Nitrogen uptake pattern of dry direct-seeded rice and its contribution to yields in northeastern Japan

ABSTRACT Dry direct-seeded rice (DDSR) cultivation is expected to reduce production costs compared with transplanted rice (TPR); however, its low N use efficiency (NUE) has hindered cost reduction. We conducted DDSR field experiments for 3 years using a standard cultivar (‘Akitakomachi’) and a high-yielding cultivar (‘Yumiazusa’) grown with a single application of coated urea (CU) or split applications of normal urea (NU) to determine factors limiting yield and NUE and compared growth and yields of TPR grown in adjacent fields. DDSR yield was lower than that of TPR at comparable N levels by 11% due to low fertilizer recovery rate, crop NUE (CNUE, yield per unit N uptake), and poor grain filling by meteorological factors. Crop N uptake at maturity in DDSR was similar to or even greater than that in TPR, but poor vegetative growth in DDSR and low crop N uptake until panicle formation (PF) resulted in limited spikelet density and CNUE compared with TPR. Analysis of the N uptake pattern suggests that enhancing early leaf development can improve N uptake until PF and thus CNUE in DDSR. NU yielded as much as CU, so it can potentially reduce the cost of production because of its lower price, but the optimal N proportion and application stage in the split application needs scrutinizing. ‘Yumiazusa’ had better initial growth, CNUE, and yield than ‘Akitakomachi’ in DDSR. To leverage these traits for further yield improvements, we also need to consider suitable phenological characteristics to ensure favorable climatic conditions during the grain-filling period. Graphical abstract


Introduction
In Japan, the population involved in agriculture decreased substantially from 11.7 million in 1960 to 2.1 million in 2015, a reduction of 86% (MAFF, 2018), while the cultivating land per farmer has been increasing. This trend is projected to continue, and labor-saving and cost-effective technologies are expected to gain prominence. Dry direct-seeded rice (Oryza sativa L.) (DDSR) cultivation is thus becoming an important alternative for rice farmers in Japan since it can reduce labor costs compared to conventional transplanted rice (TPR) cultivation (Farooq et al., 2011).
Northeastern Japan is one of Japan's main rice cultivation areas, producing approximately 29% of the national rice production (MAFF, 2021b). This region has comparatively larger paddy fields than the rest of Japan, suitable for large-scale mechanized farming. Although both wet direct-seeded rice and DDSR technologies are available, DDSR has greater labor-saving potential as larger agricultural machinery can be introduced for land preparation and planting practices (Kanmuri et al., 2017). However, DDSR technology has drawbacks preventing it from being the more popular choice in northeastern Japan. First, DDSR yields are generally lower than TPR yields. MAFF (2021a) reported that direct-seeded rice, including dry-seeded and wet-seeded, yielded 7% less than TPR (a comparison of the average of 436 field data sites). Second, DDSR is associated with low N use efficiency and requires greater fertilizer inputs (Farooq et al., 2011) thereby increasing production costs. Third, varieties suited for DDSR have not been studied in cool climates.
Many of these problems are related to the climatic conditions in northeastern Japan, located in a cool temperate zone, where rice productivity has been severely limited by low temperatures (Horai et al., 2013). DDSR seedlings are exposed to low air temperatures from an earlier stage in contrast to the conditions characteristic for TPR seedlings, which are covered or protected in nurseries such as greenhouses. One of the important advantages of ponding water in cool temperate regions is to protect rice plants from low air temperatures, particularly in the early stages of rice growth, as water is generally much warmer than air (known as the water blanket effect) (Shimono et al., 2007). Shimono et al. (2007) demonstrated that in the northern parts of Japan, simulated grain yields could be halved in the absence of surface water even without water stress because low temperatures at shoot bases hampered rice growth and yield. In DDSR, seedlings are grown under upland conditions without a water blanket for more than a month before permanent flooding. These situations possibly limit the yield potential of DDSR in cool climates.
Temperatures are also highly influential on rice N uptake Takahashi et al., 1976), particularly during the vegetative growth stage (Shimono et al., 2012). The effects of N management on the growth and yield of DDSR have frequently been studied in warm areas such as northwestern India, southern United States, and central China (Ali et al., 2015;Griggs et al., 2007;Liu et al., 2014;Mahajan et al., 2011;Qi et al., 2012;Xu et al., 2022) in the absence of cold temperature stresses. Nonetheless, evidence exists that DDSR needs a higher rate of N application than TPR for the same yield (Farooq et al., 2011;McDonald et al., 2006). Little is known about rice N uptake during early-stage DDSR growth in cool climate zones, but limited N uptake of DDSR due to low temperatures could be another reason for the lower N use efficiency (NUE), particularly, the apparent N recovery (ANR).
The use of control-release urea is empirically recommended for DDSR in northern regions, and generally, a greater amount is recommended for DDSR than for TPR (Naro, 2021). The fertilizer price of coated urea is approximately 1.7 times higher compared to normal urea per N g m −2 , thus hindering efforts toward reducing production costs. Additionally, microplastic capsules are used on polymer-coated urea, recently identified as a source of environmental pollution (Katsumi et al., 2020). Consequently, attempts to reduce its use have already started in Japan and will likely expand in the future. Although the split application of urea might represent a low-cost alternative and has been used as a basic method globally, including for DDSR use in Europe and the USA, this needs to be investigated as an effective option for DDSR in northeastern Japan.
High-yielding and good-quality cultivars are an important option for higher profitability; however, a comparison of cultivars for DDSR suitability in northeastern Japan has not been performed. In a cool climate, low temperatures limit growth duration, and without protected nursery periods DDSR takes longer to mature than TPR. Cultivars for DDSR, therefore, need to be early maturing, which is often not associated with high-yielding characteristics. Recently, however, an early-maturing, high-yielding, and good-eating quality cultivar, Yumiazusa (YA), has been bred and released by the National Agriculture and Food Research Organization (NARO) and is proven to be high-yielding for transplanting practices in cool temperate areas (Ohta et al., 2018;Yabiku et al., 2021).
Here, we examined how cultivar and N management combinations affect rice growth and yield under DDSR in a cool climate using a standard cultivar Akitakomachi (AK) and the high-yielding cultivar YA, grown under different N regimes. We focused on the following two points: (1) N uptake relevance on DDSR yields and yield components compared to those of TPR, (2) temperature-driven N uptake pattern analysis of DDSR for split application of normal urea (NU) compared to one-time polymer-coated urea (CU) application.
These results were compared with TPR experiments conducted side-by-side using the same cultivars (Yabiku et al., 2021).

