Differences in seasons and rice varieties provide opportunities for improving nitrogen use efficiency and management in irrigated rice in Kenya

Apart from nitrogen (N) rates, N use efficiency (NUE) (yield N/total input N) is affected by seasons, crop developmental stages, and varieties. Knowledge of how these factors affect NUE in rice production in Kenya is limited. Therefore, field experiments were conducted with ‘low rates’ of N (simulating farmers’ practices) of 0, 26, 52 and 78 kg N ha−1 with five varieties (MWUR1, MWUR4, IRAT109, NERICA4 and NERICA10) and higher rates of N (125, 175, and 225 kg N ha−1) simulating researchers’ doses with two lowland varieties (Basmati 370 and BW 196) and IR 72. Another experiment on NUE responses to sites, N rates and dose (split or full dose) was undertaken with the IR97 variety. With the ‘low rate’, yields increased with incremental N rates up to 52 Kg N ha−1 and declined (during cold periods, for some varieties). In this scenario, the N agronomic efficiencies (AEN) declined with increasing N but depended on sites and seasons. However, most AEN values were above 100, implying nutrient mining. In most cases (except at the Mwea site), the N utilization efficiency (NUtE) ranged from 16 to 22kg kg−1 and were not significantly affected by sources and methods of N application. In all cases, an increase in N elicited declining trends in NUtE. Moreover, N uptake efficiency ranged between 22 and 90kg kg−1 without significant variation among varieties. For the ‘high N rates’, high biomass yield resulted in higher grain yields in BW 196 and IR 72 but yield declined beyond 75 kg ha−1 N rates due to poor grain filling, particularly when a cold period coincided with booting and grain filling. We conclude that N rates, doses and rice varieties are key determinants of AEN and NUtE in contrasting rice growing seasons in Kenya. Cropping seasons and rice varieties are therefore potential key determinants of sustainable rice productivity and improved NUE in rice-based systems in the studied regions of Kenya.


