{"paper_id":"17a0a5d3-a937-40e0-b33f-e7d5506ddff5","body_text":"Radiation-Driven Prediction of Daily Irrigation Demand under \nDifferent Electrical Conductivity Scenarios in Greenhouse Tomato\nLingang Xiao¹, Yan Ma¹, Qian Feng², Xingdong Gao¹, Huifeng Shi¹, Xia Liu¹ and \nYilei Yin²,*\n¹ Institute of Agricultural Mechanization, Xinjiang Academy of Agricultural Sciences, \nUrumqi 830091, China\n² Institute of Facility Agriculture, Academy of Agricultural Planning and Engineering, \nMinistry of Agriculture and Rural Affairs, Beijing 100125, China\n*Correspondence: 276089613@qq.com\nAbstract: In soilless greenhouse tomato cultivation, daily transpiration and irrigation \ndemand are largely governed by solar radiation, while irrigation-solution electrical \nconductivity (EC) used for salinity management may further modulate plant water \nuse. This study developed a low-input, radiation-driven modeling approach to predict \ndaily irrigation demand under contrasting water–salt management scenarios. Two \ntomato cultivars were grown under four treatments: conventional baselines (CK1, \nCK2) and regulated scenarios combining irrigation volume with solution EC \n(low-water high-EC, TK; high-water moderate-EC, TC). Daily irrigation volume (I) \nand drainage were recorded, and daily cumulative radiation (G) was derived from \nphotosynthetically active radiation (PAR). Within each treatment, we compared a \nradiation-only baseline model with an EC-adjusted model and evaluated predictive \nperformance using 5-fold blocked time-series cross-validation. Results showed strong \npositive correlations between G and I across all treatments (p < 0.001). The \nEC-adjusted models achieved cross-validated root-mean-square errors (RMSE) of \n0.815–1.393 L d⁻¹ per trough and Nash–Sutcliffe efficiencies (NSE) of 0.407–0.730. \nIncorporating EC yielded a small but consistent improvement under the TK scenario \n(ΔRMSE = −0.014 L d⁻¹; ΔNSE = +0.019), whereas its effect was negligible or \nslightly negative under CK1, CK2, and TC, highlighting scenario dependence. Our \nradiation-driven framework, with an optional EC correction, offers a practical and \nscalable tool for daily irrigation forecasting and supports integrated water–salt \nmanagement in soilless greenhouse tomato production.\nKeywords: greenhouse tomato; irrigation demand; solar radiation; electrical \nconductivity; substrate cultivation; irrigation scheduling\n1. Introduction\n1.1. Greenhouse tomato production and irrigation challenges\nTomato (Solanum lycopersicum L.) is a globally significant protected \nhorticultural crop, whose economic viability hinges on yield stability, fruit quality, \nand resource-use efficiency [1,2]. With the expansion of greenhouse cultivation, \nmanagement strategies have evolved from a sole focus on yield toward integrated \napproaches that harmonize productivity, quality enhancement, and water conservation \n[3,4]. Nevertheless, greenhouse tomato systems remain heavily reliant on irrigation \n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted January 24, 2026. ; https://doi.org/10.64898/2026.01.23.701235doi: bioRxiv preprint \n\nand fertigation, rendering water management a critical—and often limiting—factor in \nthe context of escalating water scarcity and rising production costs [5–7].\nIn practice, irrigation scheduling in greenhouses frequently depends on grower \nexperience or fixed timetables, which seldom account for rapid daily fluctuations in \ncrop water demand. Suboptimal irrigation not only curtails water-use efficiency but \nalso perturbs root-zone salinity and nutrient dynamics [8,9], thereby influencing \nphotosynthetic performance, yield formation, and key fruit quality attributes such as \nsoluble solids and nutritional composition [10–12]. Hence, developing irrigation \nstrategies that dynamically respond to environmental variability while safeguarding \nyield and quality remains a pivotal challenge in modern greenhouse tomato \nproduction.