Study sites
Field experiments were conducted at the experimental farm of the NARO Tohoku Agricultural Research Center in Morioka city, Iwate prefecture, Japan (39°45'N 141° 08'E) in the 2018, 2019, and 2020 rice-growing seasons.
In each season, two fields were used: one for DDSR and the other for TPR. The soils of all experimental fields were classified as Andisols with some of their chemical properties shown in Table S1. Climatic data were obtained from the meteorological observation at the research center located 2 km from the experimental fields. Table S1 presents mean air temperature and precipitation from April to October of the experimental years (2018-2020).

Plant materials and N treatments
We used two temperate japonica cultivars: AK, an old but still widely planted cultivar in northeast Japan, and YA, a high-yielding, high-quality cultivar released in 2017 (Ohta et al. 2018). For DDSR, we had three N regimes: two with N fertilizers (CU and NU) and one without (0N). For the CU treatment, we applied 12 g N m −2 of CU fertilizer basally containing three types of slow-release fertilizer differing in the rate and pattern of N release (LP30: 30%, LPS30: 20%, LPS60: 50%, the numbers in these names indicate the days required for 80% release at 25°C (Hara, 2000), Chokuha senyou 211, Kumiai Hiryou Co. Ltd., Iwate, Japan). For the NU treatment, a total of 15 g N m −2 of normal urea was split-applied four times at 40, 66, 82, and 94 days after sowing (DAS) in 2018;at 45, 66, 80, and 94 DAS in 2019;and 52, 67, 80, and 95 DAS in 2020, at doses of 6 g N m −2 , 3 g N m −2 , 3 g N m −2 , and 3 g N m −2 , respectively. In the 0N treatment, no N fertilizer was applied. Equal amounts of P and K were applied basally to all plots at a rate of 3.4 g P m −2 and 3.1 g K m −2 , respectively, on 23 April 2018, 17 April 2019, and 15 April 2020.
All DDSR experiments were laid out in a randomized complete block design, where a factorial combination of two cultivars (AK and YA) and three N regimes (0N, CU, and NU) was randomly assigned to each of the three blocks. The size of each plot was 192 m 2 (40 m × 4.8 m, CU treatment) or 48 m 2 (20 m × 2.4 m, 0N and NU treatment) in 2018, and 81 m 2 (15 m × 5.4 m) or 124 m 2 (23 m × 4.8 m) in 2019 and 2020.
8N was considered a standard N level for TPR, comparable to the CU treatment for DDSR. 18N was used to only analyze the relationship between crop N uptake at maturity and yield components. The N, P, and K were applied basally by compound fertilizer (Iseki coat M002, Iseki Tohoku, Miyagi, Japan) in 8N and 18N at a rate of 8 g N m −2 , and P and K were applied to all plots at a rate of 3.5 g P m −2 and 4.0 g K m −2 . 18N plots were top-dressed with 2 g N m −2 of ammonium sulfate and 4 g N m −2 of coated urea (LP40, JCAM AGRI. Co. Ltd., Tokyo, Japan) at the tillering stage, 2 g N m −2 of ammonium sulfate at the panicle formation (PF) stage, and 4 g N m −2 coated urea (LP30, JCAM AGRI. Co. Ltd., Tokyo, Japan) and 3.3 g K m −2 potassium chloride at the spikelet differentiation stage. The N treatments and cultivars were laid out in a splitblock design for 3 years with three replicates each year.

