Assessment and Application of EPIC in Simulating Upland Rice Productivity, Soil Water, and Nitrogen Dynamics under Different Nitrogen Applications and Planting Windows

Abstract

A suitable nitrogen (N) application rate (NAR) and ideal planting period could improve upland rice productivity, enhance the soil water utilization, and reduce N losses. This study was conducted for the assessment and application of the EPIC model to simulate upland rice productivity, soil water, and N dynamics under different NARs and planting windows (PWs). The nitrogen treatments were 30 (N30), 60 (N60), and 90 (N90) kg N ha⁻¹ with a control (no N applied −N0). Planting was performed as early (PW1), moderately delayed (PW2), and delayed (PW3) between September and December of each growing season. The NAR and PW impacted upland rice productivity and the EPIC model predicted grain yield, aboveground biomass, and harvest index for all NARs in all PWs with a normalized good–excellent root mean square error (RMSEn) of 7.4–9.4%, 9.9–12.2%, and 2.3–12.4% and d-index range of 0.90–0.98, 0.87–0.94, and 0.89–0.91 for the grain yield, aboveground biomass, and harvest index, respectively. For grain and total plant N uptake, RMSEn ranged fair to excellent with values ranging from 10.3 to 22.8% and from 6.9 to 28.1%, and a d-index of 0.87–0.97 and 0.73–0.99, respectively. Evapotranspiration was slightly underestimated for all NARs at all PWs in both seasons with excellent RMSEn ranging from 2.0 to 3.1% and a d-index ranging from 0.65 to 0.97. A comparison of N and water balance components indicated that PW was the major factor impacting N and water losses as compared to NAR. There was a good agreement between simulated and observed soil water contents, and the model was able to estimate fluctuations in soil water contents. An adjustment in the planting window would be necessary for improved upland rice productivity, enhanced N, and soil water utilization to reduce N and soil water losses. Our results indicated that a well-calibrated EPIC model has the potential to identify suitable N and seasonal planting management options.

1. Introduction

Rice is an important staple crop, and it is a major contributor to food security worldwide [1]. Thailand is the sixth major rice-producing county [1], where upland rice is grown by small-scale farmers. Upland rice is cultivated in the northern and southern parts of Thailand. It is grown in the rainy season in southern Thailand [2] that lasts from May to October. Farmers usually grow upland rice as a sole crop, or an intercrop with young rubber plants and oil palms, as upland rice was identified as the most suitable crop for intercropping by the Rice Department and Rubber Department of Thailand [3,4]. Farmers utilize the available land in rubber and other tree plantations for growing upland rice, and upon the growth of trees, it is shifted to new areas. Due to the low production potential and land use change, there is less research and a lack of scientific evidence on optimal management for upland rice production. Important aspects of research involve agronomic management and varietal improvements to climate impact assessments. Although fertilizer management and planting time are among the easiest to adjust by rice farmers, limited and outdated information used by farmers contributes to practice range of excessive nitrogen (N) fertilization under wide planting windows (PWs), as they are concerned with obtaining a higher grain productivity. The inappropriate management of N fertilization as well as inadequate PWs can as a result contribute to yield losses, higher input costs and higher N losses in rice fields, causing environmental impacts and pollution in water bodies.

Nitrogen fertilization in rice production is highly important, as a deficiency in N in rice plants significantly impacts rice productivity [5]. Variable N application rates (NARs) have been recommended depending upon the rice type (34–69 kg N ha⁻¹) [6] and soil fertility status (48.75–82.5 kg N ha⁻¹) [7]. However, a range of 10–90 kg N ha⁻¹ NAR has been observed in southern Thailand [2,8,9,10,11]. A low N input results in yield losses, whereas a higher input results in lodging, increased insect pest infestation, higher N losses in the environment, and economic losses. The application of nitrogenous fertilizers is critical for soil fertility and rice productivity; however, excessive chemical fertilizer application in rice production is a major source of N loss in the environment [12]. Urea is a major source of N application, which is highly volatile, and urea–nitrogen use efficiency (NUE) seldom exceeds 50% [13]. According to an estimate, rice crops only recover 20–40% of applied N [14]. The fate of the remaining N is still unknown due to the complexity of the forms and pathways of N losses and the complex interactions between the crop, soil, and environment. Rice productivity is highly variable under the influence of the planting period [15] as climatic conditions change over time. Too early or delayed planted rice becomes susceptible to drought periods and higher temperatures or heavy rainfall intervals at critical crop stages. Rainfall is a critical factor that influences water availability for rice production planted at different periods. High water availability under heavy rainfall intervals triggers higher runoff and deep percolation losses. Deep percolation reduces crop water use and increases groundwater pollution due to agrochemicals in percolating water [16]. Water loss is significantly correlated with nutrient losses [17], whereas water input not only influences water productivity, but also impacts nutrient losses [18]. The excessive percolation of water also facilitates N leaching [19]. A study found that 49.7–52% of applied N leached below the 40 cm (40–100 cm) soil layer, while 48–50.3% of applied N was accumulated in the top 40 cm soil layer [20]. This indicated that heavy rainfall intervals triggered N loss in runoff and deep percolation and N movement in deep soil layers. In addition, a high N application can then result in higher N losses [21] due to the higher amount of N present in the rootzone. Conversely, the low water availability under inappropriate PWs reduces the N uptake and utilization by plants, and is usually accompanied by drought intervals. Due to the impact of climate change, variability in seasonal rainfalls increases and drought occurrence is also predicted to be severe [22]. In this scenario, an ideal PW could be a useful management option for upland rice production attained through a high input resource utilization and reduced water and N losses.

Enhanced and sustainable rice production can be achieved by understanding the factors that influence crop performance. Various techniques, such as statistical methods and the simulation of crop, soil, and environment interactions, have been used in determining various management options and in exploring the efficiency of rice production systems. Already developed crop simulation models include those that explain the conversion process of water, N and carbon balances to predict the growth, water use and nutrient uptake, the crop yields and other crop, soil, and environment components [23]. Numerous crop models have been used for a range of applications, such as the determination of optimal planting dates and N fertilization [24,25], variations in cultivar response to the environment and drought [26,27], soil water movement and N dynamics [28], and assessing the impacts of climate change [29]. The environmental policy-integrated climate (EPIC) model [30] is one of various widely used models that can simulate interactions of soil water, plant nutrient dynamics and carbon cycling in response to agricultural management practices and intercropping [31]. Globally, this model has been successfully applied in simulating cultivar performance [32], crop responses to applied nutrients and water management strategies [33,34], irrigation scheduling and management [35,36], climate change impacts on yield [37,38], and soil erosion [39]. To date, information on the field-scale application of the EPIC model to determine seasonal planting and fertilization management for upland rice has not yet been explored. Hence, the EPIC modeling approach for simulating soil water and N balance for determining management strategies could be a potential alternative to improving the productivity of upland rice systems. Therefore, this study was conducted to assess the performance and potential application of the EPIC model to simulate upland rice performance, soil water, and N balance components under four N applications and three planting windows.

