Calibration and Validation of VegSyst-CH Model to Manage Water and Nitrogen for Open-Field Lettuce in North China

Abstract

In the cold and arid regions of northern China, efficient water and nitrogen (N) management is critical for the sustainable production of leafy vegetables. Simplified models that estimate crop N and water transpiration demands using simple inputs based on climate parameters become an important method for making precise suggestions on N and irrigation application at a regional scale. This study developed and validated a regionally adapted version of the VegSyst model, named VegSyst-CH, based on a multi-year open-field experiment from 2021 to 2023. Model parameters were calibrated using data from the 2021 growing season and validated with independent datasets from 2022 and 2023. A critical N concentration (CNC) curve was established to describe the relationship between biomass accumulation and N content. VegSyst-CH, with a radiation use efficiency of 1.94 g MJ⁻¹, demonstrated high simulation accuracy for crop growth. The model showed a good predictive performance of N uptake under medium (N1) and high (N2) N treatments, with coefficients of determination (R²) above 0.80 across years and normalized root mean square error (NRMSE) values generally below 30%. The VegSyst-CH model also showed high accuracy in simulating crop evapotranspiration (ETc) over three consecutive growing seasons (2021–2023), with the dynamic trends of cumulative ETc closely aligning with measured values and the coefficients of determination (R²) consistently exceeding 0.90. These results validate the model’s robustness and applicability across different years. In conclusion, the VegSyst-CH model has strong spatiotemporal regulation capacity and climatic responsiveness, offering a robust decision support tool for precision fertilization and irrigation in open-field lettuce production in cold and arid regions.

1. Introduction

China is the world’s largest vegetable producer, where open-field cultivation is the dominant system representing 80% of the total vegetable-growing area and 65% of production [1]. Notably, the average N rate for all crops in China is 329 kg/ha, while for open-field vegetables, it is generally 2.6 times higher [2,3]. Experience-based water and fertilizer management practices make Chinese vegetable production a ‘hotspot’ for N inputs and losses to the environment, and also seriously hinder the country’s agricultural green transformation [4,5,6]. The cool and dry summer climate in northern China favors high-quality lettuce production and supports large-scale open-field cultivation [7,8]. However, limited in-season rainfall makes lettuce highly dependent on irrigation and well-timed N supply, while groundwater over-exploitation and excessive fertilization remain major challenges [9,10,11]. Although crop models driven by climate data, such as WOFOST and AquaCrop, are widely used in crop decision support systems (DSS) [12,13,14,15], their application in vegetable systems is still limited due to insufficient crop and region-specific calibration.

Recently, several user-friendly tools have been developed for practical vegetable fertigation management, such as the VegSyst model [16] and CropManage [17]. The VegSyst model is a crop growth simulation model based on thermal time and the fraction of photosynthetically active radiation intercepted (fi-PAR) by the crop. It simulates daily dry matter production (DMP), crop nitrogen uptake, and crop evapotranspiration (ETc) based on cropping dates and weather data of the Mediterranean region of Spain [18]. This model has been calibrated and evaluated for greenhouse tomato, sweet pepper, and open-field spinach and lettuce [19]. A major advantage of the VegSyst model is its minimal input data requirements, making it easy to use for farmers. Once regionally calibrated and validated, VegSyst can be integrated into a DSS as the core decision engine. Based on local weather inputs and crop status, VegSyst-DSS provides precise recommendations for irrigation volumes and amounts of applied N, facilitating on-farm adoption.

In the northern Chinese provinces of Hebei, Shanxi, and Inner Mongolia, with millions of hectares of vegetables, developing a region-specific version of the VegSyst simulation model is of major scientific and practical importance. Because of the appreciable differences in climate, cropping system and variety between northern China and the Mediterranean region, local calibration and validation of the VegSyst model is required [20,21]. The adoption of these model-based technologies to enhance the precision and efficiency of irrigation and fertilizer management will appreciably improve water and nitrogen use efficiency, reduce environmental impacts, and promote the sustainable intensification of vegetable production in this region. Here we conducted a three-year experiment to calibrate and validate the VegSyst model for lettuce grown in North China. The main objectives of this study were, for lettuce in North China’s conditions: (i) to derive parameters for canopy PAR interception, crop growth, water use, and the critical N dilution curve, using field measurements; (ii) quantifying model robustness through independent validation across multiple seasons and contrasting N management regimes; and (iii) to provide practical VegSyst-CH parameterization that can be embedded into decision support workflows to support data-driven irrigation and N application scheduling for open-field vegetable systems in this region.

2. Materials and Methods

2.1. Experiment Site

The experiment was conducted at the Yunong vegetable farm in Guyuan County, Hebei Province, China (41°38′20″ N, 115°37′35″ E). The site is located on a plateau at an altitude of 1420 m, with a cool temperate continental grassland climate. The average annual precipitation is approximately 400 mm, with more than 60% falling between June and August. The average air temperature during the lettuce-growing season (June–August) is approximately 17.9 °C.