Cultivation details
In the DDSR fields, the dry soil was plowed, powerharrowed, and laser-leveled every April. Basal fertilizer was applied approximately 3 days before sowing, following which the soil was tilled with a vertical harrow tiller and compacted using a Cambridge roller. Before sowing, seeds were coated with a thiram-based repellent (bactericide) (Kihigen R-2 flowable, Yonezawa Chemical Co., Ltd., Kyoto, Japan). Dried seeds were sown to a depth of approximately 2 cm using a high-speed seeder (Clean seeder; NTP-6AFP and NTP-8AFP, Agritechno Search Co. Ltd, Hyogo, Japan) at a row spacing of 0.3 m on 27 April 2018, 22 April 2019, and 17 April 2020. Immediately after seeding, the soil was compressed by the Cambridge roller for good seedling establishment and impermeability (Kanmuri et al., 2017). The sowing rate was 5.1 g m −2 (2018), 5.7 g m −2 (2019), and 5.7 g m −2 (2020). Flush irrigation was applied about once in 3 days to prevent drying until seedling establishment. From early June to approximately 3 weeks before harvest (late September), the field was kept flooded with 3-10 cm ponding water.
Before permanent flooding, glyphosate-potassium was applied at a rate of 4.8 mL m −2 before emergence (mid-May). Cyhalofop-butyl (0.3 mL m −2 ) and bentazone (2 ml m −2 ) were applied after emergence (early June) using a boom sprayer. In late June (after flooding), a backpack spreader was used to apply a compound herbicide containing pyriftalid, mesotrione, and metazosulfuron at a rate of 90 mg m −2 , 240 mg m −2 , and 80 mg m −2 , respectively.
For TPR experiments, the dry soil was plowed in mid-April, and basal fertilizers were broadcast on the soil surface and incorporated with a rotary tiller during early May. During mid-May, the field was submerged and puddled with a power harrow system. Seedlings were raised in a greenhouse, and at the fifth leaf stage, seedlings were mechanically transplanted on May 17 (2018), 21 (2019), and 19 (2020) with a spacing of 30 × 19 cm. The field was kept flooded from May until the end of August, after which the surface water was drained for harvest in late September.

Plant measurements, sampling, and tissue N determination
Leaf number on the main culm (leaf age, hereafter) was measured on the three plants in each replicate at oneweek intervals from emergence to flag leaf development. At the same time, tillers were counted on 0.3 m 2 (one 1 m row).
From approximately 45 DAS until harvest, we sampled aboveground plant organs weekly or bi-weekly, washed, oven-dried them at 80°C for at least 72 hours, and weighed them to determine the aboveground biomass. The sampled area at each harvest was 0.3 m 2 (one 1 m row). A part of the dried samples was ground, and tissue N concentrations were measured using an elemental analyzer (Variomax, Elementar, Germany). Aboveground crop N uptake (N up ) was calculated by multiplying the biomass and tissue N concentrations.
Harvest was done after the grain-yellowing rate reached 85%. At harvest, plants were collected from an area of 2.4 m 2 (1.2 m (4 rows) × 2 m) in each plot for yield determination. We air-dried the samples, weighed the total mass, threshed and dehulled the grains, and screened them through a 1.85 mm-thick sieve. Straw and grain moisture contents were also measured. Brown rice yield was expressed on the measured sample for yield determination, and the water content of brown rice samples was corrected to 15%.
To analyze N use efficiency, agronomic NUE (ANUE) in the fertilized plots was calculated as: where Y F and Y 0 are brown rice yields from the fertilized and 0N plots, respectively, and N F is the amount of N fertilizer applied.
The apparent N recovery (ANR) in the plots was defined as: where N up_F and N up_0 are N up in the fertilized plot and 0N plot, respectively.
We also derived crop NUE (CNUE), which was defined as yield per unit of N uptake at maturity and CNUE by fertilization (CNUEF) as:

Analysis of N uptake during the vegetative growth stage
N uptake is closely related to temperature conditions (Shimono et al., 2012) and is known to be approximated by the function of cumulative air temperature. In transplanted rice, Takahashi et al. (1976) showed that the accumulated effective thermal index (AETI) could reproduce well the N uptake pattern under different conditions (cropping seasons or years), which is defined as: where θ i � is the daily effective thermal index, which is the product of the hourly air temperature and effective air temperature coefficient (Table S2), averaged for the i-th day (Hanyu & Uchijima, 1962), and N is the number of days from start to end of the calculation. Here, we set i = 1 at the fifth leaf age -the exact date derived from the linear interpolation of the weekly observations. The hourly air temperatures (T j ) were approximated using the following cosine curve (Sun et al., 2018): where T max and T min are the daily maximum and minimum air temperatures, respectively, j is the time of day, and T j is the estimated hourly air temperature at time j.
Previous studies have shown that N uptake has two distinctive phases Takahashi et al., 1976). In phase 1, N up exponentially increases, limited by the crop growth rate, and is expressed as an exponential function of AETI: where N LA5 is N up when leaf age is 5.0 (LA5) and RNR is the relative N uptake (N up ) rate. In phase 2, N up increases linearly, limited by soil N supply.
Accordingly, N up for the vegetative growth period can be characterized by two parameters: N LA5 , and RNR. To derive these parameters, we first plotted the N up against DAS or AETI in each experimental plot (an example in Figure 4 and all plots in Figure S5). Secondly, we confirmed that N up increased exponentially with AETI during the period from the LA5 to the PF dates. Finally, we regressed logged N up on AETI starting from the LA5 date (AETI = 0 at the LA5 date) to derive N LA5 and RNR for each replicate and treatment (Equation 7). These parameters were provided for the statistical analysis. The LA5 date was computed by the relationship of leaf age and date using linear regression in each replicate. The PF date was defined as the date when the average panicle length of the main stem reached 2 mm. The average panicle length was measured every few days in each replicate of CU treatment.

Statistical analysis
Statistical analyses were performed using JMP12.0.1 (SAS Institute, 2015). P values <0.05 were considered significant. We conducted an analysis of variance (ANOVA) for DDSR yield components and N uptake parameters using a split-plot analysis, where the year was treated as the main plot, and cultivar and N treatment as the subplots. The model included three fixed effect variables (Year (Y), Cultivar (C), and N treatments (N)) and one random effect variable (block) with the interactions (Y×C, C×N, Y×N, Y×C×N). Tukey's HSD test was performed on some effects of ANOVA, with a significant level of P < 0.05.
Yield and yield components are related to plant N uptake. The analysis of covariance (ANCOVA) was computed for brown rice yield with each yield component as objective variables, N uptake at maturity stage (N MT ) as explanatory variables, and planting method (DDSR or TPR, dummy variables) as the covariance. There were 81 plots on DDSR and 54 on TPR of each experiment used for the ANCOVA.
Although N uptake before the HD is a critical determinant for spikelet density (Hasegawa et al., 1994;Sasaki, 2007), the N regimes may alter the relative contribution of N uptake at different stages. We, therefore, applied multiple regression analysis for spikelet density as a function of N uptake at PF (N PF ) and N uptake from then to the HD (N PF-HD ). Multiple regression of Model 1 was performed with spikelet density as an objective variable, N PF , N uptake at the HD (N HD ), or N PF-HD , and additional categorical variables: Planting method (DDSR or TPR), Cultivar (AK or YA), and Year (2018, 2019, or 2020). Model 2 was similarly conducted but without considering the above categorical variables.
N LA5 and RNR are indicators of N uptake during the vegetative growth period. We investigated the contribution of initial N uptake and increase rate during the vegetative growth period for N PF ; multiple regression analysis was applied for N PF using N LA5 and RNR including or excluding the categorical variable of Year (2018 and 2019). Model 3 included and Model 4 excluded Year.
The standard least-squares method was used as a fitting methodology. The adjusted coefficient of determination (Adjusted R 2 ), corrected Akaike's Information Criterion (AICc), and Bayesian Information Criterion (BIC) were obtained to compare the two models. Adjusted R 2 is the R square adjusted by the following equation to be able to compare the two different regression models with a different number of explanatory variables: where n is the sample size and m is the number of explanatory variables.
The standardized partial regression coefficient (β) was also computed to determine which parameter had the higher impact on N PF in each regression parameter of the selected model. The variance inflation factor (VIF) was used to consider multicollinearity among the explanatory variables.

Meteorological data
In April -October 2018, 2019, and 2020, mean air temperatures were 0.42°C, 0.92°C, and 0.62°C higher than the 30-year mean, respectively (Table S1). In 2019, the mean air temperature in May was 2.1°C higher than the 30-year mean, whereas the differences from the 30-year mean in 2018 and 2020 were within 1.0°C.