\n1.2. Radiation-driven irrigation demand in greenhouse systems\nAmong environmental drivers, solar radiation exerts a dominant influence on \ncanopy transpiration and carbon assimilation in greenhouse tomatoes [13,14]. Daily \nvariations in photosynthetically active radiation (PAR) directly modulate stomatal \nconductance and transpiration rates, establishing radiation as a primary determinant of \nshort-term irrigation demand [15–17]. Unlike temperature and humidity, which can be \npartially regulated in controlled environments, incident radiation is largely governed \nby external weather conditions, leading to pronounced day-to-day variability in crop \nwater requirements [18,19].\nEmpirical relationships between daily cumulative radiation and irrigation water \nconsumption have been consistently demonstrated in greenhouse tomatoes, furnishing \na practical foundation for radiation-based estimation and scheduling approaches \n[13,20,21]. Compared with mechanistic evapotranspiration models or data-intensive \nmachine-learning techniques, empirical radiation-driven models offer distinct \nadvantages in simplicity, robustness, and operational feasibility—attributes especially \nvaluable for daily irrigation scheduling in commercial settings [22–25]. However, \nmost existing radiation-based studies have been conducted under uniform fertigation \nregimes, seldom explicitly addressing potential interactions with root-zone salinity \nmanagement [26].\n1.3. Electrical conductivity and water–salt interactions\nIn soilless and substrate-based greenhouse tomato systems, the electrical \nconductivity (EC) of the nutrient solution is a key operational variable that shapes the \nroot-zone water–salt environment [27–29]. Adjustments in EC alter osmotic potential \nand ion concentrations around roots, thereby influencing water uptake, transpiration, \nand assimilate partitioning [30,31]. Numerous studies indicate that moderate elevation \nof EC, or controlled deficit irrigation, can enhance fruit quality traits such as soluble \nsolids and sugar-acid ratio, though excessive salinity may suppress yield and overall \nplant water consumption [32–34].\nCritically, the effects of EC on crop water use are not independent of irrigation \nsupply level. Under differing water-availability backgrounds, identical EC conditions \nmay exert contrasting influences on transpiration and irrigation demand, owing to \nshifts in plant water status and root-zone hydraulic properties [35,36]. Recent work \n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted January 24, 2026. ; https://doi.org/10.64898/2026.01.23.701235doi: bioRxiv preprint \n\nemphasizes the importance of integrated water–salt regulation rather than treating EC \nas a static background parameter [37–40]. Despite this, most irrigation-scheduling \nframeworks continue to neglect EC-dependent adjustments in daily irrigation demand, \nparticularly in soilless cultivation systems.\n1.4. Research gap and objectives\nAlthough radiation-driven approaches are widely employed to estimate \ngreenhouse irrigation demand, prevailing frameworks rarely incorporate EC-related \nadjustments under contrasting water–salt management scenarios. This omission \nconstrains their applicability for integrated irrigation strategies that simultaneously \naim for water conservation and fruit quality improvement.\nTherefore, this study aimed to develop a simple, practical radiation-based \nmethod for daily irrigation demand estimation under varying electrical conductivity \nscenarios in substrate-grown greenhouse tomato. Specific objectives were to:(i) \nquantify the relationships between daily cumulative radiation and irrigation demand \nunder different EC management backgrounds;(ii) elucidate how EC modifies \nradiation-driven water-demand responses; and (iii) evaluate the implications of \nEC-adjusted irrigation scheduling for yield stability and fruit quality. \nBy emphasizing operational feasibility and employing treatment-wise blocked \ntime-series cross-validation, this work provides a management-oriented irrigation \nframework that complements more complex modeling approaches reported in the \nliterature.