Introduction
In sub-Saharan Africa (SSA), rice is the most rapidly expanding food crop for both consumption and commercial purposes (Tsujimoto et al 2019). According to the USDA report (2018), the milled grain yield of rice has tripled from 9.2 to 31.5 Mt between 1990 and 2018, largely due to increased use of synthetic nitrogen (N) fertilizer. It is anticipated that with the current growth in population and rapid urbanization, there will be a need for increased rice production to meet the consumption demands (Njinju et al 2018, Fróna et al 2019. However, rice yields have stagnated in SSA for a considerable period of time (USDA 2018). In addition, the general production is low. For instance, the productivity of upland rice varieties in Kenya is at 1 t ha −1 which is way below the regional potential that is estimated at 7 t ha −1 (Magoti et al 2019). Rice production in Kenya only meets 16% of the total country's demand. The Government of Kenya's Ministry of Agriculture report (2008)(2009)(2010)(2011)(2012)(2013)(2014)(2015)(2016)(2017)(2018) indicated that consumption was increasing at a rate of 12% per annum as compared to the increase in consumption of other cereals. Furthermore, there is a rapid shift in the trend of the modern eating habits among youths, a fact that will further increase the demand of rice. There is therefore a need to reverse the trend and enhance the yield performance of rice by first identifying the bottlenecks and taking informed corrective measures. The major factors contributing towards a lower rice yield in Kenya, and in Africa in general, include, but are not limited to: a lower yield potential (according to Emitiyagoda et al (2010)) (where yield potential is 'the maximum yield that a variety can reach under the best environment') of rice genotypes, a lack of fertilizer inputs, drought and poor water control, low-nutrient soils, poor soil fertility management, a shortage of labour and a high incidence of pests, weeds and diseases (Saito et al 2013, Kajisa 2016. In developed and transition economies like China, East Asia and many European countries, crop, particularly rice, yields have increased by leaps and bounds due to increased reactive N application. Unfortunately, this has environmental costs in terms of pollution through gaseous emissions and leaching into water bodies. The flipside of the coin is too little N application in developing countries because of poor economies that can barely support N fertilizer imports. Therefore, a need to strike a balance in N application for better N use efficiency (NUE) is crucial. In the present study, the NUE is defined as grain yield per unit of the applied N fertilizer, which is a product of Nitrogen Use physiological Efficieny (NUpE) and N utilization efficiency (NUtE) (Rakotoson et al 2017). To improve NUE, it is necessary to enhance N uptake, N utilization efficiency, and the N harvest index, with many complex and interactive environment-crop, as well as physiological, mechanisms and agronomic traits (van Buerenand Struik 2017). In SSA, NUE is estimated to be above 100%, indicating nutrient mining and low productivity (Edmonds et al 2009). However, a large percentage of N losses are unaccounted for, especially in lowland rice which is partly contributed to by surface runoff, volatilization, denitrification and leaching (Fageria and Baligar 2003), hence the need for the adoption of fertilizer-responsive varieties and responsible fertilizer use strategies (Estudillo and Otsuka 2013). Crop characteristics, environmental and management factors interactively influence the NUE of a crop, i.e. crops and crop varieties differ in their ability to acquire N from the soil and respond to incremental reactive N. Furthermore, environmental factors such as temperature, rainfall and irradiance, soil and crop management practices such as nutrient management, crop rotation, cover crop, tillage, weed and pest management and irrigation are also important factors that influence the NUE of a crop (Balasubramanian et al 2004) and equally affect crop yield potential. Many farmers in SSA have no access to adequate N fertilizers and mostly apply less than recommended by both the Abuja and Malabo declarations by African heads of state. It is therefore important to test the NUE with the N rates accessible to these farmers. Indeed, Jian- Feng et al (2011) established that NUE under no, or low levels of, N were very effective, but those of N absorption efficiency and N harvest index (NHI) in rice were noneffective because their genotypic differences must be given full expression under higher N levels. Hence the further need to compare with higher rates.
Most experiments reported in SSA and part of Kenya have relied on data extractions from different databases such as FAOSTAT, IFASTAT, GFM (Lassaletta et al 2014, Zhou et al 2014, Elrys et al 2019 to estimate the NUE with less focus on specific crops and factors other than N that have a direct bearing on NUE. Therefore, in this study, we looked at NUE at a plot scale level under controlled research conditions and compared two sites as well as different varieties, in two seasons, under varying fertilizer regimes. To our knowledge, there is no other kind of study that has previously been undertaken in Kenya and the general Lake Basin region. According to Dobermann (2007), the NUE achieved in research trials may serve as a reasonable indicator of what can be targeted with good management, but this may only apply to short term experiments such as at the Ahero site, which closely simulates a farmer's field. However, for the Mwea site (a research station) where long term experiments are continuously conducted, caution in the interpretation of results is necessary to avoid overestimation of the values of control plots. We hypothesized that yields of rice are not controlled by N rates alone, instead additional factors like varieties and seasons are paramount. Therefore, this study aimed to evaluate the effects of N rates on selected local and improved varieties in two contrasting seasons and sites, and understand their N use efficiencies to form a reference for policy formulation on N management. Indeed, rice varieties, particularly landraces adapted to the local environments and selected by the farmers for their better yield, under low or zero inputs, form an interesting genetic material for the identification of donors and genomic regions for better NUE. Landraces, given their past evolutionary history and adaptation to stress environments, often outyield modern cultivars under low-input production systems, as demonstrated by Flint-Garcia et al (2003). We further hypothesized that discriminative traits related to NUE better express themselves under low input than under high input. However, testing under both low and high input can yield cultivars that are adapted to low-input conditions but also respond to high-input conditions (van Bueren and Struik 2017).

Study site and experimental design
The experiments were conducted in two sites; Kirogo farm in Mwea in Central Kenya and the Ahero irrigation scheme in Western Kenya (supplementary figure 1 (available online at stacks.iop.org/ERL/16/ 075003/mmedia)). Kirogo farm is located at the Kenya Agricultural and Livestock Research Organization (KALRO-Mwea), at 0 • 37 ′ S, 37 • 20 ′ E, about 1159 m above sea level. Rainfall ranges from 500 to 850 mm per annum in a bimodal pattern with long rains (March-June) and short rains (October-December). The temperatures for this area range between 15.4 • C and 29.5 • C. The soils are nitosols, deep, well-drained dusky-red to dark reddish-brown, friable clay with low fertility (Njinju et al 2018). The Ahero irrigation scheme is located in the middle of the Kano Plain, 25 km Southeast of Kisumu town at 0 • 08 ′ S, 34 • 58 ′ E, about 1168 m above sea level. Rainfall ranges from 1200 to 1735 mm in a bimodal pattern, long rains (March-May) and short rains (October-December). In this region, the average daily temperatures range from 15 • C to 25 • C.