Crop development of DDSR
Emergence occurred at 18-21 DAS, and plants reached the second leaf stage (LA2) at 21-27 DAS in all DDSR experiments (Table S3). Dates of PF and HD in DDSR ranged from 91 to 96 DAS and 114 to 120 DAS, respectively, later than that of TPR and similar across different years and cultivars. The period from the HD to maturity tended to be longer for YA than for AK. The results of tiller numbers and biomass growth are reported in Figures S3 and S4.
Leaf age growth was impacted by the N regimes from the early stages (approximately 50 DAS; Figure  S1). Days from sowing to LA5 were 2-10 days earlier in CU than in 0N and NU (Table 1). The difference in leaf age between 0N and CU persisted until the final leaf development (0.4-1.7 fewer total leaves on the main culm in 0N than in CU (Table 1). About 60% of this difference was accounted for by the difference at the seedling stage ( Figure S2). Leaf age increase in the NU treatment was initially slow but recovered from 50 DAS ( Figure S1), narrowing the gap with CU to approximately 0.25 averaged at the end of the experiment (Table 1).

Yield and yield components of DDSR
DDSR brown rice yield under CU conditions was 452 g m −2 averaged for the 3 years and two cultivars (89% of the average yield of transplanted rice under 8N conditions of the same varieties) ( Table 2, Table S4, and  Table S5). DDSR brown rice yield differed significantly   (Tables S4 and S5). DDSR, dry direct-seeded rice; TPR, transplanted rice. AK and YA represent the cultivar's names, 'Akitakomachi' and 'Yumiazusa', respectively. 0N, CU, NU, and 8N represent the following N treatments: no N treatment, one-time coated urea treatment (12 g N m-2), split application of normal urea (15 g N m-2, 6, 3, 3, 3), and conventional compound fertilizer treatment (8 g N m-2), respectively. Different letters denote significant differences at P < 0.05 by Tukey HSD test. among years, with the 2019 crop showing a considerably lower yield than the 2018 and 2020 crops (P = 0.0174, Table S4). The yields of CU and NU were not significantly different on Tukey's HSD test; however, the NU/CU ratio tended to be lower than 1.0 (Table 2). Of the two cultivars on DDSR, YA had an 8% greater yield than AK, averaged over treatments and years (P = 0.0083), although the cultivar effect was not consistent among years (P = 0.0062 for the Y×C interaction, Table 2 and  Table S4). The yield advantage of YA over AK was apparent in relatively high-yielding years (16% in 2018 and 12% in 2020, Table S4), while AK had a higher yield in 2019 averaged over the three N regimes.
Panicle density and spikelet density were comparable between CU and NU (Table 2), although leaf age ( Figure S1) and tiller number ( Figure S3) progress differed between the two fertilized plots. The NU/CU ratio of panicle and spikelet density was smaller in 2018 than in the other 2 years (Table S4). Conversely, the NU/CU ratio for ripened spikelet (spikelet density × grain filling) and 1000-grain weight was greater in 2018, thus compensating for the limited sink size under NU conditions. The N regime significantly affected 1000-grain weight (P = 0.0163, Table 2), with NU having the heaviest mass, followed by CU and 0N.
The yield advantage of YA over AK was associated mainly with the greater panicle density (8%) averaged over years and N regimes. YA had slightly fewer spikelets per panicle and grain filling than AK, but these differences did not offset the yield advantage of YA. The cultivar differences in these yield components were generally consistent across different N regimes and years (Table S4).

N use efficiency and crop N uptake at key growth stages
N MT of DDSR was 7.9, 11.2, and 11.8 g N m −2 in 0N, CU, and NU, respectively, averaged over 3 years and two cultivars (P < 0.0001, Table 3). It was also similar to or even greater than crop N uptake observed in TPR  (Table S6). DDSR, dry direct-seeded rice; TPR, transplanted rice. 0N, CU, NU, and 8N represent the following N treatments: no N treatment, one-time coated urea treatment (12 g N m-2), split application of normal urea (15 g N m-2, 6, 3, 3, 3), and conventional compound fertilizer treatment (8 g N m-2), respectively. AK and YA indicate the cultivar's names, 'Akitakomachi', and 'Yumiazusa', respectively. PF, HD, MT indicate the panicle formation, heading, and maturity stage, respectively. Different letters denote significant differences at P < 0.05 by Tukey HSD test.
plants undergoing comparable N treatments ( Figure 1). Furthermore, N PF and N MT in DDSR had no significant differences between cultivars, although N HD had a significant Y×C effect (Table 3). N HD in DDSR tended to be higher than in TPR, whereas N PF was lower in both 0N and fertilized plots than in TPR (Figure 1). This resulted in a greater N uptake between PF and HD in DDSR than in TPR plants. In the two DDSR fertilized plots, CU enhanced N uptake by PF more than NU, while NU overtook CU before HD (Figure 1).
The ANR was significantly different between years (P = 0.0181; Table 3), with lower values in 2018 than in the other 2 years (Table S6). The ANR was similar between CU and NU (P = 0.4943), averaging 0.27 in CU and 0.32 in NU for the 3 years, 82% in CU and 97% in NU relative to the 8N conditions of TPR (Table 3, Table S6). Conversely, ANUE was significantly lower in the NU than in the CU (P = 0.0023, Table 3; 28% decrease) for both cultivars. The DDSR/TPR ratio for ANUE was 0.68 when averaged for CU (Table 3). CNUE was significantly different between the N treatments on ANOVA (P = 0.0012; Table 3, Table  S6), and CNUE of the CU treatment was significantly higher than that of the 0N and NU treatments (Tukey's HSD test, P < 0.05). The CNUE of YA was significantly higher than that of AK (P = 0.049, Table 3). CNUEF, the criteria of NUE that excluded soil N supply, showed no significant difference among treatments. However, the CNUEF of the CU was consistently higher than that of the NU (Table 3).