2. Materials and Methods

2.1. Field Experiments

An experimental field of the Faculty of Natural Resources, Prince of Songkla University, in Songkhla province of Southern Thailand (7°00′14.5″ N, 100°30′14.7″ E) was selected for the current study during the rice growing seasons of 2018 and 2019. The climatic conditions of the study area were highly variable, with an annual average temperature of 27.9 °C and an average annual rainfall of 2066.7 mm [40]. An aromatic and nonglutinous upland rice cultivar, Dawk Pa–yawm, was used, commonly grown in upland rice-growing regions in Thailand. Soils were sampled at 0–30, 31–60, and 61–120 cm soil depths prior to planting. Observed preplantation soil data are presented in Table S1. Details of experimental treatments, including three nitrogen application rates (NARs), including control and three planting windows (PWs) are provided in

Table 1. Details of experimental treatments, including nitrogen application rates (NARs) and planting windows during first (2018–2019) and second (2019–2020) crop growing seasons.

Nitrogen Application Rates		Planting Windows	First Season	Second Season
N0	Control—No N applied	Early	30 August	1 September
N30	30 kg N ha−1	Moderately delayed	26 September	6 October
N60	60 kg N ha−1	Delayed	31 October	3 November
N90	90 kg N ha−1

. Recommended cultural practices were followed for crop management. Further details on the experimental procedures, field plot management, experiment design and replicates, and crop management practices, including planting, irrigation, and the application of treatments, can be accessed in our previous study, and were employed in the modeling study as provided by Hussain et al. [4].

2.2. Data Collection, Observations, and Computations

Agronomic data, plant, and soil sampling for N determination concentrations were obtained at maturity during the harvest for each planting window. Details of agronomic data collection, including phenology, yield, and above-ground biomass production, were provided in a recently published study [4]. Soil sampling to determine the soil profile nitrate was performed at maturity at 0–30, 31–60, and 61–120 cm depths. Chopped rice straw samples and rice grain samples were first oven dried at 65 °C to achieve a constant weight and passed through a 1mm sieve to determine N concentrations in plant parts [41]. Soil and plant samples were used to determine N concentrations using the Kjeldahl method [42]. Nitrogen contents for plant N uptake and soil N were determined by multiplying respective weights and observed concentrations as outlined by Hammad et al. [43]. Meteorological data, including daily minimum and maximum temperatures (°C), wind speed (m·s⁻¹), solar radiation (MJ·m^–2·day⁻¹), rainfall (mm·day⁻¹), and relative humidity (%), were collected from the Agrometeorology–Agricultural Information Center of Kho Hong, Hat Yai, situated 1.8 km from the experimental location. Weather data (Figure S1) were then used as input for the simulation study. Data of irrigation water applied at specific events in each planting window and daily rainfall (mm) data were added to obtain the total water input for corresponding PWs.

2.3. Soil Water Contents and Evapotranspiration

Soil water contents were recorded in all experimental plots designated for each N application rate in each planting window during both seasons using a PR2 profile probe and HH2 m (PR2/6, Delta–T Devices Ltd., Cambridge, UK) [44]. One PR2 access tube per plot was installed in the center of the experimental plot. The PR2 access tubes were specially constructed thin-walled fiberglass tubes, which enhanced the electromagnetic field into surrounding soil. The PR2/6 soil moisture probe consisted of a sealed polycarbonate rod 25.4 mm in diameter, having electronic sensors at six fixed positions (10, 20, 30, 40, 60, and 100 cm). The PR2 probe could be moved from one access tube to another, which enabled the measurement of soil water contents in different experimental plots. The probe used electromagnetic signals to record the permittivity of soil water contents. Soil water contents were recorded normally in the morning on a weekly basis starting from the planting date until the maturity of each planting window. Observed soil water contents at two soil layers (0–30 cm and 30–60 cm) were compared with the simulated soil water contents. Actual seasonal evapotranspiration (ET) was computed using FAO-56 Penman–Monteith (PM) method and equation [45] for each treatment.

2.4. Model Description

WinEPIC (v.0810), which is a Windows-based graphic user interface (GUI) application of the EPIC model [46,47] that supports the executions of the EPIC model, was used in this study and the performance of the EPIC model was assessed. EPIC is one of the most extensively used crop models for simulating global crop production, providing useful outputs (i.e., crop performance under different stressed and nonstressed environments, water balance, nutrient balance, and economics). The EPIC model was developed by USDA-ARS to estimate soil productivity in the 1980s in the United Sates [47,48]. It simulates approximately more than one hundred crops using unique attributes of each crop [49]. EPIC has also been applied in rice research, including simulating yield and biomass [31], global yield estimations [50], water, and nutrients [51,52,53]. It has been further modified to simulate biophysical processes under different environmental conditions and adapted to analyze the sustainable productivity in complex cropping systems and decision support. The EPIC model is highly capable of simulating weather and climatic impacts, soil processes, including erosion, nutrient dynamics, tillage, and the management of crop growth and yield.

The EPIC model requires input data, including crop parameters, crop management practices, soil properties, field information, and daily weather data. Major outputs relevant to this research were components of crop productivity (grain yield, crop biomass, etc.) and soil water and N balance components. Considering the crop growth module in EPIC, the simulation of crop growth and yield was dependent upon crop parameters, which are unique characteristics for each crop [52]. To analyze soil water dynamics, water movement is a major process in EPIC, which is influenced by various factors, including evapotranspiration, surface runoff, subsurface lateral flow, and deep percolation. Fluctuations in soil water contents were considered in EPIC and water balance components were computed. The model also considered various levels of stresses, including water and nutrients. Fertilization information, including application rate, depth, and interval, were provided in the management input, following which the difference between average annual N uptake and N amount present in the root zone were computed and N requirement was determined. Prior to application of the EPIC model, calibration of crop parameters and Parm coefficients was necessary for better simulations [38].

2.5. Model Calibration and Validation

To determine the performance of the EPIC model in predicting the observed attributes based on applied N rates on upland rice planted under different planting windows, crop yields, nitrogen uptake, evapotranspiration, and soil water were analyzed. Suitable methods and operations were chosen in EPIC to compute desired variables. The EPIC model provided various options, including the FAO-56 Penman–Monteith (PM) model [45] and Priestley–Taylor evapotranspiration method [54], to determine potential evapotranspiration. The FAO-56 Penman–Monteith (PM) model was selected in this study to estimate evapotranspiration, as the same was used to compute the observed ET. In the first step, the model was calibrated using the initial soil analysis data for soil properties, weather data, the crop data, ET, and soil prolife nitrate observed from treatment with highest nitrogen application rate (N90) in the moderately delayed planting window (PW2) from two growing seasons. Crop parameters were calibrated according to the observed parameters and by adjusting the coefficient’s values to narrow down the gap between simulations and observations. The model is highly sensitive to crop parameters in the simulation of growth and yield [31]; therefore, modifications were performed according to the field observations, maximum observed crop performance, and considering the literature values. Parameters that were not modified during calibration were used as default values in EPIC. To improve the simulation accuracy for nitrogen-related parameters, available data were utilized. NO₃–N in irrigation water and rainfall was analyzed from the water samples collected from three rainfall events and irrigations, and averaged concentration was used as model input. Adjustment in N-related Parm coefficients was also performed to better simulate N dynamics. Following calibration, model evaluation was performed by comparing the goodness between simulated and observed parameters for remaining treatments in both seasons, and additional simulated components of soil water and N balance were also studied. To determine the simulation accuracy, statistical indicators of relative percentage difference, coefficient of multiple determinations (R²) in linear regression, normalized root mean square error (RMSEn) (1) [55], and the index of agreement (d) (3) [56] were computed. The RMSE (2) was first calculated to compute the RMSEn. These statistical indicators are tools to determine the goodness of simulations. Considering relative percentage difference, a positive measure indicated an overestimated result, whereas a negative measure indicated an underestimated result. R² ranging from 0 to 1 indicated the linearity among the simulations and observations [31]. With respect to RMSEn, low value and a high d-index value approaching 1 were desired to produce a good fit among simulations and observations. Values of RMSEn less than 10% indicated strong agreement and were considered excellent and good, and fair if the values ranged between 10–20% and 20–30%, respectively, whereas RMSEn values greater than 30% specified poor results [55].