Experiment 1-effect of fertilizer application method on rice yields and NUEs
The experiments were conducted at the Mwea and Ahero sites. The experiments were laid out in a randomized complete block design (RCBD) in factorial arrangements. Factor one comprised two forms of N fertilizer, sulphate of ammonia (SA) and urea at various levels (0, 25, and 50 kg ha −1 ). Factor two comprised two methods of applications: full dose and two splits. The treatments were replicated three times. In this experiment, the plot sizes were 3 × 4 m. The rice variety used was IR97 (obtained from the National Irrigation Board (NIB) of Kenya). Threeweek-old seedlings were transplanted at a spacing of 25 × 15 cm in 36 plots, each plot had a total of 320 plants (table 1). IR97 is nonaromatic and is the only one that can grow in Ahero due to attacks from birds.  BW 196 and IR 72) constituted the second factor. The N fertilizer application rates were calculated by subtracting the basal application from the top-dressing rates. Treatments were replicated three times. Trials were conducted between February-August 2014 and January-June 2015.

Experiment 3-effect of lower rates of N fertilizer application on rice yields and NUEs
The 'low N' experiment was laid out in RCBD with a factorial arrangement at the Mwea site. Factor one was rice varieties (IRAT109, MWUR1, MWUR4, NERICA4 and NERICA10) in five levels while factor two treatment was N rates (0, 26, 52 and 78 kg N ha −1 ) (four levels). In this experiment, seeds were directly sown at a spacing of 20 × 15 cm by dibbling. Each plot had five rows. Calcium ammonium nitrate fertilizer was randomized in each block and a blanket application of zinc sulphate was applied at a rate of 25 kg ha −1 . The fertilizer treatments were carried out in two equal splits with the first split being applied at the tillering stage (21 d after sowing), and the second split applied after 45 d (at the panicle initiation stage). Treatments were replicated three times. This was carried out for two seasons, May-October 2015 and October 2015-April 2016. The water was supplied by irrigation throughout most of the growing periods at the Mwea site. Being a research site, water was not limited, unlike in Ahero where there was regulation of water through the constant opening and closing of canals followed by redirecting to the necessary farms. In both sites the water was supplied by NIB.

Data collection and analysis 2.2.1. Plant harvesting
Harvesting was done at about 40-50 d after heading when 80% of the grain had reached the hard dough stage. A net plot of 1.00 m 2 was harvested at the middle of each plot comprising of about 25 hills, leaving out the border rows.

Biomass collection, weighing and analysis
The straw from the harvested plants was oven-dried for 24 h at 72 • C (until constant weight). The aboveground biomass was determined by weighing the oven-dried straw of the samples.

Yield and yield components
The harvested panicles were dried, threshed and the weight of the total grain weight obtained after drying to 13% moisture content. Filled grains and unfilled grains were also weighed separately. Total grain yield in tonnes per hectare was estimated using the following formula:

Plant tissue analysis
Plants for data collection were selected and marked randomly. Within the five rows in a plot, the outer rows were not considered while sampling. Leaves and stems from the sampled plants were oven-dried at 70 • C for 24 h, then ground to powder using a Heiko vibrating mill. The N content (%) was separately analysed from the straw and seeds in each treatment for No.
Nitrogen use efficiency indicator Formula Apparent N recovery rate (ANR): The net increased total N uptake by the plant N fertilizer application against the total amount of N fertilizer applied Agronomic N efficiency (NpUE): Net increased yield against the net increased N uptake with N fertilizer application all genotypes after harvest. The N content from plant tissue was obtained using the Kjeldahl method using standard procedure (Horwitz 2010).

Soil analysis
Soils were randomly sampled at 0-20 cm depth using an auger, mixed thoroughly and subsampled. The subsamples were air-dried, ground, and sieved using a 2.0 mm sieve. The samples were thereafter kept in a well-labelled plastic container for further analysis. The soil pH (water) was determined using a pH meter (ASTM 1995). Organic matter was determined by a wet oxidation/chromic acid digestion method (Walkey and Black 1934). The total N in the soil was determined using the Kjeldahl method (Bremmer 1960) (supplementary table 1).

Meteorological data
Data on temperatures, rainfall and solar radiation were obtained from a weather data station at Kirogo farm in Mwea, and where gaps existed, they were filled by data accessed and extracted using the 'Data Download' web tool (Hegewisch and Abatzoglou 2020) (supplementary figures 2 and 3).

Nitrogen use efficiency
Biomass and grain yield results were used to estimate the NUE characteristics using the following modified formulas described by Good et al (2004), Xu et al (2012) and table 2.

Data analysis
Data on the yield characteristics and NUE indicators were organized in excel spreadsheets and subjected to various statistical procedures (analysis of variance, Welch two-sample t-test, and Pearson correlations) using R software version 3.4.4.

