mean a higher A_n (Fotovat et al., 2007). With the increased clay content, the [N]_leaf increased accordingly, thereby increasing the A_n and WUE_ET. Consistent with this, SPAD was significantly positively correlated with WUE_ET in this experiment ( Figure 5 ). In addition, if the increased WUE_i was only linked to SPAD, then a positive correlation between SPAD and δ¹³C_leaf associated with time-integrated WUE_i would be expected. However, in the present study, there was no clear relationship between SPAD and δ¹³C (p>0.05, Table 2 ), indicating that other factors may control increased WUE_ET, such as decreased stomatal conductance in response to abiotic stress (Desrochers et al., 2022). As mentioned previously, SPAD had a significant positive correlation with clay content ( Figure 1A ). Under water-saving irrigation, with the increase of clay content, larger cracks formed in field could stretch and tear the roots, which is similar to the effect of root pruning (Bordoloi et al., 2020), which could decrease both stomatal conductance and transpiration due to the pruning-induced root signals (Ma et al., 2008; Feng et al., 2022). In this study, the large cracks observed under the same AWD regime with high clay content might suggest a strong stress signal generated by root pruning. Consequently, the stomatal conductance of rice was speculated to be reduced accordingly. Meanwhile, previous studies have shown that reduced g_s would lead to a reduction in the loss of ammonia and nitrous oxide, hence decreasing δ¹⁵N in leaf (Yousfi et al., 2012). Therefore, increased WUE_ET and WUE_T caused by reduced g_s (Yan et al., 2020) was expected to be negatively correlated with δ¹⁵N_leaf, which is consistent with the results of this study.

Another reason for the negative relationship between ¹⁵N and WUE_whole-plant might be the decreased N O 3 − transport from root to shoot of plants exposed to stress-condition (Yousfi et al., 2012). Due to the fractionation induced by nitrate reductase (NR), N O 3 − not assimilated in the roots would be enriched in ¹⁵N and exported to shoots for assimilation, causing an increased δ¹⁵N in shoots relative to roots (Zhang et al., 2017). Hence, stress conditions would restrict N O 3 − transport from the roots to the shoots, therefore increasing ¹⁵N in root while decreasing it in the shoots compared with the flooding irrigation (Yousfi et al., 2012). In our study, the clay contents in soil were positively correlated with SPAD ( Figure 1A ). Moreover, as shown in Supplementary Figure 1 , the soil clay contents were also positively correlated with crack volume under AWD. As suggested aforementioned, higher crack volumes under AWD might imply a strengthened abiotic stress for crop, and consequently reduced N O 3 − from root to shoot, and ultimately resulted in the decrease of δ¹⁵N in shoot. Thus, a negative relationship between SPAD and shoot δ¹⁵N was likely to be found ( Table 2 ). In this study, the C and N allocation to roots were significantly increased with the elevated soil clay content ( Table 1 ), which further demonstrated that the large crack reduced the export of N O 3 − from roots to shoots. Moreover, the WUE_ET showed a significant positive correlation with SPAD ( Figure 5 ), but a negative correlation with δ¹⁵N ( Figure 6 ). Hence, the SPAD values tend to be negatively correlated with the δ¹⁵N ( Table 2 ) and further supported that the increased WUE was probably due to a combination of A_n and g_s. However, it should be noted that ¹⁵N in plants can also be influenced by the variations of soil nutrients (Tang et al., 2017), which was limited by the occurrence of cracks. In this case, the discrimination process for ¹⁵N during the N uptake tends to be slight, resulting in an increased ¹⁵N in plants while the WUE was decreased by the crack. Therefore, a significant negative relationship between the WUE and δ¹⁵N was found ( Figure 6 ), but further trials are needed to examine the actual underlying mechanism.