\n2. Materials and Methods\n2.1. Experimental Site and Environmental Conditions\nThe experiment was conducted from November 2023 to May 2024 in Glass \nGreenhouse A2 at the Facility Agriculture Research Base of the Academy of \nAgricultural Planning and Engineering, Ministry of Agriculture and Rural Affairs, \nlocated in Yongqing County, Langfang City, Hebei Province, China. The greenhouse \ncovered approximately 5000 m² and was equipped with automated \nenvironmental-control systems (internal/external shading, fan-pad cooling, heating, \nsupplemental lighting, and ventilation). The daytime temperatures were controlled \nbetween 22°C and 26°C, and night-time temperatures ranged from 18°C to 22°C. \nRelative humidity was maintained at 60%-70%, and environmental data (such as \nradiation and temperature) were continuously recorded using an on-site automatic \nweather station.The complete daily environmental monitoring and irrigation records \nare provided in S1 Dataset.\n2.2. Plant Materials and Cultivation Method\nTest varieties were ‘Jingdan No. 8’ (cherry tomato) and ‘Jingcai No. 8’ \n(strawberry tomato). Soilless substrate cultivation used a 3:1 (v/v) coconut coir:perlite \nmix. Each substrate trough measured 1.0 m × 0.15 m × 0.12 m (length × width × \nheight), with four plants per trough (plant spacing 30 cm, row spacing 55 cm) (Figure \n1). Drip irrigation was applied using municipal tap water, with nutrient solution \nadjusted for pH and EC before irrigation. Prior to planting, coconut coir was fully \n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted January 24, 2026. ; https://doi.org/10.64898/2026.01.23.701235doi: bioRxiv preprint \n\nhydrated and pretreated with standard nutrient solution to minimize initial water-salt \nvariability.\nFigure 1. Schematic diagram of the substrate cultivation trough and drip irrigation setup for \ngreenhouse tomato experiments.\n2.3. Experimental Design and Treatments\nA randomized block design was used with four water-fertilizer treatments: \nconventional water-salt baseline for each variety (CK1 and CK2) and corresponding \nregulated scenarios (low-water high-EC: TK; high-water moderate-EC: TC). Each \ntreatment had three replicates (12 plots total).\nFor model development, daily measurements were averaged across the three \nreplicate troughs within each treatment, yielding one daily observation per treatment \nper day (approximately 180 days per treatment).\nConsidering varietal differences in water demand and nutrient management, a \n“variety–water-salt scenario pairing” framework was adopted. ‘Jingdan No. 8’ \nreceived conventional baseline (CK1) and low-water high-EC (TK) to analyze supply \nlevel and EC interactions under the same variety. ‘Jingcai No. 8’ received \nconventional baseline (CK2) and high-water moderate-EC (TC) to examine EC \nadjustments under enhanced supply. CK1 and CK2 served as practical baselines for \ntheir respective varieties. Although initial design intended pH adjustment for CK2, \nactual irrigation pH remained similar across CK1 and CK2 (Table 1); thus, CK2 was \npositioned as the baseline for ‘Jingcai No. 8’ without treating pH as an independent \nfactor.\nTable 1. Experimental treatment setup.\nTreatment Variety Irrigation Volume \n(L d⁻¹ per trough) EC (dS m⁻¹) pH\nCK1 Jingdan No. 8 4.88 3.00 6.05\nTK Jingdan No. 8 4.14 3.31 6.08\n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted January 24, 2026. ; https://doi.org/10.64898/2026.01.23.701235doi: bioRxiv preprint \n\nTreatment Variety Irrigation Volume \n(L d⁻¹ per trough) EC (dS m⁻¹) pH\nCK2 Jingcai No. 8 4.53 3.21 6.08\nTC Jingcai No. 8 5.65 3.18 6.28\nTarget irrigation volumes, EC, and pH were maintained via flow meters (continuous recording) \nand online/manual adjustments for stability throughout the trial.\n2.4. Measurements and Data Collection\nIrrigation volume was recorded in real time by flow meters; drainage was \ncollected and summarized daily. Nutrient solution EC and pH were measured using \nconductivity (INESA DDS-307A) and pH meters (PHS-3C), calibrated periodically.