Before transplanting, soil samples were collected from the 0–20 cm layer. The soil was classified as sandy loam with slightly alkaline reaction and moderate fertility. Prior to transplanting, five soil samples were randomly collected from the 0–20 cm plow layer across the experimental field. The samples were analyzed for the following properties: pH (1:2.5 w/v, soil:water), electrical conductivity (EC) (1:5 w/v, soil:water), bulk density (ring-knife method), alkali-hydrolyzed nitrogen (alkali diffusion method), available phosphorus (Olsen method), exchangeable potassium (sodium acetate extraction), and soil organic matter (potassium dichromate oxidation). All analytical procedures followed the methods described by Bao [22]. The soil had 18.28 g/kg organic matter, 79.98 mg/kg alkali-hydrolyzed nitrogen, 31.33 mg/kg available phosphorus (Olsen-P), and 161.81 mg/kg available potassium (exchangeable K), with a pH of 8.51 and a bulk density of 1.36 g/cm³.

2.2. Crop

The experimental field is part of a long-term open-field vegetable trial from 2021 to 2023, with one crop for each year. Iceberg lettuce (Lactuca sativa L., cv. “Shooter 101”), a variety commonly grown in Guyuan County in Hebei Province, was used in this study. Raised beds 80 cm wide and 15 cm high, covered with plastic mulch, were established. Each bed contained two plant rows, with a planting density of 6.67 × 10⁴ plants ha⁻¹. Lettuce seedlings, approximately four weeks old, were transplanted in early June each year. Surface drip irrigation was used throughout the growing period, and all top-dressed fertilizers were applied via a fertigation system. Basal fertilizers were incorporated into the soil prior to transplanting.

The experiment included three N fertilization treatments: N0 (no N applied, 0 kg/ha), N1 (low N rate, 180 kg/ha), and N2 (high N rate, 360 kg/ha). The plot size was 160 m² (8 m × 20 m), and a randomized block design with three replicate plots for each treatment was used. Basal fertilizers included both chemical and organic sources. An NPK compound fertilizer (15–15–15; N–P₂O₅–K₂O) was applied at a rate providing 90 kg/ha N, 90 kg/ha P₂O₅, and 90 kg/ha K₂O to the N1 and N2 treatments. In addition, organic fertilizer (N–P₂O₅–K₂O = 1.8–1.3–1.7%) was incorporated at 6.0 t/ha in 2021 and 3.0 t/ha in 2022 and 2023, supplying approximately 108 kg/ha N, 78 kg/ha P₂O₅, and 102 kg/ha K₂O in 2021, and 54 kg/ha N, 39 kg/ha P₂O₅, and 51 kg/ha K₂O in 2022–2023. The N0 treatment received no N input. To ensure a balanced supply of phosphorus and potassium in the N0 treatment, calcium superphosphate (18% P₂O₅) and potassium sulfate (52% K₂O) were applied as basal fertilizers before transplanting at rates equivalent to those used in the N1 and N2 treatments. The total phosphorus and potassium application rates were kept consistent across all treatments, each at 225 kg/ha.

Top-dressed fertilizer was applied using fertigation through drip irrigation with urea ammonium nitrate solution (UAN), ammonium polyphosphate (APP) and potassium dihydrogen phosphate (KH₂PO₄). Top-dressing occurred three times during each crop, at 10, 20 and 30 days after transplanting, with applications of 15%, 50% and 35% of the total top-dressed N amount (90 kg/ha and 270 kg/ha for N1 and N2 treatments, respectively). The total irrigation amounts in 2021, 2022, and 2023 were 83.3 mm, 200 mm, and 245 mm, respectively. The irrigation amounts and frequencies were determined based on routine irrigation practices of the experimental farm.

Plant sampling was performed at five representative growth stages: seedling, vegetative, rosette, heading, and harvest, at approximately 11, 21, 31, 41 and 52 days after transplanting, with small differences between years. At each stage, five representative plants were randomly selected per plot. Fresh weight was measured in the field, and dry matter production (DMP) was determined after oven-drying samples at 105 °C for 30 min to deactivate enzymes, followed by 70 °C until constant weight (typically 12–24 h, depending on sample size). The dried samples were ground, digested using the H₂SO₄-H₂O₂ method, and analyzed for total N concentration (N%) by the Kjeldahl method [22].