Varietal responses to N forms and doses on rice grain yield and yield components in two sites
There were no significant statistical differences amongst different N application methods (p > 0.05) for sulphate of ammonium (whether split or full dose) at different rates (figure 1). A similar trend was observed with (UR) under the split application. The application of UR at 50 kg ha −1 under full dosage elicited the highest grain yields compared to the control. In addition, the application of 25 kg ha −1 in two split doses at the Mwea site resulted in higher grain yield (p ⩽ 0.05) compared to the control and this was evident irrespective of the N sources (figures 2(C) and (D)). However, significant differences (p ⩽ 0.05) were only observed between the highest rate (50 kg ha −1 N) and the control for urea at full dose (figure 2(B)).

Nitrogen utilization, physiological use efficiencies, apparent recovery and agronomic efficiency
NUtE was not influenced by the method of N application (whether split or full dose) but showed a decrease with increase in levels of N fertilizers (table 3). These differences were not significant (p > 0.05). Unfertilized plots had the highest NUtE of 23.47 followed by ammonium sulphate at 25 Kg N ha −1 (20.29) when applied in split applications and the lowest NUtE was observed in urea (50 Kg N ha −1 ) in split applications with a value of 17.31 at the Ahero site (supplementary  table 2). In Mwea, a similar trend was observed with NUtE where higher N levels elicited the lowest values, while the control (unfertilized)-to-low N application had higher NUtE when applied in split methods. N physiological efficiency (NpUE) exceeded 100% in some treatments under urea (25 Kg N ha −1 ) in Ahero when applied in full dose (220.76) while in Mwea the same rate applied in splits resulted in 111.03 NpUE.
In Ahero, there were no significant differences among AE N values (p > 0.05), with values ranging from 7.90 to 45.02 (table 3), without clear trends in N rates and methods of application (full dose or split). At the Mwea site, the values were three-fold to five-fold higher than at the Ahero site. As compared to full dose, all the split applications had values of more than 100, with the highest values registered with 25 kg ha −1 N (irrespective of N source). The ANR showed variations with N application methods and rates at both the Ahero and Mwea sites. The ANR values ranged from 0.42 to 2.2 at the Ahero site, while the values at Mwea were from 3.41 to 8.17. At the Ahero site, the interactive effects of N rates and dosage (split/full) had no significant effects on the NUE components. However, at Mwea, the N rates had a significant effect on NUtE, ANR, AE N and NHI (Nr * * * ) but no effect on NpUE (Nr ns ). Furthermore, dosage (split/full) did not affect all these NUE components (Do ns ) (table 3).

Nitrogen use efficiency components in two seasons with different varieties and N rates
The treatments with lower rates (0-78 kg N ha −1 ) showed varietal differences in NUtE (p ⩽ 0.05) with the control (unfertilized plots) having the highest NUtE in both seasons. There was a decreasing trend of NUtE with increase in N rates. The varieties MWUR1 and MWUR4 showed the lowest NUtE values in both seasons under highest N supply (supplementary table 3). The NpUE values had wide variations with some varieties exhibiting negative NUE values at different N rates of application during season 1. At 78 kg N ha −1, MUR1 (season 1) had the extreme value of NpUE of −289.04 which could indicate high loss of applied N to the environment or a high extent of soil mining that degrades the soil quality, or these values were just outliers. NER-ICA 4 in unfertilized plots exhibited values close to the optimal NpUE of 61.05. In season 2 there was a clear declining trend in NpUE with the increase in N rates and this was evident in all varieties. Compared to varieties, the effects of N rates on NpUE values were more pronounced in both seasons. Moreover, The variations in yields in the two seasons were not consistent among the varieties and N rates, probably due to several outliers and poor mean distributions ( figure 3 and table 4). On the contrary, trials with higher fertilizer rates (75-225 kg ha −1 ) showed yield differences between seasons under different fertilization regimes and were also affected by varieties ( figure 4 and supplementary table 4). The varieties BW 196 and IR 172 had significantly lower yields (p ⩽ 0.05) in season 2 compared to season 1 under different fertilizer rates (figures 4(B(iii)) and (B(iv))). The booting and initiation of flowering coincided with the warmer part of the growing period during season 2 for the Basmati 370 variety, while for BW 196 and IR 72 varieties, these critical developmental stages coincided with warmer parts of season 1, which partially explains the patterns of grain yield reported in figure 5 (see supplementary figures 2 and 3).