RMSEn (1) and the RMSE (2) for simulated or predicted and observed values for a given dataset with n observed values were defined as,

$$\mathrm{RMSEn}=\frac{\mathrm{RMSE}\times 100}{\overline{O}}$$

(1)

$$\mathrm{RMSE}=\sqrt{\frac{ {\sum }_{i=1}^n(S_i-O_i{)}^2}{n}}$$

(2)

where S_i indicates the simulated value, O_i indicates the observed value, n indicates the number of observations, and $\overline{O}$ indicates the overall mean of observed values.

The value of d-index (3) was computed with the following equation,

$$d=1-\frac{{\sum }_{i=1}^n(S_i-O_i{)}^2}{{\sum }_{i=1}^n(\left|S_i^{\prime }\right|+\left|O_i^{\prime }\right|{)}^2}$$

(3)

where n indicates the number of observations, S_i indicates the simulate value, and O_i indicates the observed value; S_i′ = S_i − $\overline{O}$ and O_i′ = O_i − $\overline{O}$.

3. Results

3.1. Model Calibration

The EPIC model was calibrated to assess its performance and to quantify the impacts of the different NARs and PWs on upland rice responses, soil water, and N dynamics. Parameters, including grain yield, aboveground biomass, harvest index, grain and total crop N uptake, actual seasonal evapotranspiration (ET), and soil profile nitrate, observed from the highest N application rate (N90) under a moderately delayed planting window (PW2) for two seasons were compared. Adjustments in crop parameters and Parm coefficients (

Table 2. Calibrated EPIC model parameters and adjusted parm coefficients for simulating upland rice (cv. Dawk Pa–yawm) response using data from highest nitrogen application treatment (N90) in moderately delayed planting (PW2) of two growing seasons. Default values are presented in the brackets and the Parm coefficients as well as crop parameters not shown in the table were used with default values.

Parameter	Description	Value
WA	Biomass energy ratio	25 (25)
HI	Crop-specific potential harvest index defined as proportion of rice grain in the aboveground biomass under optimal conditions	0.46 (0.5)
DMLA	Maximum potential leaf area index	5.40 (6.0)
RLAD	LAI decline factor	0.80 (0.5)
RBMD	Biomass/energy decline rate	1.50 (0.5)
HMX	Maximum crop height (m)	1.44 (0.8)
RDMX	Maximum root depth (m)	1.20 (0.9)
CNY	Fraction of nitrogen in yield (kgs·kg−1)	0.0103 (0.0136)
PPLP1L	Plant population at first point (plants m–2)	105 (125)
PPLP2L	Plant population at second point (plants m–2)	313 (250)
RFN0	Average concentration of nitrogen in rainfall (ppm)	1.50 (0.8: 0.5–1.5)
CNO30	Concentration of NO3–N in irrigation water (ppm)	2.50 (0–1000)
Parm (20)	Microbial decay rate, adjusted soil water–temperature–oxygen equation	0.50 (0.5–1.5)
Parm (27)	Lower limit nitrate concentration, maintained soil nitrate concentration	0.50 (0–10)
Parm (63)	Upper limit of N concentration in percolating water (ppm)	100 (100–10,000)

) resulted in a good agreement between the simulated and observed values for the studied attributes (

Table 3. Comparison of simulated and observed parameters, percentage differences and RMSEn (%) using dataset from highest nitrogen application treatment (N90) in moderately delayed planting (PW2) from two growing seasons for EPIC model calibration.

Parameters	First Season (2018–2019)			Second Season (2019–2020)			RMSEn
	Simulated	Observed	Difference %	Simulated	Observed	Difference %
Grain yield (kg ha−1)	5351.00	5271.84 ± 46.50	1.49	3813.00	3764.11 ± 106.00	1.29	1.0
Aboveground biomass (kg ha−1)	10,490.00	11,484.71 ± 80.40	−9.05	9156.00	8314.93 ± 234.20	9.63	9.3
Harvest index (-)	0.42	0.46 ± 0.01	−8.88	0.40	0.45 ± 0.00	−11.76	9.9
Grain N uptake (kg ha−1)	58.00	62.00 ± 0.50	−6.66	40.20	40.28 ± 1.10	−0.19	5.5
Total N uptake (kg ha−1)	104.90	113.87 ± 0.60	−8.20	82.40	73.04 ± 2.10	12.05	9.8
Evapotranspiration (mm)	675.98	682.40 ± 7.90	−0.95	644.90	661.60 ± 11.50	2.60	1.9
Soil profile nitrate (kg ha−1)	48.50	46.56 ± 0.90	4.08	51.30	48.39 ± 1.10	5.84	5.2

Observed data are presented as means with ± standard errors of 3 experimental replications and computations.

). Considering all the studied parameters, the percentage difference ranged between −9.05 and 4.08% for the first season and between −11.76 and 12.05% for the second season, whereas the RMSEn was less than 10%, which indicated a good agreement between the simulated and observed values (Table 3).

3.2. Upland Rice Productivity

The upland rice performance was highly influenced by the impact of N application and plant management. Maximum grain yields were observed under N90 in PW2. The simulated grain yield with the N addition compared to the control increased by 24–70% at PW1, 34–94% at PW2, and 38–96% at PW3 during the first season, and by 10–37% at PW1, 15–76% at PW2, and 6–42% at PW3 during the second season (Figure 1A,B). In the model validation, which was performed for all treatments except for N90 at PW2, the simulated grain yield was 9% higher for N0, N30, and N60, and 7% higher for N90 at PW1 in the first season, while it was 6%, 5%, 1%, and 10% higher in the second season, respectively. The simulated grain yield for PW2 was 7%, 9%, and 6% higher in the first season, and 12%, 8%, and 7% lower in the second season for N0, N30, and N60, respectively. Considering the delayed planting PW3, the grain yield was underestimated for all NARs in both seasons by 9%, 10%, 13%, and 8% in the first season and by 10%, 9%, 8%, and 9% in the second season for N0, N30, N60, and N90, respectively (Figure 1A,B). The statistical relationship and degree of agreement between the simulated and observed grain yield for all planting windows and NARs are illustrated in Figure 1A,B.