\nLight conditions were monitored with quantum sensors (LI-190R, LI-COR, \nUSA) placed at greenhouse top, recording PAR at 1-min intervals. Daily cumulative \nradiation (G, MJ m⁻² d⁻¹) was integrated from PAR. Specifically, PAR (μmol m⁻² s⁻¹) \nwas integrated to daily light integral (DLI, mol m⁻² d⁻¹) and converted to energy units \nusing a conversion factor of 0.218 MJ per mol photons . At the daily aggregation \nscale used for modeling, no missing values were present; therefore, no interpolation or \nexclusion was required. As a prespecified quality-control rule, if PAR gaps had \noccurred, gaps ≤10 min would have been linearly interpolated, whereas days \nwith >5% missing PAR would have been excluded from modeling.\nPhotosynthetic parameters (net photosynthetic rate Pn, transpiration rate Tr, \nstomatal conductance Gs, intercellular CO₂ concentration Ci) were measured using a \nportable system (LI-6400XT) on three representative plants per treatment (three \nmeasurements per plant, averaged). Photosynthesis measurements are available in S4 \nDataset.\nYield was recorded at the plot level (one cultivation trough per replicate) at each \nharvest (fruit number and total fresh weight). Yield was expressed as g plot⁻¹ per \nharvest (and kg plot⁻¹ for seasonal cumulative yield). Per-harvest yield and fruit \nnumber data are available in S2 Dataset.\n2.5. Data Processing and Statistical Analysis\nData were organized in Microsoft Excel and analyzed using SPSS (version 25.0). \nTreatment effects were assessed via one-way ANOVA at α = 0.05, with Duncan’s \nmultiple range test applied for post-hoc comparisons. Pearson correlation coefficients \nwere computed to evaluate linear associations, with significance levels denoted as *p \n< 0.05, **p < 0.01, and ***p < 0.001.\nTo quantify the combined influence of radiation and irrigation-solution EC on \ndaily irrigation volume, an EC-adjusted linear regression model was fitted separately \nfor each treatment:\nI = β0 + β1G + β2ECirr + ε     (1)\nA radiation-only baseline model was also fitted for comparison:\nI = α0 + 𝛼1G  + ε            (2)\nwhere I is daily irrigation volume (L d⁻ ¹ per trough), G is daily cumulative \n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted January 24, 2026. ; https://doi.org/10.64898/2026.01.23.701235doi: bioRxiv preprint \n\nradiation (MJ m ⁻² d⁻¹) integrated from PAR, ECirr is the electrical conductivity of \nthe irrigation solution (dS m⁻¹), βi and 𝛼i are regression coefficients, and ε is the \nrandom error term.\nPrior to final model fitting, a 3σ residual rule was applied within each treatment \nto exclude outliers: an initial model was fitted, residuals were calculated, and \nobservations with |residual| > 3σ were removed. The model was then refitted on the \nfiltered dataset.\nTo evaluate out-of-sample predictive performance while accounting for temporal \nautocorrelation, a blocked 5-fold time-series cross-validation was implemented. Daily \nobservations were ordered chronologically and partitioned into five contiguous time \nblocks. In each fold, one block served as the validation set, while the remaining four \nblocks constituted the training set. The 3σ outlier removal was applied solely to the \ntraining data within each fold. Predictions from all folds were pooled, and \nperformance metrics—root-mean-square error (RMSE), mean absolute error (MAE), \nand Nash–Sutcliffe efficiency (NSE)—were computed from the pooled out-of-sample \npredictions.\n3. Results\n3.1. Correlations among Key Irrigation–Drainage Variables\nDaily-scale relationships among irrigation, drainage, and radiation variables were \nexamined using Pearson correlations, including irrigation volume (I), \nirrigation-solution electrical conductivity (EC irr) and pH, drainage volume (D), \ndrainage EC (EC dra) and pH, and daily cumulative radiation (G). The correlation \nstructure is illustrated in Figure 2 (underlying data in S1 Dataset).\nFigure 2. Correlation matrix of key irrigation, drainage, and radiation variables at the daily scale. \nColor intensity indicates correlation strength and direction. Asterisks denote significance (* p < \n0.05, ** p < 0.01, *** p < 0.001). I, irrigation volume (L d⁻¹ per trough); D, drainage volume (L \nd⁻¹ per trough); G, daily cumulative radiation (MJ m ⁻² d⁻¹); EC irr, electrical conductivity of \n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted January 24, 2026. ; https://doi.org/10.64898/2026.01.23.701235doi: bioRxiv preprint \n\nirrigation solution (dS m ⁻¹); pH irr, pH of irrigation solution; EC dra, electrical conductivity of \ndrainage solution (dS m⁻¹); pHdra, pH of drainage solution.