Above-ground nitrogen uptake (N_a, kg·ha⁻¹) was calculated as:

N_a = N_conc (%) × DMP (kg ha⁻¹)

(1)

2.3. Climate Data Collection

A portable microclimate automatic weather station with a CR1000 data logger (Campbell Scientific, Inc., Logan, UT, USA) was installed in the experimental field to continuously monitor environmental and soil conditions. The weather station recorded air temperature, relative humidity, solar radiation, rainfall, and wind speed, as well as soil moisture measured with a TDR (CS655, Campbell Scientific, Inc.) monitoring system. The TDR system was equipped with 20 cm long probes, positioned vertically, to measure soil volumetric water content and determine changes in soil water storage (ΔW) within the 0–20 cm root zone. All sensors were recorded every 30 min, and daily averages or integrated values were calculated. These high-frequency meteorological and soil datasets were then used as essential inputs for the VegSyst-CH model to simulate daily photosynthetically active radiation (PAR), reference evapotranspiration (ETo), and crop water demand. For example, solar radiation and daily air temperature were used to estimate photosynthetically active radiation (PAR) and cumulative thermal time (CTT), which were subsequently applied to calculate radiation use efficiency (RUE). In addition, daily relative humidity, wind speed, and rainfall were employed to estimate reference evapotranspiration (ETo), water balance, crop water requirements, and actual crop water consumption [23].

Daily mean air temperature and rainfall data from 2021 to 2023 were collected and presented in Figure 1. The average air temperatures and total rainfall during these three growing seasons were 18.0 °C and 179 mm (2021), 18.4 °C and 115 mm (2022), and 20.0 °C and 77.1 mm (2023), respectively. No notable incidences of pests or diseases were observed during the three cropping periods.

2.4. Calibration and Validation of the VegSyst Model

The original VegSyst simulation model (V1) was developed by [16,24] in Spain for greenhouse-grown vegetable crops such as tomato, cucumber, and pepper, and was later integrated into the VegSyst-DSS decision support system [25]. Subsequently, an adaptation known as VegSyst-Outdoors was developed for open-field vegetable crops such as lettuce [19]. In the current study, the VegSyst-Outdoors model was calibrated and validated for open-field iceberg lettuce under Chinese agro-environmental conditions. From this point forward, the adapted version is referred to as VegSyst-CH. Data from the 2021 growing season were used for both calibration and validation, while data from the 2022 and 2023 seasons were used for independent validation of the model.

Table 1. Symbols, units, and description of parameters in the VegSyst-CH model.

Symbol	Units	Description
Tbase	(°C)	Base temperature
fi-PAR		Fraction of photosynthetically active radiation (PAR) intercepted by the crop at day i
f0		Fraction of PAR intercepted by the crop at transplanting or emergence
ff		Maximum fraction of PAR intercepted by the crop
RTTi		Relative thermal time at day i
RTT0.5		Relative thermal time at which fi-PAR = 0.5 × (f0 + ff)
CTTi	(°-day)	Cumulative thermal time at day i
CTTf	(°-day)	Cumulative thermal time at maximum PAR interception
CTTmat	(°-day)	Cumulative thermal time at maturity (end of the crop)
fi-SR		Fraction of solar radiation (SR) intercepted by the crop at day i
ff-SR		Maximum fraction of SR intercepted by the crop
α		Equation fitting coefficient (shape coefficient)
DMPi	(g m−2)	Above-ground dry matter production at day i
RUE	(g MJ−1 PAR)	Radiation use efficiency
kcini		Initial crop coefficient
kcmax		Maximum crop coefficient
kcend		Crop coefficient at maturity (end of the crop)
a		N content in dry biomass when DMP = 1 t ha−1
b		Equation fitting coefficient
PARi	(MJ m−2 d−1)	Daily PAR intercepted by the crop at day i
SRi	(MJ m−2 d−1)	Daily integral of solar radiation at day i
ETc	(mm d−1)	Daily crop evapotranspiration
Ncrit	(%)	Critical N content

lists the symbols and their definitions used throughout the paper.

2.4.1. Crop Growth Parameter Calibration

(1)

PAR VegSyst simulates the fraction of intercepted photosynthetically active radiation (f_i-PAR) from thermal time, which is estimated from the average of daily maximum and minimum air temperatures following the methodology described by [19]. Daily intercepted PAR (PAR_i) was calculated from daily values of f_i-PAR and the integral of daily PAR which was obtained from the product of the integral of daily solar radiation. The detailed procedures and equations are listed in

Table 2. Model calculation equations and instructions.

Equation	Instructions
PARi = 0.45 × SRi × fi-PAR	Daily incident PAR was assumed to be 45% of measured solar radiation (SRi).
RUE = DMPi/PARi	RUE values were determined as the slope of the linear relationship between experimental values of intercepted PAR and DMP during the cropping season.
ETc = EToi × kci	ETo was estimated using the FAO Penman–Monteith equation.
fi-SR = 1 − exp [ln(1 − fi-PAR)/1.4]	1.4 is the ratio between the extinction coefficients of PAR and solar radiation-based
kci = 1 + (kcmax − 1) × (fi-SR/ff-SR)	-
Nc = a × DMPb	a and b are species-specific coefficients fitted from field data

.