Like the grain yield maximum, the aboveground biomass was observed under N90 in PW2. The simulated aboveground biomass with the N addition compared to the control increased by 25–74% at PW1, 34–80% at PW2, and 29–88% at PW3 during the first season, and by 19–59% at PW1, 26–62% at PW2, and 21–46% at PW3 during the second season (Figure 1C,D). The aboveground biomass was underestimated by EPIC at PW1 for all NARs in both seasons by 12%, 14%, 14%, and 11% in the first season and by 13%, 16%, 14%, and 13% in the second season for N0, N30, N60, and N90, respectively. Under PW2, the aboveground biomass was underestimated in the first season by 8%, 17%, and 3% and overestimated in the second season by 10%, 8%, and 9% for N0, N30, and N60, respectively. Considering the delayed planting PW3, the aboveground biomass was oversimulated for all NARs in both seasons by 11%, 4%, 8%, and 13% in the first season and by 12%, 11%, 8%, and 9% in the second season for N0, N30, N60, and N90, respectively (Figure 1C,D). The statistical relationship and degree of agreement between the simulated and observed aboveground biomass for all planting windows and NARs are presented in Figure 1C,D).

The harvest index followed a similar trend with the grain yield and aboveground biomass, and EPIC overestimated the harvest index for PW1 by 9%, 9%, 10%, and 12% during the first season in N0, N30, N60, and N90, respectively. The harvest indexes were underestimated by 16%, 10%, and 8% in the N0, N30, and N60 treatments, whereas they were oversimulated in N90 by 6% during the second season. Considering the moderately delayed planting (PW2), the harvest index was matched for N0 in the first season; however, it was oversimulated by 12% in N30 and underestimated by 10% in the N60 treatment. In the second season, the model underestimated the harvest index by 20%, 11%, and 9% in the N30, N60, and N90 treatments, respectively. In the delayed planting PW3, the harvest index was matched for N0 during the first season, whereas it was oversimulated by 1%, 4%, and 2% for the N30, N60, and N90 treatments, respectively. In the second season, the model underestimated the harvest index by 10%, 3%, 10%, and 11% for N0, N30, N60, and N90, respectively (Figure 1E,F). The statistical relationship and degree of agreement between the simulated and observed harvest index for all planting windows and NARs are presented in Figure 1E,F.

3.3. Water Balance Components

3.3.1. Total Water Input and Evapotranspiration

A model evaluation for the soil water balance components was performed. The total water input was computed as the rainfall received and irrigation water applied at each planting window. The highest rainfall was received in PW1, and the amount of rainfall decreased with delayed planting. PW3 received the lowest rainfall and highest irrigation water input (Table S2). The FAO-56 Penman–Monteith (PM) method was used to compute the actual seasonal evapotranspiration (ET) for each planting window and N application rate. Considering the EPIC model’s performance for replicating seasonal ET, generally, it was slightly underestimated for all N application rates at all planting windows in both seasons. For PW1, the seasonal ET was underestimated by 2% in N0 and 3% in N30, N60, and N90 during the first season, and by 2% in N0, N30, and N60 and by 3% in N90 during the second season. Similarly, for PW2, it was underestimated by 2% in N0 and N30 and 3% in N60 during the first season, and by 1% in N0 and 2% in N30 and N60 during the second season. For PW3, the ET was also underestimated by 2% in N0 and N30, 3% in N60, and 4% in N90 during the first season, and by 1% in N0 and N30, 2% in N60, and 3% in N90 during the second season. The statistical indicators showed a good agreement between the simulated and observed seasonal ET for all planting windows and N application rates (Figure 2).

There were good agreements among the observed and simulated ETs, and the highest ET was observed at N90 in all planting windows in both seasons (Figure 2). Therefore, the simulated ET was used for the comparison of the water balance. The simulated seasonal ET varied among N application rates and planting windows; however, like the observed data, the maximum seasonal ET was observed in N90 under all planting windows (

Table 4. Water balance components, including total water input (irrigation and precipitation), simulated evapotranspiration, surface runoff, and deep percolation losses, for upland rice grown under different planting windows and nitrogen application rates during first season (2018–2019) and second season (2019–2020).

Growing Season	Planting Window	Nitrogen Application Rate	Total Water Input	Actual Evapotranspiration	Surface Runoff	Deep Percolation Losses
		kg ha−1		<mm>
2018–2019	PW1	0	1257	627.3 (+3)	331.2 (+40)	255.4 (+13)
		30	1257	638.6 (0)	329.3 (+41)	247.1 (+22)
		60	1257	648.6 (−2)	327.6 (+40)	239.8 (+23)
		90	1257	657.5 (−3)	327.3 (+40)	228.4 (+21)
	PW2	0	1092	607.3	236.3	225.6
		30	1092	638.6	233.8	201.9
		60	1092	661.6	233.8	195.2
		90	1092	676.0	233.0	188.6
	PW3	0	857	551.4 (−9)	139.5 (−41)	157.2 (−30)
		30	857	568.0 (−11)	134.5 (−42)	155.3 (−23)
		60	857	579.6 (−12)	133.5 (−43)	154.5 (−21)
		90	857	599.3 (−11)	131.6 (−44)	150.8 (−20)
2019–2020	PW1	0	1156	611.7 (+2)	305.6 (+51)	226.4 (+21)
		30	1156	620.8 (−1)	304.1 (+54)	216.1 (+17)
		60	1156	625.8 (−1)	302.3 (+55)	208.3 (+13)
		90	1156	631.8 (−2)	300.0 (+53)	202.3 (+17)
	PW2	0	978	601.8	202.4	186.9
		30	978	624.4	197.4	185.4
		60	978	634.5	194.4	183.6
		90	978	644.6	196.0	172.6
	PW3	0	749	556.9 (−7)	91.7 (−55)	139.1 (−26)
		30	749	565.5 (−7)	87.2 (−56)	137.1 (−26)
		60	749	569.1 (−10)	85.4 (−56)	136.0 (−26)
		90	749	572.8 (−11)	81.6 (−58)	136.0 (−21)

Note: Percentage changes in simulated actual evapotranspiration, surface runoff, and deep percolation losses were given in brackets for each treatment, computed and compared with respect to the nitrogen application rates under moderately delayed planting window (PW2) in two growing seasons.

). Compared to N0, the simulated seasonal ET in PW1 was 2%, 3%, and 5% higher during the first season and was 1%, 2%, and 3% higher during the second season in N30, N60, and N90, respectively. In PW2, the simulated seasonal ET was also 5%, 9%, and 11% higher in the first season and 4%, 5%, and 7% higher in the second season in N30, N60, and N90, respectively, compared to N0. Compared to N0, the PW3 was 3%, 5%, and 9% higher in the first season and 2%, 2%, and 3% higher during the second season in N30, N60, and N90, respectively (Table 4).

Considering the change in the simulated seasonal ET at PW1 and PW3 relative to the N application rates at PW2, the simulated actual ET in PW1 was 3% higher in N0, similar to that in N30, 2% and 3% less in N60 and N90, respectively, during the first season, and during the second season, it was 2% higher in N0, 1% less in N30 and N60, and 2% less in N90 than the seasonal ET simulated at PW2 in N0, N30, N60, and N90, respectively. However, the simulated seasonal ET at PW3 decreased in both seasons. In the first season, it was 9–12% less, and in the second season, it was 7–11% less than that simulated at PW2 for different NARs, respectively (Table 4).