Relationship between crop N uptake at maturity and yield components
We examined the relationships between N MT and yield components and compared them between DDSR and TPR to determine whether crop N utilization for the yield formation process differed by planting methods (Figure 2). The brown rice yield increased almost linearly with N MT , but the intercept was significantly greater for TPR than DDSR, confirming that DDSR had a lower CNUE than TPR (Figure 2a). Similarly, spikelet density had a strong positive relationship with N MT in both TPR and DDSR, while it tended to be lower in DDSR than in TPR for the same N MT (Figure 2b). Grain filling (%) and 1000grain weight in DDSR were not correlated with N MT and were consistently lower than those in TPR (Figure 2c, d).
The 1000-grain weight in DDSR was not modeled by N MT , planting method, and their interaction (R 2 = 0.2982; Figure 2d).

Analysis of the relationships between N uptake and spikelet density
Spikelet density was also significantly correlated with N PF and N HD , and the regression lines were different between TPR and DDSR at both stages (Figure 3, significantly different intercepts; P = 0.0009 with N PF and P = 0.0244 with N HD ). However, there was a clear difference between the two stages, where the regression line for DDSR was higher than that for TPR at PF, and it was reversed at HD (Figure 3). The N uptake for each N treatment, planting method, and cultivar. Note: 0N, CU, NU, and 8N represent the following N treatments: no N treatment, one-time coated urea treatment (12 g N m −2 ), split application of normal urea (15 g N m −2 , 6, 3, 3, 3), and conventional compound fertilizer treatment (8 g N m −2 ), respectively. DDSR and TPR represent the dry direct-seeded rice and transplanted rice, respectively.
The difference in the spikelet-N relationship between planting methods and growth stages could be related to the different crop N uptake patterns. We, therefore, applied a multiple regression analysis to examine the relative contribution of N uptake on spikelet density at different growth stages (Table 4). We found that both N PF and N PF-HD significantly contributed to spikelet density, while the planting method (DDSR or TPR), year (2018, 2019, or 2020), or cultivar (AK or YA) had no significant effects  The relationship between spikelet density and N uptake at N PF or N HD under different planting methods of DDSR and TPR. The planting method difference with N HD was smaller than that with N PF and was reversed. The interaction between N HD and the planting method was not significant. Note: DDSR and TPR represent the dry direct-seeded rice and transplanted rice, respectively. AK and YA represent the cultivar's names, 'Akitakomachi' and 'Yumiazusa', respectively. In our analysis of covariance, the regression lines were different between TPR and DDSR at the PF (left side, P = 0.0009) and HD (right side, P = 0.0244).
( Table 4, Model 1). A simplified model without these categorical variables (Model 2) demonstrated that the coefficient for N PF was nearly twice as large as that of N PF-HD . The differences in crop N uptake patterns and sensitivity of spikelet density to N uptake at different growth stages accounted for much of the variation in spikelet density due to the N application method, planting method, year, and cultivar.