Daily intercepted PAR (PAR_i, MJ m⁻²) was calculated as the product of incident PAR, which was considered to be 45% of measured solar radiation (SR_i), and the simulated fraction of intercepted PAR (f_i-PAR):

PAR_i = 0.45 × SR_i × f_i-PAR

(2)

(2)

RUE The original VegSyst-Outdoors and VegSyst-CH models calculate dry matter production (DMP, g m⁻²) for each day (i) as the product of PAR_i (MJ m⁻²) and radiation use efficiency (RUE):

RUE = DMP_i/PAR_i

(3)

where RUE (g MJ⁻¹) is the radiation use efficiency. For lettuce, RUE values were determined as the slope of the linear relationship between experimental values of intercepted PAR and DMP during the cropping season. When there were different slopes of this relationship during the season, the maximum slope was used. The measured DMP data for each growth stage were obtained from the plant sampling described in Section 2.2.

(3)

CTT Cumulative thermal time (CTT) is the accumulated value of temperature (daily average temperature values) up to a given day, obtained from the CAMPBELL portable microclimate automatic weather station.

2.4.2. N Parameter Calibration

N critical curve A critical crop N content (N_crit), defined as the minimum total crop N content (N%) associated with maximum DMP, at a given time, was used to simulate critical crop N uptake. Data from the 2021–2023 experiments were pooled to derive the CNC relationship. Following Liang et al. [26], N_crit points were selected from observations under non-N-limiting conditions (mainly the higher N treatments) by constructing the upper boundary of the relationship between crop N concentration and accumulated above-ground DMP. N_crit is a necessary parameter for modifying the VegSyst model across different ecological zones. Critical crop N uptake (kgN ha⁻¹) was calculated as the product of DMP (kg ha⁻¹) and N_crit. Critical crop N contents were used to obtain critical N content values from the curve of accumulated above-ground dry matter production.

N_crit = a × DMP^b

(4)

Parameter a represents the crop N content in dry biomass when DMP equals 1 t ha⁻¹, whereas b is the coefficient obtained from the equation fitting process.

2.4.3. Water Parameter Calibration

(1)

Crop evapotranspiration (ETc) Soil water storage dynamics were continuously monitored using the time-domain reflectometry (TDR) system installed with the experimental meteorological station. This approach follows the methodology of [27,28], where TDR measurements were used in combination with the soil water balance principle to estimate field crop evapotranspiration (ETc). Daily crop evapotranspiration (ETc) was calculated using the water balance equation:

ETc = P + I − R − D − ΔW

(5)

where P is precipitation (mm), I is irrigation (mm), R is surface runoff (mm), D is deep drainage or downward flux across the lower boundary (mm), and ΔW is the change in soil water storage (mm).

The experimental plots were flat and managed with a drip irrigation system to not exceed field capacity. Under such conditions, surface runoff (R) was negligible because irrigation was applied slowly and uniformly. Deep drainage (D) was assumed to be negligible because irrigation maintained soil moisture within the root zone without exceeding field capacity. With a deep groundwater table, no significant upward flux occurred at the lower boundary. Therefore, when the soil profile covers the active root zone and water inputs are moderate, the lower boundary can be reasonably treated as a zero-flux boundary, as commonly applied in field evapotranspiration studies [29,30].

(2)

Reference evapotranspiration (ETo) and crop coefficient (kc) Reference evapotranspiration (ETo) was estimated from meteorological data using the FAO Penman–Monteith equation [31]. The daily crop coefficient (kc) values were then derived as:

$$\mathrm{kc}={\displaystyle \frac{\mathit{ETc}}{\mathit{ETo}}}$$

(6)

The observed ETc and kc values were subsequently compared with VegSyst-CH model simulations at 10-day intervals, corresponding to the soil water balance calculation frequency, to evaluate model accuracy across different growth stages.

Furthermore, crop evapotranspiration (ETc) is calculated by the model [16,24], considering: (a) daily values of reference crop evapotranspiration (ETo) estimated with the Penman–Monteith equation [31], and (b) daily kc values calculated from initial (kc_ini), maximum (kc_max) and final (kc_end) crop coefficient values based on the fraction of intercepted solar radiation (f_i-SR).

2.5. Statistical Evaluation

The datasets collected—including microclimate variables, intercepted PAR, ETc, biomass, and plant N concentration—were integrated into the VegSyst-CH model. Model performance was evaluated by comparing simulated and observed dry matter production, crop N uptake, and ETc using the pooled dataset across the three seasons and N treatments. The coefficient of determination (R²), root mean square error (RMSE), and normalized RMSE (NRMSE) were used as performance indicators. R² values were calculated with SPSS 20.0, and RMSE and NRMSE with Microsoft Excel. Statistical significance of linear regressions was assessed using two-sided t-tests in SPSS 20.0, with significance levels of p < 0.05, 0.01, and 0.001. Figures were prepared using Origin Pro 2021.