3.3.2. Surface Runoff

Simulated surface runoff slightly decreased in the treatments with high N fertilization, and it was highly variable under different planting windows (Table 4). In contrast to the N application rates, the planting windows were the main factor influencing surface runoff. The simulations indicated that the highest surface runoff losses occurred in N0 compared to N90 under all planting windows during both seasons (Table 4). The simulated surface runoff losses during PW1 were 1% less in N30, N60, and N90 during the first season and were 1%, 1%, and 2% less during the second season in N30, N60, and N90, respectively, as compared to N0. In PW2, the simulated surface runoff losses were also 1% less for N30, N60, and N90 during the first season and 2%, 4%, and 3% less during the second season in N30, N60, and N90, respectively, compared to N0. Considering PW3, the surface runoff was 4%, 4%, and 6% less during the first season and it was 5%, 7%, and 11% less in the second season in N30, N60, and N90, respectively, as compared to N0.

In a comparison of the simulated surface runoff losses during PW1 and PW3 with respect to the N application rates during PW2, it was found that the surface runoff losses during the first season in PW1 were 40–41% higher, and in the second season they were 51–55% higher than the surface runoff losses that occurred during PW2 for the different NARs, respectively. However, the surface runoff losses for PW3 decreased during both seasons, in the first season being 41–44% and in the second season being 55–58% less than the that which occurred during PW2 for the different NARs, respectively (Table 4).

3.3.3. Deep Percolation Losses

Deep percolation losses decreased in the high N application treatments, and they were highly variable under different planting windows (Table 4). Similar to other water balance components, the planting window was the main factor influencing deep percolation losses compared to the N application treatments. The model’s results indicated that the highest deep percolation losses occurred in N0 under all planting windows during both seasons (Table 4). The simulated deep percolation losses during PW1 were 3%, 6%, and 11% less in the first season and 5%, 8%, and 11% less during the second season in N30, N60, and N90, respectively, as compared to N0. In PW2, the simulated deep percolation losses were 10%, 13%, and 16% less during the first season and were 1%, 2%, and 8% less during the second season in N30, N60, and N90, respectively, as compared to N0. Considering PW3, the deep percolation losses were 1%, 2%, and 4% less during the first season and were 1%, 2%, and 2% less during the second season at N30, N60, and N90, respectively, as compared to N0.

In a comparison of simulated deep percolation losses during PW1 and PW3 relative to the N application treatments in PW2, it was found that the deep percolation losses during the first season in PW1 were 13–23% higher in the first season and in the second season were 13–21% higher than the deep percolation losses simulated for PW2 at the different NARs, respectively. However, the deep percolation losses in PW3 decreased during both seasons and were 21–30% less in the first season and 21–26% less than that simulated in PW2 during the second season for the different NARs, respectively (Table 4).

3.3.4. Soil Water Contents

There were no considerable differences in the observed as well as simulated soil water contents among the N application treatments. Therefore, we used average soil water contents and compared them for planting windows in both seasons. The simulated and observed soil water contents for the two soil layers (i.e., 0–30 cm and 30–60 cm) are presented for all the planting windows in both growing seasons in Figure 3. Overall, there was a good agreement between the simulated and observed soil water contents for both seasons. The simulated soil water contents were generally overestimated. For PW1, the values for the RMSEn and d-index were 15.6% and 0.81 and 13.0% and 0.62 for the first season and 13.3% and 0.88 and 13.8% and 0.81 for the second season for the 0–30 cm and 30–60 cm soil layers, respectively. The soil water contents for PW2 were well simulated in both seasons. The RMSEn and d-index were 16.4% and 0.87 and 15.5% and 0.79 for the first season and 10.8% and 0.96 and 15.2% and 0.91 for the second season for the 0–30 cm and 30–60 cm soil layers, respectively. In comparison to PW1 and PW2, the soil water contents for PW3 were fairly simulated for both seasons. The RMSEn and d-index were 16.6% and 0.91 and 20.6% and 0.81 for the first season and 19.3% and 0.84 and 17.4% and 0.84 for the second season for the 0–30 cm and 30–60 cm soil layers, respectively.

The experimental observations indicated that the soil water contents were comparatively higher in the 30–60 cm soil layer for all planting windows in both seasons (Figure 3B,D). The results showed that fluctuations in the soil water contents were higher in PW1 and PW2, in which the higher rainfall events were observed. A visual comparison indicated that fluctuations in the simulated soil water contents were higher in the upper 0–30 cm soil layer (Figure 3A,C), whereas the model showed less fluctuations in the 30–60 cm soil layer (Figure 3B,D), particularly for PW3 in both seasons.

3.4. Nitrogen Balance and Components

3.4.1. Nitrogen Uptake

Nitrogen uptake varied among NARs and PWs; however, like the grain yield and aboveground biomass, the maximum N uptake was observed in N90 at PW2. The simulated N uptake also increased with increased N additions. The simulated grain N uptake with the N addition compared to the control increased by 37–142% at PW1, 38–109% at PW2, and 28–116% at PW3 during the first season and by 16–64% at PW1, 14–62% at PW2, and 8–67% at PW3 during the second season (Figure 4A,B). In the model evaluation, which was performed for all the treatments except N90 at PW2, the simulated grain N uptake was overestimated in both seasons for PW1, by 10%, 14%, 6%, and 16% in the first season and by 14%, 11%, 8%, and 8% in the second season for N0, N30, N60, and N90, respectively. The simulated grain yield for PW2 was also overestimated for N0, N30, and N60 in both seasons, except for being underestimated for N60 in the second season. There were oversimulations of the grain N uptake of 14%, 6%, and 8% in the first season for N0, N30, N60, respectively, and 14% and 11% for N0 and N30 in the second season. The grain N uptake was undersimulated by 3% in the second season for N60. Considering delayed planting PW3, the grain N uptake was oversimulated for all NARs in both seasons by 28%, 17%, 15%, and 16% in the first season and by 22%, 19%, 14%, and 12% in the second season for N0, N30, N60, and N90, respectively (Figure 4A,B). The statistical relationship and degree of agreement between the simulated and observed grain N uptake for all planting windows and NARs are presented in Figure 4A,B.

The simulated total plant N uptake also increased with the increased N rate. The simulated total plant N uptake with the N addition compared to the control increased by 43–124% at PW1, 34–124% at PW2, and 47–119% at PW3 during the first season and by 11–70% at PW1, 26–109% at PW2, and 5–29% at PW3 during the second season (Figure 4C,D). The simulated total plant N uptake was oversimulated in both seasons for PW1 by 19%, 18%, 5%, and 9% in the first season, and by 24%, 11%, 16%, and 10% in the second season for N0, N30, N60, and N90, respectively. The simulated total plant N uptake for PW2 was oversimulated for N0 by 6% and underestimated for N30 and N60 by 12% and 1%, respectively, in the first season. In the second season, the total plant N uptake for PW2 was oversimulated for the evaluated N application rates, including N0, N30, and N60 by 14%, 10%, and 10%, respectively. Considering delayed planting PW3, the total plant nitrogen uptake was oversimulated for all NARs in both seasons by 31%, 30%, 15%, and 16% in the first season and by 38%, 22%, 7%, and 2% in the second season for N0, N30, N60, and N90, respectively (Figure 4C,D). statistical relationship and degree of agreement between the simulated and observed total plant N uptake for all planting windows and NARs are presented in Figure 4C,D.