Regression analysis of yield and nitrogen rates in different seasons
In season 1 (supplementary figures 4(a(i))-(e(i))) there were poor relationships between fertilizer rates and yields (grain and total biomass) and this was irrespective of varieties (R 2 values ranged from 0.11 to 0.42). Under high N fertilization, the relationship between yields and N rates showed exceptionally low R 2 values in season 1 (ranging between 0.1 and 0.5) depending on variety (supplementary figures 5(a)-(c)). In season 1, the biomass yield showed an increasing trend with an increase in N levels, while the grain yield dropped with an increase in N (starting after 75 kg ha −1 ).The relationship between yield and N rates for BW 196 was R 2 = 0.50 and R 2 = 0.15 for total biomass and grain yields in season one and two, respectively. In season two, the BW 196 variety showed an increase in both grain and biomass yield but showed a plateau at 150 kg ha −1 N. In both seasons, there was a decrease in filled grain as the N rate increased, but this was pronounced in the second season (supplementary figure 6). This has implications on N utilization and general NUE.

Correlation of yield parameters and nitrogen use efficiency characteristic for different rice varieties at low and high nitrogen levels
In all the five varieties (IRAT109, MWUR1, MWUR4, NERICA4 and NERICA10), there were positive correlations between N rates and grain yields, total biomass, harvest index, ANR and AE N , while the correlation of N rates on NU t E and NPUE were negative. However, in terms of NHI, some varieties had positive correlations (IRAT109, MWUR4 and NER-ICA10) with N rates while others had negative correlations (MWUR1 and NERICA4) (supplementary table 5). This trend was similar in season 1 (data not shown). In season one, both Basmati 370 and IR 72 had negative correlations between the N rates and grain yield, harvest index and NHI, while the correlations of N rates and total biomass and N uptake were positive in these varieties. On the contrary, BW 196 registered a negative correlation for N rates with grain yield and harvest index. However, this variety had a negative correlation of N rates with the harvest index and NHI, while the correlation of N rates to total biomass and N uptake were positive. In season two, both Basmati 370 and IR 72 showed negative correlations for the N rates with grain yield, harvest index and NHI. The correlation of N rates to total biomass and N uptake were also negative in these varieties. Just like the other varieties, BW 196 registered a negative correlation on N rates with grain yield and harvest index. However, this variety had a positive correlation of N rates with the NHI, while the correlation of N rates to total biomass and N uptake was negative (supplementary tables 5-7).