3. Results

3.1. VegSyst-CH Model Calibration

The key calibration parameters are listed in

Table 3. Parameters of calibration used for lettuce in the VegSyst-CH model. Lowercase letters “a” and “b” represent the fitted coefficients of the critical N dilution curve (Nc = a × DMP^b).

Crop Growth Parameters	Values	Critical N Curve Parameters	Values	Key Water-Use Parameters	Values
CTT0 (°-day)	0	a	3.2219	kcini	0.41
CTTmat (°-day)	655	b	−0.394	kcmax	1.01
RUE (g MJ−1 PAR)	1.94			kcend	0.97

. Three crop growth parameters, three water use parameters, and two critical N curve parameters were recalculated. Model calibration was performed to improve prediction accuracy using comprehensive datasets from the 2021 growing season, including meteorological observations (air temperature, relative humidity, solar radiation, rainfall, and wind speed), as well as field measurements of crop growth, evapotranspiration, and plant N concentration. As the same lettuce cultivar type was used throughout this study and differences in growth characteristics among cultivars were minimal, only the 2021 dataset was used for calibration.

Based on previous studies, the base temperature for lettuce was set at 4 °C. Simulation analysis revealed that the cumulative thermal time (CTT) effectively captured the crop canopy’s capacity to intercept photosynthetically active radiation (PAR). In this study, the fraction of intercepted PAR (fi_-PAR) of the model increased with CTT and closely matched the observed data (Figure 2). The measured fi-PAR reached a maximum value of 0.8 at 655 degree days when the crop was harvested.

3.1.1. Dry Matter Production and N Uptake

There was good agreement between the simulated and measured DMP in the 2021 lettuce crop for the three treatments (Figure 3). The agreement was better for the N2 and N3 treatments, with slopes of the linear regression close to one and the associated R² values close to 0.90. For the N0 treatment, the measured DMP was less than the simulated ones in the early stages, with lower R² values of 0.79 (

Table 4. Linear regression between VegSyst-CH model simulations and measured DMP values.

Year	Treatment	n	Fitted Equation	R2	RMSE	NRMSE (%)
	N0	6	y = 0.980x − 0.392	0.789 *	0.445	21.6
2021	N1	6	y = 0.898x + 0.222	0.898 **	0.108	5.3
	N2	6	y = 0.827x + 0.659	0.887 **	0.357	17.5
	All	18	y = 0.901x + 0.163	0.882 **	0.335	16.3
	N0	4	y = 0.791x − 0.231	0.765 *	0.853	31.7
2022	N1	4	y = 0.926x + 0.159	0.948 **	0.159	5.9
	N2	4	y = 1.004x + 0.109	0.960 **	0.122	4.5
	All	12	y = 0.907x + 0.012	0.865 **	0.506	18.8
	N0	6	y = 0.962x − 0.533	0.888 **	0.657	22.2
2023	N1	6	y = 1.041x − 0.124	0.959 **	0.057	1.9
	N2	6	y = 1.055x − 0.141	0.924 **	0.112	3.8
	All	18	y = 1.019x − 0.266	0.921 **	0.386	13.1
Multi-year	N0	6	y = 1.055x − 0.141	0.902 **	0.647	25.2
	N1	6	y = 0.955x + 0.104	0.989 **	0.063	2.5
	N2	6	y = 0.968x + 0.227	0.970 **	0.152	5.9
	All	18	y = 0.942x − 0.015	0.899 **	0.385	15.0

Note: “n” represents average content at 10-day interval. * and ** indicate statistical significance at p < 0.05 and p < 0.01, respectively. Multi-year refers to the average data optimization across 2021, 2022, and 2023.

).

Using the established equations in Section 2.4.1, the radiation use efficiency (RUE) value calculated in this study was 1.94 g MJ⁻¹ (Table 3), lower than the RUE of ‘Crisphead’ lettuce 2.1 g MJ⁻¹ reported by Giménez [19], showing a small difference between the different regions.

3.1.2. N Dilution Curve

To precisely simulate the relationship between N content and DMP, a critical N concentration curve (CNC) with a power function was determined. The two key parameters of the power-law equation used in CNC modeling are presented in Table 3 and Figure 4. It showed a good relationship between DMP and plant N concentration for open-field lettuce in northern China (R² = 0.726, p < 0.01).

3.1.3. ETc and kc

In this study, field-measured crop evapotranspiration (ETc) was obtained using the soil water balance method based on high-frequency TDR measurements. The measured ETc values showed good agreement with model-simulated values (Figure 5). The corresponding daily crop coefficients (kc) were calculated for each growth stage based on measured ETc. The kc values were 0.41 during the initial stage, 1.01 at peak growth, and 0.97 in the late stage (Figure 6). These kc values were in close agreement with the model’s assumed parameters, exhibiting only a slight deviation at the harvest stage.

