Objectives To address the core issues that the fixed nutrient supply mode in traditional protected tomato cultivation fails to match the dynamic nutrient demands of crops and that the light-temperature coupled water and fertilizer supply model is lacking, this study aimed to optimize the VegSyst model and establish a dynamic nutrient management scheme suitable for tomatoes grown in Chinese greenhouses.

Methods The experiment was conducted from 2024 to 2025 in large-scale multi-span greenhouses for intensive tomato production. Three cropping cycles were set up, with 11 small-fruited tomato varieties and 12 large-fruited tomato varieties selected. The cultivation adopted the coconut coir substrate mode combined with an integrated precision water and fertilizer irrigation system. Environmental data (photosynthetically active radiation, PAR; temperature), plant growth indicators (plant height, leaf area, dry matter production, DMP) and nutrient uptake (N, P, K, Ca, Mg, S) of tomatoes were systematically monitored throughout the entire growth period. Data from the first cropping cycle were used for model parameter calibration, those from the second cycle for model construction, and those from the third cycle for model verification. Key parameters of the VegSyst model were calibrated by integrating thermal time (accumulative temperature) and canopy PAR interception data, and variety-specific nutrient dilution curves were established. Combined with the monitoring data of electrical conductivity (EC), pH value and ion concentration of the drainage solution, a closed-loop nutrient regulation strategy of “model prediction–drainage monitoring–dynamic regulation” was constructed. The model performance was evaluated using four indicators, namely root mean square error (RMSE), relative error (RE), Willmott’s consistency index (d) and coefficient of determination (R²).

Results The cumulative thermal time required for large-fruited tomatoes to reach maximum PAR interception (CTT_f = 1382℃·d), maximum PAR interception rate (f_f = 0.968) and radiation use efficiency (RUE = 4.33 g/MJ) were all higher than those of small-fruited tomatoes (CTTf = 1276℃·d, ff = 0.947, RUE = 3.36 g/MJ). All nutrient dilution curves fitted the power function relationship (%Nutrient = aDMP^b). Specifically, the nitrogen dilution curves for small-fruited and large-fruited tomatoes were%N = 5.26DMP^−0.24 (R²=0.93) and%N = 4.75DMP^−0.16 (R² =0.86), respectively. The model exhibited excellent precision in simulating DMP and the uptake of N, P, K, Ca, Mg and S, with all evaluation indicators meeting the acceptance criteria for agricultural simulation models (RE ≤ 0.19, d≥ 0.964, R²≥ 0.871). Based on the weekly nutrient supply recommendations generated by the model and the closed - loop regulation strategy, precise nutrient supply was achieved. Compared with the traditional management mode, the nutrient input of small - fruited and large - fruited tomatoes was reduced by 53.6%–82.0% and 38.5%–77.5%, respectively. In addition, the nutrient use efficiency of small - fruited tomatoes was significantly increased from 26.7%–40.0% under the traditional mode to 41.7%–87.0%, and that of large - fruited tomatoes was significantly raised from 26.3%–42.3% to 43.0%–62.3%. A linear positive correlation was observed between the economic yield and nutrient uptake of both small - fruited and large - fruited tomatoes, which provided a quantitative basis for predicting nutrient input based on target yields.

Conclusions Based on the growth and climate monitoring data of large-fruited and small-fruited greenhouse tomatoes cultivated locally, this study completed the systematic calibration of key parameters of the VegSyst model by adopting the stepwise calibration method, and successfully realized the localization adaptation of this model in the soilless cultivation scenario of greenhouse tomatoes in China. After calibration, the simulation accuracy of the model for dry matter production and the absorption dynamics of N, P, K, Ca and Mg is higher than the standard for production application, and the prediction accuracy for the absorption dynamics of S also meets the production requirements. This model can provide core model support for the construction of the closed-loop nutrient management scheme of “model prediction - return flow monitoring - dynamic regulation”. Subsequent research needs to further verify the universality of the model and promote its in-depth integration with intelligent decision-making systems.