Grain yield and yield components as affected by nitrogen rates and forms in different rice varieties
The optimal management of N is critical for high yields in rice and there are recommendations for three stages of application, including 40% before transplanting followed by two 30% doses at the panicle initiation and grain filling stages, respectively (Olfati et al 2012). The split N applications undertaken in our experiment considered these key rice phenological stages that differentially coincide with these critical rice N demands (supplementary figures 2 and 3). The split application was more significant at Mwea than at the Ahero site. In Mwea, the interaction between split application and fertilizer sources (urea or sulphate of ammonium) were significantly high, though the advantage of splitting was more pronounced for urea. Urea has been reported to have a lower fertilizer use efficiency compared with other N sources (Zaman et al 2008) since it is highly susceptible to loss pathways, particularly with NH 3 emissions as soon as it is applied to croplands, especially under warm climatic conditions like at the Ahero site, hence the advantage of split application. Interestingly, the rice yield for the control treatment (unfertilized plots) was much higher (three folds) for the Mwea site than those of Ahero. Being a research centre, the residual N (both organic and inorganic forms) is expected to be higher. Notably, researchers (Liu et al 2004, Feng et al 2006 had previously postulated luxury consumption of N, a probable nexus between higher N concentration in rice straw and high indigenous N supply of soil (INS).
In the current study, the average grain yield in the control (unfertilized) plots at Mwea was higher than 3.5 t ha −1 , while at Ahero it was 1.2 t ha −1 , and the lower grain yield at the Ahero site may be attributed to the lower yield potential and N uptake efficiency of the rice genotypes under very hot conditions, probably with poor pollen fertilization. The splitting of N inputs throughout the growing season of rice crops contributes to the maintenance of synchrony between the crop N demand and N availability that leads to the harvesting of a considerable grain yield (Kumar et al 2018). According to Arthanari et al (2007), rice crops require a sustained supply of N in the soil until the reproductive stages to enhance maximum yields. A lack of adequate N is a major constraint among many local farmers, especially in Kenya (unless receiving fertilizer subsidies or getting supplies from cooperative societies). As reported by Li et al (2008), N in soils at the heading stage helps plants to have higher photosynthetic rates, delayed leaf senescence and significantly increases grain filling. In another study done by Sheng et al (2002), the strategy of stressed panicle fertilizer application including application of 30% at tillering and 70% at panicle initiation without any prior application of basal fertilizer, resulted in improved NUE and high grain yield due to maintenance of the soil N supply. The rice crop is known to have variable demand of N at major critical growth stages and this may explain the slight yield advantage at Mwea under split application (supplementary table 2).
The lower grain production with increased fertilizer application in different seasons had been reported earlier at the Mwea irrigations scheme by Njinju et al (2018). Their results showed the negative impact of increasing N on grain filling and HI in Basmati 370 and BW 196 varieties, especially when N rates were above 75 kg ha −1 . Their work concentrated only on N with respect to yield, but did not relate to the implication of declining yield at higher N to NUE and N management. Surprisingly, in the current study, the Basmati 370 variety yielded higher than the two varieties (BW 196 and IR 72), particularly in the second season ( figure 5). The trends were, however, inconsistent with the first season where BW 196 and IR 72 had higher yields. Basmati's booting and flower initiation coincided with a warmer period as compared to the two varieties during the second season, while the opposite occurred in season 1 (figure 4see supplementary figures 2 and 3). Whereas the differences in temperatures may appear marginal, the cumulative degree days are substantial, particularly towards the reproductive stages of growth. Such a marginal role of temperatures on the yield of different rice varieties have been reported by Yang et al (2019), albeit with warm temperatures. At the Mwea site, the increase in grain yield with N showed stagnation at some point but was more prominent with higher fertilizer rates (supplementary figure 6). The plateauing/levelling off of yield with N application is not unique to N and rice production. Typically, crop yields and nutrient uptake rates have been reported to show gradual decline with additional nutrient supply (reactive N in our case), reaching a ceiling and eventually declining with further addition (supplementary figures 4 and 5). The level of this ceiling is determined by the environment-genetic yield potential. At low levels of nutrient supply, rates of increase in yield and nutrient (N) uptake are large because the nutrient of interest is the primary factor limiting growth-Liebig's law of minimum (Mengel and Kirkby 2001). This could imply poor utilization of N by the crop at the higher supply of N (see supplementary tables 9 and 10) especially during a window of cold season during the rice booting stage, due to curvilinear return to the conversion of plant N to grain as the yield approaches the ceiling at higher levels (Djaman et al 2016). For the BW 196 and IR 72 varieties, N application was optimal at around 75 kg ha −1 in the second season and additional N beyond this point led to a decrease in grain yields and possible loss to the environment. These findings indicate that the current N fertilizer amount (75 kg N ha −1 ) is sufficient for IR 72 and BW 196 in the second season at the Mwea site and implies that the additional N is not efficiently utilized but lost through leaching, emissions or volatilization or any other pathway into water bodies or the atmosphere (Bijay- Singh et al 1995, Lassaletta et al 2014. The maturity period is therefore a crucial varietal 'cold-escaping' determinant, especially at the critical stage of growth, hence higher yields. Unfortunately, higher N rates have the potential to extend the vegetative stages of certain rice varieties; with the possibility of critical biological stages coinciding with a cold (or even a hot) period, hence poor N uptake and utilization with consequent low NUE and possible N losses. This information is not only important for economic reasons but also important to avoid environmental hazards associated with the over-application of N fertilizers as this also affects NUE as propounded by Lassaletta et al (2014). The synchrony of the crop planting dates and N rates is therefore important in planning for proper N utilization and improved yield, especially during the off-season when water is plentiful but coincides with a cold period (Samejima et al 2020). This could probably be achieved through transplanting rather than direct planting to shorten the period the plant is in the field, with the hope of escaping the cold period.
It is not only low temperatures that affect rice production. High temperatures during some parts of season one and some sections of season two (at the Mwea site) could have affected the process of grain yield, resulting in low R 2 values. This is concurrent with the argument by Shi et al (2017) who demonstrated that high temperatures impair grain filling, leading to poor seed set and reduced seed grain weight. At high temperatures, there is a higher possibilities of N loss through volatilization hence reducing the N available for enhanced grain and biomass formation. As reported by Lai et al (2019), higher soil temperatures may trigger processes of denitrification and lead to high emission of nitrous oxide to the environment. Therefore, with the current study, a high rate of fertilization could have resulted in more environmental loss and less yield due to variations in temperatures (either soil and/or ambient).