Crop N dynamics and model parameters
Crop N uptake increased almost exponentially until about 100 DAS (PF stage) and linearly until maturity, with noted differences between years (Figure 4 and S5). When crop N was plotted against AETI (Equation 5 and Equation 6), the differences narrowed ( Figure S5), thus confirming that crop N uptake patterns can be well represented by AETI, Figure 4. The relationship between rice N uptake and DAS or AETI from LA5. Note: Graph for YA-CU (Yumiazusa and coated urea) treatment. Supplemental material ( Figure S5) shows the relationships of all treatments. DAS, days after sowing; AETI, accumulated effective thermal index (see material and methods section); PF, panicle formation. N up is the N uptake of the rice plant. N LA5 is the N uptake at fifth leaf age. RNR is the relative N uptake rate. N LA5 and RNR were derived as the intercept and coefficient of the exponential function during from the date of LA5 (5.0 of leaf ages) to the panicle formation date. Table 4. The effects of N uptake on spikelet density at different growth stages (PF and HD), planting methods (DDSR and TPR), cultivars (AK and YA), and years (2018, 2019, and 2020 Note: Model 1 includes categorical variables: planting methods, cultivars, and years, and Model 2 excludes them. DDSR, dry direct-seeded rice; TPR, transplanted rice. AK and YA indicate the cultivar's names, 'Akitakomachi', and 'Yumiazusa', respectively. particularly during the vegetative growth period with the model parameters (N LA5 and RNR).
N LA5 was higher in CU and NU than 0N (Table 5). RNR had a different Y×N effect (P = 0.0052) and tended to be higher in CU. YA had a higher N LA5 than AK (P = 0.0353) while exhibiting similar RNR (P = 0.2046, Table 5).
Model 3 with the Year term outperformed Model 4 without it, as evidenced by higher adjusted R 2 and lower AICc and BIC, confirming the significant Year effect (Table 6). In both Models 3 and 4, RNR and N LA5 had significantly positive effects on N PF (P < 0.01, Table 6), with the standardized partial regression coefficient (β) being consistently higher for N LA5 than for RNR. The β value for N LA5 was three times greater (0.87) than that of RNR (0.29) in Model 4 (Table 6). VIFs in Models 3 and 4 were <4; thus, serious multicollinearity was not suggested between N LA5 and RNR (Table 6).

Discussion
Our three-year field experiment confirmed that DDSR yield was lower than TPR by an average of 11% between AK and YA (Table 2), similar to that reported by Shinoto et al. (2021). Furthermore, the ANUE was lower in the DDSR than in the TPR by 42% (Table 3), in line with the observations that DDSR requires a greater amount of fertilizer to be productive (Otani et al., 2013). In this section, we first explore possible reasons for the low ANUE in DDSR  The standardized partial regression coefficient. compared to TPR. Then, we discuss how N regimes and cultivars affect production efficiency in DDSR cultivation.

Why is the ANUE of DDSR low? (comparison with TPR)
ANUE can be considered a product of fertilizer N recovery and the efficiency of converting crop N into grain yield. We predicted that the former, represented by ANR, was largely responsible for the lower ANUE in DDSR relative to TPR because a large fertilizer loss was expected for the DDSR, particularly in an Andosol with a high percolation rate (Kanmuri et al., 2017). However, our analysis indicated that both ANR and CNUE were lower in the DDSR than in the TPR, both contributing to the low ANUE. The low CNUE of DDSR was mainly induced by two factors. One was spikelet production per unit of N HD or N MT being limited in the DDSR relative to the TPR. As shown in Figure 1 and Table 3, N MT was comparable between DDSR and TPR, but the N uptake pattern was considerably different. While 8N of TPR took up 39% of the total crop N before the PF stage, CU of DDSR was only 29% at PF while accelerating the N uptake toward HD and ultimately surpassing that observed in the TPR at the HD (Table 3 and Figure 1). The multiple regression analysis of spikelet density on N PF and N PF-HD revealed that the contribution of N PF was nearly double that of N PF-HD (Table 4), suggesting that a smaller N PF and larger N PF-HD in DDSR leads to low spikelet production efficiency per unit N uptake. These results indicate that enhancing N PF might be effective in increasing CNUE and ANUE. Alternatively, Model 1 in Table 4 showed that the N PF and N PF-HD relationship was not significantly influenced by the planting method (DDSR or TPR), thus allowing us to show the universality of the relationship between spikelet density and N uptake for rice grown under different planting methods.
Secondly, the unfavorable climatic conditions for DDSR during the grain-filling period are compared with TPR and these conditions are limited source capacity for grain growth. The heading date of DDSR occurred approximately 9-17 days later than that of TPR, resulting in lower solar radiation and air temperature during the grain-filling period for DDSR plants. This is reflected in the poor grain filling and 1000-grain weight at all N levels ( Figure 2). Therefore, optimum growing seasons for DDSR need to be determined, accounting for cultivar phenological properties and site-specific climatic conditions.