3.2. Validation of the VegSyst-CH Model

3.2.1. Validation of DMP

For all treatments over three years, the simulated DMP values from the VegSyst-CH model were compared with the measured DMP values using linear regression (Figure 7). There was consistently good agreement between simulated and measured DMP values across all data. The overall linear regression for all data had a coefficient of determination (R²) value close to 0.90, with a statistical significance at the p < 0.01 level. The slope was close to one, and the intercept values were close to zero.

For each year, the measured values from the N0, N1, and N2 treatments were fitted to the simulated DMP, resulting in R² values ranging from 0.76 to 0.96, indicating significant differences between the N0 treatment and the N1 and N2 treatments in 2021 and 2022. Over the average of three years, R² ranged from 0.90 to 0.97, and RMSE ranged from 0.063 to 0.65; no significant difference occurred between treatments.

3.2.2. Validation of Crop N Uptake

The VegSyst-CH model simulates N uptake based on key environmental variables, including solar radiation, air temperature, and air humidity. To evaluate the model’s accuracy, the simulated N uptake values were compared with field-measured crop N uptake data. Figure 8 presents the comparisons between the VegSyst-CH model simulations and field measurements for three different N treatments for the three growing seasons of 2021, 2022, and 2023. In general, the model had good simulation performance for the N1 and N2 treatments, except at the harvest stage in 2021 (Figure 8a). For the N0 treatment, the simulated values were higher than the measured values (Figure 8b,c) which is to be expected because the model simulations assume that the crop is not N-deficient while the N0 treatment was N-deficient. In addition, in the first experimental year there was high residual soil mineral N from a preceding commercial crop. Interannual climatic differences may have contributed to this discrepancy. Overall, across the three years, the model exhibited consistent and satisfactory simulation of crop N uptake (Figure 8d and Figure 9), resulting in an R² of 0.8715 and a significant p-value of <0.01 for the linear regression model.

The linear regression analysis results for the simulated and measured crop N uptake, by year and across years, are summarized in

Table 5. Linear regression between VegSyst-CH model simulated and measured N uptakes.

Year	Treatment	n	Fitted Equation	R2	RMSE	NRMSE (%)
	N0	9	y = 0.734x + 7.814	0.848 **	13.68	25.9
2021	N1	9	y = 1.688x − 32.419	0.838 **	14.88	28.2
	N2	9	y = 1.556x − 30.093	0.883 **	11.36	21.5
	All	27	y = 1.292x − 18.233	0.788 **	13.38	25.4
	N0	9	y = 0.423x + 2.678	0.909 **	24.79	40.3
2022	N1	9	y = 0.789x + 3.924	0.918 **	8.48	20.6
	N2	9	y = 0.940x + 6.262	0.899 **	8.06	19.6
	All	27	y = 0.717x + 4.288	0.806 **	15.82	28.5
	N0	9	y = 0.515x + 9.159	0.851 **	19.55	29.4
2023	N1	9	y = 0.424x + 17.803	0.832 **	17.40	25.1
	N2	9	y = 0.658x + 32.190	0.811 **	18.05	26.4
	All	27	y = 0.532x + 19.717	0.743 *	18.35	27.0
Multi-year	N0	9	y = 0.515x + 12.254	0.828 **	15.23	34.2
	N1	9	y = 0.755x + 13.707	0.954 **	9.43	11.6
	N2	9	y = 0.872x + 19.251	0.863 **	14.49	26.6
	All	27	y = 0.714x + 15.070	0.872 **	15.17	18.2

Note: “n” represents average content at 10-day interval. * and ** indicate statistical significance at p < 0.05 and p < 0.01, respectively. Multi-year refers to the average data optimization across 2021, 2022, and 2023.

. The results indicated that significant R² values (p < 0.01) were found between simulated and measured values for all treatments in 2021 and 2022, while slightly lower but still significant linear relationships (p < 0.05) were observed in 2023. The coefficient of determination (R²) values exceeded 0.80 across all treatments and years, indicating a strong agreement between the simulated and measured data. Moreover, except for the N0 treatments from 2022 and the multi-year assessment, the normalized root mean square error (NRMSE) values were consistently below 30%, suggesting acceptable simulation performance. The multi-year averages of R², RMSE, and NRMSE further demonstrated the robustness and reliability of the VegSyst-CH model in simulating crop N uptake in lettuce under N1 and N2 treatments across different years (Figure 9).

3.2.3. Crop Evapotranspiration (ETc) Validation

The accuracy of the VegSyst-CH model for simulating crop evapotranspiration (ETc) was assessed using linear regression of simulated and measured values. When taking the three years together, the linear equation relation had a R² value of 0.91 with a significance level of <0.001, and the slope close to one and the intercept values close to zero, which demonstrated very good agreement and high accuracy (Figure 10). In each of the years 2021, 2022 and 2023, simulated daily ETc was further evaluated (

Table 6. Linear regression between VegSyst-CH model simulated and measured ETc.