Variation in yield and N uptake among different varieties
There were clear differences in yield responses of different rice varieties when supplied with different N rates (figures 1, 4 and 5). This was not surprising since N levels have been previously reported to affect the yield of many rice varieties (Flint-Garcia et al 2003, Jahan et al 2020. Yoshida et al (2006) reported a greater production of spikelet, possibly due to less translocation of photoassimilates from leaves to spikelet under increased N rates, instead of being channeled to the grains. Their results agree with our current work where more N led to more grains but after some level, there was a higher percentage of unfilled grains and less of filled grains (having reached the yield plateau) especially for moderate and high N rates at the Mwea site (supplementary figure 6). Varietal differences in yields and N rates had previously been reported by Matsunami et al (2013) in their previous experiment with 70 rice varieties. These results seem to concur with our results and are also in agreement with those of Gweyi-Onyango (2018). Moreover, Gewaily et al (2018) reported differential responses of varities with N rates, with the optimal N rate being observed at 220 kg N ha −1 for late maturing varieties as compared to 165 kg ha −1 as optimal for early maturing varieties. In their experiment, the early maturing variety had a lower yield than the late maturing varieties. In our current study, the yield declined with fertilizer rates beyond 125 kg ha −1 for the 'high rate' treatments. Unlike our case where the cold period seemed to limit yield (despite the high N uptake) (figure 4, table 5) the warm temperature in Egypt did not seem to limit the yield of rice in the study by Gewaily et al (2018). However, Yang et al (2019) reported the negative effect of temperatures, depending on the growth duration of the genotype in question. The N uptake also varied among varieties, with Basmati having a higher N uptake during the second season, with uptake increasing with N rates (reaching an optimal value at 175 kg ha −1 andleveling off thereafter). However, in the first season, BW 196 was the most superior variety in terms of N uptake. BW 196 and Basmati showed a similar response to N rates, N uptake and grain yields. However, the low yielding variety (IR 72) showed consistently lower yields, which was also reflected in low N uptake in both seasons (figure 4 and table 5). The higher uptake rates coincided with higher grain yields during the season when a particular variety had the highest grain yield (BW 196 during the first season and Basmati during the second season). Our current findings are in agreement with other findings which showed that rice genotypes show significant variability for N uptake (external efficiency) and N utilization (internal efficiency) with yield being predominantly determined by the N uptake process (Singh et al 1998, Witcombe et al 2008. N uptake is key in the determination of NUE, hence the choice of varieties may be a key consideration in the proper management of N.