3.4.2. Net N Mineralization

The nitrogen input affected the net N mineralization under all planting windows. The nitrogen input from rainfall and irrigation was also different due to changes in rainfall and irrigation water among planting windows (Table S2). The highest rainfall was received in PW1 followed by PW2 and PW3. Therefore, the N input from the total water input also followed a similar trend, and the highest N from the total water input was received at PW1, followed by PW2 and PW3 (Table S2). The net N mineralization simulations indicated that an increase in the N application rate increased the N mineralization, and the highest N mineralization occurred at N90 compared to N0 under all planting windows during both seasons (

Table 5. Nitrogen balance components, including nitrogen (N) from net mineralization, nitrate loss in surface runoff, nitrate leaching, N volatilization, and soil profile nitrate simulated for upland rice grown under different planting windows and N application rates during first season (2018–2019) and second season (2019–2020).

Growing Season	Planting Window	Nitrogen Application Rate	Net N Mineralization	Nitrate Loss in Surface Runoff	Nitrate Leaching	N Volatilization	Soil Profile Nitrate
		kg ha−1
2018–2019	PW1	0	57.7 (−13)	13.4 (+38)	53.6 (+5)	9.9 (−6)	29.6 (+7)
		30	61.5 (−12)	15.5 (+48)	54.9 (+3)	10.5 (−15)	42.9 (+5)
		60	65.8 (−11)	19.4 (+63)	58.5 (+7)	11.2 (−15)	47.7 (+5)
		90	76.2 (−11)	22.0 (+42)	61.1 (+9)	12.3 (−9)	48.6 (+4)
	PW2	0	66.5	9.7	50.8	10.6	27.6
		30	69.5	10.4	53.5	12.3	40.7
		60	73.6	11.9	54.6	13.2	45.3
		90	85.9	15.5	56.0	13.6	46.6
	PW3	0	48.5 (−27)	8.4 (−13)	33.3 (−34)	7.5 (−30)	28.5 (+3)
		30	53.3 (−23)	9.1 (−13)	36.7 (−31)	8.5 (−31)	40.4 (−1)
		60	66.2 (−10)	10.7 (−10)	38.2 (−30)	9.2 (−30)	42.8 (−5)
		90	70.5 (−18)	13.0 (−16)	43.0 (−23)	10.2 (−25)	45.8 (−2)
2019–2020	PW1	0	57.1 (−9)	6.6 (+2)	63.8 (+27)	8.3 (+1)	24.8 (−3)
		30	59.6 (−12)	11.1 (+20)	65.6 (+28)	9.8 (0)	41.9 (+1)
		60	66.5 (−12)	14.5 (+40)	66.7 (+27)	10.6 (−39)	46.7 (+2)
		90	75.9 (−7)	16.2 (+41)	67.3 (+27)	11.3 (−39)	49.6 (+2)
	PW2	0	62.9	6.5	50.3	8.2	25.5
		30	67.4	9.3	51.2	9.8	41.6
		60	75.6	10.3	52.7	11.7	45.6
		90	81.9	11.5	53.1	12.3	48.4
	PW3	0	47.8 (−24)	4.5 (−13)	30.6 (−39)	7.0 (−15)	37.0 (+45)
		30	48.8 (−28)	9.9 (+7)	31.9 (−38)	8.6 (−13)	44.9 (+8)
		60	61.2 (−19)	10.8 (+4)	32.6 (−38)	9.4 (−19)	46.6 (+2)
		90	66.2 (−19)	13.8 (+20)	34.7 (−35)	10.4 (−16)	48.6 (0)

Note: Percentage changes in nitrogen balance components are presented in brackets for each treatment, which were computed and compared with respect to the nitrogen application rates under moderately delayed planting (PW2) in two growing seasons.

). In PW1, the simulated mineralization was 7%, 14%, and 32% higher in the first season and 4%, 16%, and 33% higher in the second season at N30, N60, and N90, respectively, as compared to N0. Under PW2, the simulated mineralization was 5%, 11%, and 29% higher in the first season and 7%, 20%, and 30% higher in the second season at N30, N60, and N90, respectively, as compared to N0. Considering PW3, the simulated mineralization was 10%, 37%, and 45% higher in the first season and 2%, 28%, and 38% higher in the second season at N30, N60, and N90, respectively, as compared to N0.

A comparison of the net N mineralization at PW1 and PW3 with respect to the N application rates in PW2 indicated that the highest net N mineralization occurred under PW2 (Table 5). The net N mineralization in PW1 was 11–13% and 7–12% less in the first season and second season, respectively, compared to the net N mineralization that occurred in PW2 at different NARs. Similarly, the net N mineralization in PW3 was 10–27% and 19–28% less in the first season and the second season, respectively, compared to the net N mineralization that occurred in PW2 at different NARs (Table 5).

3.4.3. Nitrate Loss in Surface Runoff

The nitrate loss in the surface runoff varied with planting windows and increased with an increase in the N fertilization rates (Table 5). The simulated nitrate loss in the surface runoff indicated an increase in the N application rate that generally increased the nitrate loss, and the highest nitrate loss occurred at N90 as compared to N0 under all planting dates during both seasons (Table 5). In early planting (PW1), the simulated nitrate losses were 16%, 45%, and 64% higher in the first season and were 67%, 118%, and 144% higher in the second season at the N30, N60, and N90 application rates, respectively, compared to N0. Under PW2, the simulated nitrate losses at N30, N60, and N90 were 7%, 23%, and 59% higher in the first season and were 42%, 58%, and 76% higher in the second season, respectively, compared to N0. Considering delayed planting (PW3), the nitrate losses were 8%, 27%, and 54% higher in the first season and 121%, 140%, and 207% higher in the second season at N30, N60, and N90, respectively, compared to N0.

A comparison of nitrate loss in surface runoff at PW1 and PW3 compared to the N application rates at PW2 indicated that the highest nitrate loss occurred under PW1 (Table 5). In the first season, the nitrate loss in surface runoff in PW1 was 38–63% higher, and in the second season it was 2–41% higher than the nitrate loss in the surface runoff that occurred in PW2 at different NARs. Conversely, the nitrate loss in surface runoff in PW3 varied in both seasons. The nitrate losses were 10–16% less in the first season and were 13% less at N0 and 4–20% higher in the second seasons than those that occurred in PW2 at different NARs (Table 5).

3.4.4. Nitrate Leaching

Nitrate leaching increased with the increase in N fertilization and was affected by different planting windows (Table 5). An increase in the N application rate generally increased the nitrate leaching losses, and the highest nitrate leaching loss occurred at N90 as compared to N0 during all planting windows in both seasons (Table 5). In PW1, the simulated nitrate leaching was 2%, 9%, and 14% higher in the first season and 3%, 5%, and 5% higher in the second season at N30, N60, and N90, respectively, as compared to N0. Under moderate delayed planting (PW2), the simulated nitrate leaching was 5%, 8%, and 10% higher in the first season and 2%, 5%, and 6% higher in the second season at N30, N60, and N90, respectively, compared to N0. Delayed planting (PW3) resulted in 10%, 15%, and 29% higher simulated nitrate leaching in the first season and 4%, 7%, and 13% higher simulated nitrate leaching in the second season at N30, N60, and N90, respectively, compared to N0.

A comparison of the nitrate leaching losses at PW1 and PW3, relative to the N application rates at PW2, indicated that the highest nitrate leaching loss occurred under PW1 (Table 5). In the first season (PW1), nitrate leaching was 3–9% higher than PW2, while in the second season it was 27–28% higher than in PW2 at different NARs. However, nitrate leaching decreased for both seasons at PW3, being 23–34% less and in the second season and 35–39% less than that which occurred in PW2 at different NARs (Table 5).