How can we improve DDSR productivity? (comparison among treatments on DDSR)
Grain yield was not statistically different between the two fertilized treatments imposed on DSSR but was, on average, 4% lower in NU than CU (Table 2), which could be attributed to N uptake patterns, and N use efficiency in the two N regimes.
The ANR in DSSR was not different between CU and NU. This was unexpected because CU has been known to improve ANR in TPR (Shoji & Kanno, 1994;Wang et al., 2018) and in direct seeded rice in warmer region (Cheng et al., 2022) compared with non-coated N application. Frequent top-dressing of NU in this study might have reduced N loss, leading to comparable ANR of NU to CU, although ANR for DSSR was generally lower than for TPR (Table 3), highlighting the need for improvements regarding N application methods for DDSR. On the other hand, ANUE, CNUE, and CNUEF in the NU treatment were generally lower than in the CU treatment (Table 3) although the difference in CNUEF was not significant. This could be attributed to different N uptake patterns between the two treatments, similarly to the difference between DSSR and TPR discussed above; NU had a lower N PF but greater N PF-HD to reach a similar N uptake at maturity (Table 3). These results suggested that the advantages of coated urea in this study were the efficiency of brown rice production per N uptake rather than N recovery. In other words, normal urea management may be improved by split application tailoring to the plant's N demand, which can help increase CNUE and CNUEF.
N uptake before the PF stage is critical for spikelet production and yield determination of DDSR (Table 4). We analyzed the N uptake pattern on DDSR until PF and confirmed that the two parameters (N LA5 and RNR) of the exponential function of (AETI) depicted well the increase of N uptake during the vegetative period, as reported by Takahashi et al. (1976). Both parameters significantly influenced the N PF, with N LA5 exhibiting a slightly higher effect than RNR (Table 6). The relative importance of N LA5 over RNR was greater in Model 4 without the year term than in Model 3 with it, suggesting that N uptake during the early stage, even N uptake at LA5, contributes to the year-to-year variation in the DSSR yield.
Both N LA5 and RNR could be improved by the N application (Table 5). While CU had a greater effect on RNR than NU, albeit not statistically significant, the effect appeared significantly different between years ( Table 5). The reason for the significant Y×N interaction on RNR remains unclear, but environmental conditions need to be considered to stably enhance RNR through N management. On the other hand, the effect of N application on N LA5 was consistent in both years, with a similar enhancement in CU and NU (Table 5). N application before LA5 enhanced leaf development, shortening the days to the LA5 stage (Table 1). These initial differences persisted and accounted for 60% of the differences in the final leaf number ( Figure S2). The delay in leaf development could limit canopy light interception and biomass growth to secure N PF for spikelet production, leading to lower yields. With the NU treatment, there seems to be room for increasing N LA5 (Table 1). However, normal urea fertilizer at an early stage may be lost for the most part by ammonia volatilization (Griggs et al., 2007;Qi et al., 2012) and denitrification (Casey et al., 2018) with DDSR. Further research is needed to increase N LA5 by NU to the same level as that of CU. Cultivar selection is another option for the improvement of DDSR yields. In our study, YA (high-yielding cultivar) retained its yield advantage under dry direct-seeding practices. The YA brown rice yield was significantly higher than that of the AK (Table 2, P = 0.0083). This could be attributed to the following traits: greater N LA5 (Table 5, P = 0.0353), ANUE (Table 3, P = 0.0023), CNUE (Table 3, P = 0.0012), and more panicles ( Table 2, P = 0.0937), and 1000-grain weight ( Table 2, P = 0.0043). However, YA is slightly later-maturing than AK, resulting in a significantly lower grain-filling percentage and a lower yield in 2019 than that of AK (Table S5). The DDSR in a cool climate may not provide YA with the appropriate conditions to express its high-yielding abilities because the grain-filling is restricted by unfavorable air temperatures or solar radiation, especially at the grain-filling stage. Further research is needed for DDSR yield constraints associated with meteorological conditions in a cool climate.

Conclusions and remaining issues
Our three-year field experiment confirmed that DDSR yield was lower than TPR by an average of 11%. We have revealed three important constraints limiting grain yield and N use efficiency of DDSR in a cool climate: Low fertilizer N recovery, low spikelet production per N input, and poor grain filling as compared with TPR. The low fertilizer N recovery in DSSR was expected because of the high percolation of the paddy field used, but CNUE, defined as yield per unit of N uptake at maturity, was also low, limiting DDSR yields. The low CNUE of DDSR was associated with the crop N uptake pattern; DDSR had a greater crop N at maturity but a smaller N PF than TPR, resulting in a low spikelet production per N input. N management could improve N PF by enhancing early leaf development and relative N uptake rate during the vegetative growth period. The split application of urea yielded almost as much as the coated-urea application, so it can potentially reduce the cost of production because of its lower price. However, some challenges remain to determine the optimal N proportion and application stage in the split application that can increase early leaf development and CNUE. A high-yielding cultivar, YA, has some favorable traits such as better initial growth and greater CNUE than a standard cultivar, AK. To leverage these traits for yield improvements, we need to select suitable phenological traits to ensure favorable climatic conditions during the grain-filling period.