Year	n	Fitted Equation	R2	RMSE	NRMSE (%)
2021	5	y = 0.928x − 0.251	0.951 **	0.642	15.1
2022	4	y = 0.979x − 0.146	0.957 *	0.407	8.9
2023	4	y = 0.923x + 0.401	0.923 **	0.461	9.8
Multi-year	13	y = 0.981x − 0.142	0.907 ***	0.524	11.7

Note: “n” represents soil moisture average content at 10-day interval; *, **, and *** indicate statistical significance at p < 0.05, p < 0.01, and p < 0.001, respectively.

). The regression slopes ranged from 0.92 to 0.98, indicating a consistent but small underestimation of ETc, and R² was above 0.9 with a different significance level, showing the best performance in 2021 and 2023. Overall, the model demonstrated reliable performance in simulating daily ETc across multiple seasons, supporting the potential application of the model for water management and irrigation scheduling in similar agroecological zones.

4. Discussion

Precise N and water management are crucial measures to alleviate the pressure from water resource shortages and to support environmental protection at a regional scale. This study calibrated and validated the VegSyst-CH model for simulating DMP, N uptake, and ETc of open-field lettuce production under the cold–dry climatic conditions of northern China. Using three years (2021–2023) of environment field data under different N levels, we demonstrated that VegSyst-CH provides reliable simulation of crop N demand and water use dynamics. The average coefficient of determination values (R²) over the three years were 0.90 for DMP, 0.87 for N uptake and 0.91 for ETc at a significance level of 0.01, indicating strong model adaptability and stability under local field conditions. Our findings highlight the importance of the VegSyst-CH model for the study area (Bashang) to precisely manage N and water application.

4.1. Local Paramaterization of the VegSyst Model and Regional Applicability

The calibrated CNC for iceberg lettuce in northern China was N_c = 3.2219 × DMP^−0.394, which was slightly different from the previously reported equation under Mediterranean conditions [19], where the slope parameter was steeper and initial N requirements were higher. This discrepancy can be attributed to climatic differences between regions. Lower temperatures, shorter photoperiods, and greater diurnal variation in northern China limit canopy expansion and biomass accumulation, resulting in a reduced N requirement per unit of dry matter compared to greenhouse and open-field crops in Spain. These findings emphasize the importance of developing region-specific CNC functions when applying VegSyst in open-field systems with contrasting environmental conditions.

The calibrated radiation use efficiency (RUE) in the VegSyst-CH model was 1.94 g MJ⁻¹, lower than the 2.8–3.0 g MJ⁻¹ typically observed in the Mediterranean area [32]. This difference likely results from the complex meteorological variability in open-field conditions, especially in high-latitude cold regions, where radiation, temperature, and relative humidity often deviate from optimal ranges, limiting photosynthetic efficiency. Nevertheless, this RUE value aligns well with measured values for open-field lettuce in high-latitude cold regions. For example, Bhattacharya [33] reported an RUE range of 1.6–2.2 g MJ⁻¹ for open-field lettuce under similar environmental conditions, and [19] reported an RUE of 2.1 g MJ⁻¹ for a similar lettuce type grown in open-field conditions in Spain. These comparisons verify the regional representativeness of the VegSyst-CH calibration results.

The VegSyst-CH model employed a single growth period fi-PAR–thermal time (CTT) growth curve, reducing the number of fitted parameters and improving its practical applicability. The simulation results indicated a maximum fi-PAR value of 0.80, suggesting that the model effectively captures the canopy PAR interception capacity. This finding aligns with the conclusions of VegSyst applications in open-field crops in Spain [19]. This simplification enhances the model’s utility under farm conditions where frequent destructive sampling is impractical. These results demonstrate the strong regional adaptability of VegSyst-CH. It achieves predictive accuracy comparable to international studies (e.g., Spain) while using readily available climatic inputs and minimal parameters.

Regarding N dynamics, VegSyst-CH effectively simulated crop N uptake under moderate (N1) and high (N2) N inputs, but overestimated N uptake under N-deficient conditions (N0 treatment). These deviations are clearly reflected in Figure 8. This bias arises because VegSyst assumes optimal N availability and does not explicitly account for physiological responses to N limitation, such as reduced leaf area development, lower chlorophyll concentration, and decreased photosynthetic capacity. Similar results have been reported in other VegSyst studies [23]. In addition, it should be noted that N0 does not necessarily represent “zero N supply” in the field. In the first experimental year (2021), residual soil fertility may have been relatively high due to historical over-fertilization by local farmers, and soil N mineralization can further contribute to plant-available N, leading to measured N uptake patterns that differ from those expected under strictly N-free conditions. Moreover, interannual climatic variability (e.g., differences in temperature, rainfall, and radiation during 2021–2023) can modify growth rates and the timing of N demand, contributing to stage-specific discrepancies between simulations and observations. This issue is particularly pronounced in open-field systems due to greater uncertainties in N supply and stronger soil N mineralization fluctuations [26]. Future improvements could incorporate dynamic stage-specific parameters or integrate crop nutritional diagnostics to enable “corrective management” and enhance the precision of model-based recommendations [34,35].