3.4.5. Volatilization

The simulated N volatilization indicated an increase with the increase in N fertilization, and it varied under different planting windows (Table 5). The results indicated that an increase in the N application rate generally increased the N volatilization losses, and the highest N volatilization loss occurred at N90 as compared to N0 under all planting windows during both seasons (Table 5). The simulated N volatilization losses at PW1 were 6%, 13%, and 24% higher in the first season and were 18%, 27%, and 35% higher in the second season in N30, N60, and N90, respectively, as compared to N0. In PW2, the simulated N volatilization losses were 16%, 24%, and 28% higher in the first season and were 20%, 43%, and 50% higher in the second season at N30, N60, and N90, respectively, compared to N0. Considering PW3, the N volatilization was 14%, 23%, and 36% higher in the first season and 23%, 35%, and 48% higher in the second season at N30, N60, and N90, respectively, as compared to N0.

A comparison of the simulated N volatilization losses at PW1 and PW3 with respect to the N application rates in PW2 indicated differences in N volatilization losses. In the first season, the PW1 volatilization losses were 6–15% less, while in the second season were 1% higher at N0, with no difference at N30, and were 10% and 9% less at N60 and N90 than the N volatilization losses that occurred in PW2 at different NARs. However, the N volatilization losses in PW3 decreased by 25–33% in the first season and by 13–19% in the second season compared to PW2 for different NARs (Table 5).

3.4.6. Soil Profile Nitrate

The soil profile nitrate was depleted under low N fertilization. The maximum soil profile nitrate was observed in N90 at all planting windows in both seasons. A higher N application resulted in a higher soil profile nitrate (Figure 5). Considering the EPIC model’s performance for simulating the soil profile nitrate, generally, it underestimated for all N application rates at all planting windows in both seasons. For PW1, the soil profile nitrate was underestimated by 12%, 7%, 4%, and 4% in the first season and by 13%, 5%, 4%, and 6% in the second season for N0, N30, N60, and N90, respectively. Similarly, the soil profile nitrate for PW2 was underestimated by 16%, 10%, and 6% in the first season and by 12%, 5%, and 3% in the second season for N0, N30, and N60, respectively. It was also underestimated in PW3 by 10%, 7%, 1%, and 6% in the first season and by 6%, 5%, 4%, and 1% in the second season for N0, N30, N60, and N90, respectively.

Since there was a good statistical agreement between the simulated and observed soil profile nitrate for all planting windows and N application rates (Figure 5), we used the simulated soil profile nitrate for the comparison of N balance between treatments. The simulated soil profile nitrate varied among N application rates and planting windows. The maximum soil profile nitrate was observed at N90 under all planting windows. Compared to N0, the simulated soil profile nitrate at PW1 was 45%, 61%, and 64% higher in the first season and 69%, 88%, and 110% higher in the second season at N30, N60, and N90, respectively. In PW2, the simulated soil profile nitrate was also 47%, 64%, and 69% higher in the first season and 63%, 79%, and 90% higher in the second season at N30, N60, and N90, respectively, as compared to N0. Considering PW3, the soil profile nitrate was 42%, 50%, and 61% higher in the first season and 21%, 26%, and 31% higher in the second season at N30, N60, and N90, respectively compared to N0 (Table 5).

Considering the change in the simulated soil profile nitrate at PW1 and in PW3 with respect to the N application rates in PW2, the simulated soil profile nitrate in PW1 was 4–7% higher in the first season and was 3% less at N0 and 1%, 2%, and 2% higher in the second season at N30, N60, and N90, respectively, compared to the soil profile nitrate simulated for PW2 at different NARs. The simulated soil profile nitrate at PW3 was 3% higher at N0 and 1%, 5%, and 2% less during the first season at N30, N60, and N90, respectively, and was 2–45% higher during the second season for N0–N60 and similar at N90 to that of the simulated soil profile nitrate at PW2 for different NARs (Table 5).

4. Discussion

In this study, the performance of the EPIC model was assessed for its potential use as a decision support tool in the management of N fertilizer application and planting period by simulating the upland rice yields, soil water, and N balance components under different N application treatments and PWs. The model run using the default input parameters did not produce the desired results, and an adjustment in the soil and crop parameters and Parm coefficients was necessary. Similar findings were also reported in a calibration and validation study of the EPIC model in a maize production [32]. Therefore, in the first step, the model was calibrated with the observed data from the N treatment with the highest application rate (N90) in the moderately delayed planting window (PW2) from two growing seasons. EPIC indicated a sensitivity to the soil parameters and Parm coefficients and a comparatively higher sensitivity to the crop input parameters in simulating the upland rice yields. Le et al. [31] also reported that EPIC was highly sensitive to the crop parameters in a simulation of growth and yield responses. An adjustment in the model input parameters then resulted in a better simulation accuracy and the model was able to replicate the observations. This was in line with the statement of Xiong et al. [50], who reported that a good agreement could be obtained through the calibration of the model with site-specific information on crop and management.

Following the model calibration, the model was validated with the remaining treatments and was able to replicate the observed data on the grain yield, aboveground biomass, harvest index, crop and grain N uptake, seasonal ET, and soil profile nitrate. Although there were variations in overestimations and underestimations in comparison, there was a good statistical agreement between the simulated and observed data. The ability of the EPIC model to simulate rice yields under various management options is well documented [31,49,52]. Worou [57] and Le et al. [31] found that EPIC was able to simulate rice productivity and, generally, the results were overestimated. Simulations for crop performance also followed a similar trend with observed data, and the simulated and observed data indicated that an increase in N application resulted in increased crop performance and productivity. The increased crop performance was due to the positive role of N fertilization and enhanced yield components, as reported earlier [4], and is also well documented [58,59]. The model results also confirmed that the change in planting period caused variation in the yield response of upland rice in respective N application treatments. This indicated that PW influenced N utilization and the model was sensitive to the changes in weather conditions that prevailed at different PWs.

To explore the effects of different NARs and PWs, we focused on the water and N balance components that indicated the major changes that occurred in the soil water and N balance during the crop growth period of each PW. The maximum agronomic performance of upland rice was observed in N90 at PW2; therefore, the soil water and N balance components of all NARs in PW1 and PW3 with respect to the NARs in PW2 were compared. Besides yields and the N uptake of upland rice that increased with the increase in NARs under all PWs, we observed that PW was the major factor impacting the crop performance, soil water, and N dynamics. Evapotranspiration was higher in PW1 and PW2 because of high water availability compared to PW3. Amin et al. [17] also observed that ET was higher under the treatments with high soil water availability. A highest ET occurred in PW2 followed by PW1 and PW3, indicating improved crop performance at PW1 and PW2 and increased water utilization. The highest rainfall was received in PW1, where most heavy rainfall intervals occurred during both seasons, and the total water input was also higher in PW1. This resulted in higher surface runoff (40–55%) compared to PW2. Similarly, deep percolation losses were also higher in PW1 compared to PW2 during both seasons. However, the runoff and deep percolation losses decreased at PW3 in both seasons. This was not because of a higher moisture utilization by the crops, but because of the lower rainfall occurrence at PW3. The lowest rainfall was also received at PW3. Our results were in line with the findings of Amin et al. [17], who observed that there was a strong relationship between the total water input and percolation losses.