Nitrogen utilization, physiological use efficiencies, apparent recovery and agronomic efficiency
The values of NUtE in this study (table 4) were relatively lower compared to what was reported by De-Xi et al (2007) that ranged from 30.9% to 45.9%, which can be attributed to low soil N status and insufficiency of the applied N inputs. The values reduced as N levels increased, with 78 kg N ha −1 eliciting the lowest values. Low NUE values with an increase in N were reported by Elrys et al (2020). They calculated five scenarios and recommended S2 (EqDamount that was able to achieve self-sufficiency) which was 77 kg N ha −1 per year. Therefore the 78 kg N ha −1 in our study is only for one season, meaning this is double the amount. But other than 77 kg N ha −1 , they proposed the need for the application of additional organic N (37 kg animal manure and 11 kg BNF). The integration of inorganic and organic N may therefore be more sustainable than our approach where we only looked at inorganic N. However, the findings of Elrys et al (2020) were general in terms of regions and crops. In most rice farming, intensification is practiced and is one of the high valued crops that is supplied with a higher amount of N, hence it is possible to supply N rates beyond the 77 kg N ha −1 per year projection. The findings of the current study can be used to develop appropriate N management strategies for a rice-based cropping system in Kenya to prevent further degradation of the soil quality status that can be achieved through advocating for the application of the right source of N fertilizers applied at the right time. To achieve optimal NUE, plants should have enhanced uptake of N from the soils and use it more efficiently in yield production (Ju et al 2015). Our results agree with those of Eagle et al (2000) who reported extremely low N utilization efficiencies in rice that were probably influenced by the extreme losses, particularly in flooded environments without proper water regulations. The study also confirmed that with an increase in soil N supply, a decrease in utilization efficiency has an implication on N management. With high availability of N, additional N uptake is partitioned to the straw and less is made available to the grain, hence resulting in low efficiencies. Contrary to this, with low N, especially in unfertilized plots (as evident from this study) (table 4 season 1-Nericas 4 and 10 as well as IRAT109), the little available N is distributed to the grain, contributing to high utilization and high NUE ratios (Eagle et al 2000). According to Xu et al (2016), fertilizer application should be done judiciously; considering the 4 R principles (right rate, right time, right place, and right form) to ensure the applied N sufficiently meets the specific crop needs. Due to existing challenges among the rural farmers, low N inputs and NUE remain a grey area requiring more research. Therefore, the focus on strategies and technologies for NUE improvement among local farmers remains a key area of concern in light of environmental management, since this has the potential to reduce N loss.
Our results on AE N (tables 3 and 4) varied with varieties as well as locations. These findings agree with those of Djaman et al (2016) who reported AE N values ranging from 26% to 105% depending on rice varieties and locations. However, the current findings are contrary to the results of Thind et al (2018) who reported a higher AE N and ANR when fertilizers were applied in three splits in rice fields. These differences could have been influenced by test varieties, stated soil physiochemical properties like bulk density, structure, texture, and organic matter, as they effect the NUE components (Koudjega et al 2019). Besides, our low efficiencies could be due to the method used to apply the fertilizer, either in split or in full doses, as clearly demonstrated by Baral et al (2020), where the deep placement of urea in lowland rice was effective in promoting plant N uptake, reducing loss and consequently resulting in higher ANR and AE N values. This is not surprising since previous reports show that the recovery of N in soil-plant systems seldom go above 50%, which means that the N applied gets lost through different pathways (Abbasi et al 2003). This increasing gap between the N applied and taken up by crops has been attributed to a monumental reduction in NUE, as reported by the excellent publication of Raza et al (2018) in Pakistan. Moreover, Jing et al (2007) argued that N fertilizer recovery is not only dependent on crop growth and genotype, but rather on management practices like the adoption of 4 R stewardship in nutrient management and the surrounding environmental conditions that have effects on the volatilization and leaching losses. The 4 R stewardship 'implies applying the right source of fertilizer (N for our case), at the right rate, at the right time, in the right place' (Xu et al 2016, Ahmad et al 2018. For fertilizer NUE, there is the need to create more awareness to match N supply as close as possible with the demands of the specific crops and environments; facts that are also in concurrence with the findings of Yang et al (2019) who looked at the effects of warmer temperatures on rice growth and yields. According to Kombali et al (2016), fertilizer N recovery can be optimized through the application of frequent smaller doses.
From our findings, rice varieties respond differently to N in different seasons, and these observations are in agreement with other reports that point out the influence of N form, timing, crop variety N needs and the environment on NUE (Djaman et al 2018, Gweyi-Onyango 2018, Ntinyari andGweyi-Onyango 2018), hence these factors play pivotal roles in determining yield and NUE and are linked with other associated factors that may affect N management. The fertilizer subsidy can increase availability and access (Marenya and Barrett 2009) at the right time, which has a direct effect on yield, as demonstrated by the excellent reviews of Masso et al (2017). Besides, there have been reported cases of fertilizer adulteration and inoculants (Jefwa et al 2014) and this definitely affects crop growth performance and consequently introduces errors in the estimation of NUE with the potential of affecting the management and policy advice at local and regional scales. Indeed, previously suggested interventions in SSA, such as improved seeds, balanced fertilisation, organic inputs, liming materials, and water management (e.g. Ahero site) have the potential to double NUE, and particularly AE N , when novel agronomic practices are adopted, as argued by Vanlauwe et al (2015).

Conclusions
This is the first study to have attempted to evaluate the NUE from the perspectives of rice varieties and seasons in Kenya. N fertilizer management is very important for improving NUE, crop productivity and the reduction of N losses to the environment. Improving NUE to a desirable range remains critical as high values indicate that soil mining is contributing to the declining soil N status and fertility levels. In the current study, cropping season (warm vs cool), rice verities, N application methods, rates and source had a significant impact on the NUE and its components. Results indicated that an interaction of cropping season and rice varieties is crucial in maintaining rice productivity and NUE in a rice-based cropping system in the study regions of Kenya. Limited or low N conditions provide a better opportunity for selecting for NUE, particularly for double cropping seasons where the cold period becomes a challenge. Whereas it is implicit that most rice varieties are bred for higher N responses, we conclude that most of the N applied beyond 125 kg N ha −1 is not fully utilized by the plant (unless there is integration between inorganic and organic N inputs) and hence may be an environmental threat and have cost implications for the farmers. The findings of the current study can therefore be used to develop appropriate N management strategies for the rice-based cropping system. However, additional studies under various N management practices using similar cultivars and N rates in different seasons and at different sites in Kenya would give better insights.

Data availability statement
Option #1: the data that support the findings of this study are openly available at the following URL/DOI: [insert web link or DOI to the data].
All data that support the findings of this study are included within the article (and any supplementary files).