The calibrated VegSyst-CH model effectively captured the water use characteristics of lettuce under local open-field and drip irrigation conditions. Compared to the FAO-56 [31] standard kc values for lettuce (0.45–0.50, ~1.0, ~0.9), the observed kc values (0.41, 1.01, and 0.97) showed similar temporal patterns but slightly lower variability, reflecting adaptations to the cooler and drier climate and the precise irrigation scheduling typical of northern China. These values are also consistent with those reported for the Bashang region of Zhangjiakou in China by [36], who found that local kc values for open-field vegetables were generally lower than the FAO-56 standards due to reduced evaporative demand and the adoption of sub-mulch drip irrigation practices. The simulated daily ETc values showed a strong agreement with field measurements, with R² ranging from 0.92 to 0.96 across years, confirming that VegSyst-CH effectively captured the seasonal pattern of water consumption in lettuce under open-field drip irrigation.

4.2. Model Improvement and Potential Implications

In terms of practical application, VegSyst-CH demonstrates both spatial and temporal adaptability. Spatially, the model uses locally measured climate data that are generally consistent within small agricultural regions. This relative climatic homogeneity means that a locally calibrated VegSyst-CH model can be extended to surrounding farms within the same agroecological zone. Thus, the model offers a feasible path for scaling precision management from plot-level applications to regional-level deployment, potentially supporting zonal irrigation and fertilization policy planning.

A key challenge for direct day-to-day scheduling is that VegSyst-CH requires meteorological inputs that are not known in advance, which limits its use for real-time future fertigation decisions without weather forecasts. In practice, VegSyst-CH is most suitable for (i) in-season diagnosis and adjustment using observed weather up to the current date, and (ii) pre-season planning using representative weather scenarios. To support pre-season planning, we constructed an “average climate year” based on three years of observations, which provided reasonable predictions of seasonal N uptake and water demand and showed a strong agreement with measurements. This suggests that in regions with relatively stable interannual variability, average climate scenarios can guide seasonal fertilizer and irrigation planning when real-time data are unavailable.

Importantly, the current VegSyst-CH framework cannot (a) provide accurate real-time scheduling for future dates without forecast meteorological inputs, and (b) fully capture crop physiological responses under severe N deficiency (as reflected by the N0 overestimation), highlighting the need for stress response modules or diagnostic corrections.

Further, some factors originating in field-specific conditions were not explicitly represented in the model and likely led to small deviations. For example, high wind speeds common in northern China’s open-field environments can enhance crop evapotranspiration beyond canopy-based estimations, while the use of plastic mulch can modify soil temperature and surface evaporation dynamics. Additionally, spatial heterogeneity in rainfall and infiltration across plots may have caused short-term discrepancies between observed and simulated ETc. These points offer clear directions for future refinement through the inclusion of parameters accounting for wind, mulching, and dynamic canopy cover.

Looking forward, VegSyst-CH could evolve into a powerful platform for adaptive resource management. With appropriate parameter adjustments, it can be expanded to other crops, mulch types, and cropping systems, eventually developing into a modular and universal decision support system (DSS). Integrating VegSyst-CH with UAV-based remote sensing, soil moisture and nutrient sensors, and local weather stations would enable rapid in-season recalibration and real-time decision-making. Previous studies have shown that combining modeling tools with sensor-based monitoring significantly improves input efficiency and sustainability [37,38]. In this context, VegSyst-CH has the capacity to serve not only as a field-scale tool for individual farm management, but also as a regional-level platform supporting climate-smart, site-specific irrigation and fertilization planning under variable environmental conditions. Future research should prioritize expanding parameter calibration across diverse systems and integrating VegSyst-CH with emerging digital agriculture technologies, ultimately supporting sustainable intensification and smart resource use at both farm and policy-making scales.

5. Conclusions

This study calibrated and validated the VegSyst model for open-field lettuce production in northern China, resulting in the regionally adapted VegSyst-CH version. The model showed strong accuracy in simulating DMP, N uptake, and crop evapotranspiration over three years of field trials. The calibrated parameters, including radiation use efficiency and the critical N concentration curve, aligned well with measured values under local environmental conditions. The VegSyst-CH model demonstrated high robustness and practical applicability, offering a valuable decision-making tool for improving water and N use efficiency in open-field vegetable systems. Its future application potential lies in supporting farm-scale precision management and, with further development, extending to regional-scale decision support for sustainable agricultural practices.