The total water input and mainly rainfall events impacted soil water contents under different PWs. In general, EPIC was able to predict soil water contents, but the simulation accuracy was comparatively low (RMSEn: 15–27%) compared to other assessed attributes. In PW1 and PW2, where the rainfall and total water input were high, the model indicated fluctuations in the soil water contents. However, under PW3, where the rainfall and total water input were low, the model indicated less fluctuations from a visual comparison of the soil water contents, particularly for the 30–60 cm soil depth. This indicated that the model’s robustness in simulating soil water contents under water-limited conditions was affected. At the early crop stages in all PWs, both the simulated and observed soil water contents in the 0–30 and 30–60 cm soil depths were higher and around the field capacity; however, at the mid stages, and particularly at the lateral crop stages, these were decreased near to the wilting point. The soil water contents in PW1 and PW2 were most suitable for plant growth, as most rainfall events occurred in PW1 and PW2, whereas there was less rainfall in PW3, which triggered the need for a high amount of supplementary irrigation. The maximum rainfall at PW1 caused higher runoff losses, whereas in PW2, the crops received a better rainfall distribution, thus, retaining ideal soil water contents at critical crop stages. Delayed planting (PW3) encountered most dry intervals, and there was a need for frequent irrigation that required more supplemental irrigation. Soil water contents also dropped to near wilting point at crop reproductive stages in PW3, particularly at the upper soil layer in both seasons, and possibly caused by the drought stress. In addition, the occurrence of higher temperature and less rainfall at PW3, particularly at middle and lateral crop stages, impacted crop performance, which resulted in low yields. Rice is highly sensitive to drought stress [60], especially at critical crop stages, and variable reductions in crop yields under ranges of water stresses have been well documented [8,40,61]. Considering the results, an adjustment in PW to the favorable time would help to attain the maximum soil water utilization and would stabilize yields. Candradijaya et al. [15] stated that rice yield is highly vulnerable to the planting period. It was also noted that a delay in planting of approximately 20 days from the actual planting time would enhance rice yields in future climate scenarios [49]. Similar results were also reported by Kassie et al. [62] in Ethiopia, indicating that a delay in the planting period increased the crop yields. However, the delay in planting is highly dependent upon the region-specific predicted weather conditions, as seasonal rainfall patterns are highly variable over different regions, particularly in southern Thailand. Our data indicated that too delayed planting resulted in low yields due to less rainfall occurrence and high temperatures at reproductive stages. An adjustment in planting for improved rainwater utilization was recommended in recent research [63]. In addition, a climate change study conducted in Thailand also predicted that a delay in planting would increase rice yield by 23% in the 2050s [64]. The results of a study conducted by Lar et al. [50] based on the EPIC model also suggested that changing the planting period is a good strategy in managing rice yield reductions in the future.

The soil water status not only influences water use efficiency, but also influences the nutrient balance and utilization [65,66]. The low soil water availability at PW3 possibly impacted the N uptake by plants. The simulated and observed data indicated that the N uptake by upland rice was significantly reduced at PW3 compared to PW1 and PW2. Heavy rains at early crop stages at PW1 and PW3 possibly depleted the nitrate in the soil profile in the upper layer due to a lower tiller formation, higher runoff, and deep percolation losses. Nitrogen loss at early stages and reduced N availability at middle and lateral stages possibly impacted the crop performance at PW3. A recent study confirmed that there is a strong relationship between higher soil water losses and nutrient losses [17]. Therefore, managing the water input and the improved utilization of soil water could help achieve a better supply and reduce the loss of nutrients. A study of N balance indicated that the net N mineralization was also affected at PW1 and PW3 when the nitrate loss in the surface runoff was higher compared to PW2. A low net N mineralization was possibly because of the higher N losses at PW1 and reduced soil water availability at PW3. Similarly, N losses in the surface runoff and leaching were higher at PW1 and lower at PW2 and PW3 due to a moderate rainfall distribution at PW2 and reduced rainfall at PW3. Nitrogen leaching is percolation-driven in the field [28], and a significant impact in water availability on nitrate leaching has been observed in previous studies [17,67]. Frequent water availability also increases the N leaching losses [28]. A slight shift in PW according to the predicted weather conditions could result in reduced nitrate leaching, as the results confirmed that nitrate leaching losses were higher at PW1, which received the maximum rainfall and total water input, while it was reduced at PW2, which received a moderate distribution of rainfall and low irrigation water. A comparable recommendation was reported by Peng et al. [67], who stated that a decrease in percolation could considerably decrease N leaching losses in rice fields. Nitrate leaching was reduced at PW3 due to reduced rainfall and soil water availability. In addition, the low water availability reduced the N uptake by plants, resulting in a reduction in crop yield at PW3. The N volatilization was also affected by the effect of PW; however, PW1 and PW2 had a similar trend. Considering soil profile nitrate, at PW2, the maximum N uptake by the crops occurred, resulting in reduced soil profile nitrate compared to PW1 and PW3. In summary, significant relationships among water and N balance components and high-water input resulted in high N losses, and low water availability resulted in reduced N availability to the crops at critical times, impacting crop yields. A moderate distribution of rainfall at PW2 resulted in less supplemental irrigation, reduced water loss, and, hence, reduced N losses. A shift in the planting period to coincide with ideal rainfall periods, improved soil water utilization, and N uptake by the crops ultimately resulted in higher production and reduced environmental impacts.

5. Conclusions

Upland rice aboveground biomass, yields, harvest index, and N uptake increased with an increase in the N application rate. The maximum crop performance for productivity and N utilization was observed at N90 under moderately delayed planting (PW2). The model calibration with the observed data from the highest N application treatment (N90) resulted in improved model predictions, and there was a good agreement between the simulated and observed parameters, including grain yield, aboveground biomass, harvest index, grain and total plant N uptake, seasonal evapotranspiration, and soil profile nitrate. An acceptable range (good to excellent) of statistical indicators, including RMSEn (<30%) and d-index (approaching unity), indicated that the EPIC model was able to predict the observed results well. The water and N balance components were greatly influenced by the impact of PW as compared to the N application rates. Therefore, it is important to adopt an optimal PW and schedule N applications accordingly. In a comparison of PW1 and PW3, the maximum net N mineralization, reduced nitrate loss in surface runoff, and reduced nitrate leaching were observed under PW2. Similarly, the highest evapotranspiration, reduced surface runoff, and deep percolation losses were also observed at PW2. The EPIC model was also able to simulate fluctuations in soil water contents. Considering the upland rice productivity and soil water and N dynamics, PW2 was identified as the most suitable planting period for rainfed upland rice production. Climatic conditions and seasonal rainfall patterns were highly variable over the years. Therefore, considering the upland rice performance results to the planting window, soil water, and N dynamics and model predictions, it is recommended that the N application should be performed according to the adopted planting windows, and the EPIC model could be utilized in predicting the most suitable seasonal planting period and N application rate under the predicted climatic conditions. In addition, it should be noted that the EPIC model indicated a higher sensitivity to crop parameters; therefore, crop parameters should be adjusted, particularly for diverse genotypes, prior to the model’s application.