\nI was strongly and positively correlated with D (r = 0.70, p < 0.001), consistent \nwith the coupled dynamics of substrate water input and outflow. G was also positively \ncorrelated with I (r = 0.65, p < 0.001), supporting the dominant role of \nradiation-driven transpiration in determining day-to-day irrigation demand.\nAmong solution properties, EC irr was moderately negatively correlated with \nirrigation pH (r = −0.32, p < 0.001). In addition, G was negatively correlated with \nECirr (r = −0.37, p < 0.001), suggesting that EC tended to be reduced on \nhigh-radiation days, potentially as an operational adjustment to mitigate salt \naccumulation risk.\n3.2. Daily Irrigation Demand Prediction Models\nDaily-scale empirical regression models were developed separately for CK1, \nCK2, TK, and TC, using daily accumulated radiation (G) and irrigation water \nelectrical conductivity (EC irr) as predictors and daily irrigation amount (I) as the \nresponse variable.\nDrainage electrical conductivity (EC dra) was not statistically significant during \ncandidate-variable screening (p > 0.05) and was therefore excluded from the final \nmodels.\nCK1:\nI = 2384.62 + 2065.40G - 421.35ECirr\nR² = 0.79, n = 179; G showed a highly significant positive effect (p < 0.001), \nwhile ECirr showed a significant negative effect (p = 0.025).\nCK2:\nI = 605.17 + 1687.12G + 259.51ECirr\nR² = 0.71, n = 180; G showed a highly significant positive effect (p < 0.001). \nThe effect of ECirr was positive but marginal (p = 0.052).\nTK:\nI = 1252.60 + 995.08G + 360.98ECirr\nR² = 0.52, n = 180; G showed a highly significant positive effect (p < 0.001), \nand ECirr showed a significant positive effect (p = 0.032).\nTC:\nI = 3511.60 + 1555.54G - 191.25ECirr\nR² = 0.61, n = 182; G showed a highly significant positive effect (p < 0.001). \nThe effect of ECirr was negative but not significant (p = 0.279).\nOverall, the four models consistently indicate that G is the dominant driver of \nday-to-day irrigation variation, whereas the direction and significance of ECirr \ndepend on the water management scenario.\n3.3. Model Performance Evaluation\nModel performance was evaluated using the coefficient of determination (R²), \nroot mean square error (RMSE), mean absolute error (MAE), and Nash–Sutcliffe \n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted January 24, 2026. ; https://doi.org/10.64898/2026.01.23.701235doi: bioRxiv preprint \n\nefficiency (NSE) (Table 2). RMSE and MAE quantify daily scheduling errors in \nphysical units (L d⁻¹ per trough), whereas NSE summarizes how well the model \nreproduces day-to-day variability relative to the observed mean.\nTable 2. Performance indicators of radiation-driven daily irrigation models under different \ntreatments.\nTreatment n R² RMSE (L d⁻¹ per \ntrough)\nMAE (L d⁻¹ per \ntrough) NSE\nCK1 179 0.79 1.01 0.79 0.79\nCK2 180 0.71 0.93 0.72 0.71\nTK 180 0.52 0.75 0.57 0.52\nTC 182 0.61 1.13 0.85 0.61\nAcross treatments, the fitted radiation-driven models achieved RMSE values of \n0.75–1.13 L d⁻¹ per trough and MAE values of 0.57–0.85 L d⁻ ¹ per trough (Table 2), \nindicating that daily irrigation demand was typically estimated within approximately \n0.6–0.9 L d⁻¹. For the in-sample fitted models, NSE values numerically matched R² \nbecause both metrics were computed as 1 − SSE/SST on the same dataset with an \nintercept term; therefore, RMSE and MAE are particularly informative for operational \nscheduling accuracy.\nPerformance differed among water–salt scenarios. CK1 exhibited the strongest \nfit (R²/NSE = 0.79), suggesting a stable radiation–irrigation response under \nconventional management for ‘Jingdan No. 8’. CK2 and TC showed intermediate \nexplanatory power (R²/NSE = 0.71 and 0.61, respectively), whereas TK had the \nlowest explained variance (R²/NSE = 0.52), implying larger residual variability under \nthe low-water high-EC regime. Notably, TK also showed the smallest absolute errors \n(RMSE = 0.75 L d⁻¹; MAE = 0.57 L d⁻¹), consistent with the lower irrigation supply \nlevel limiting the magnitude of day-to-day fluctuations (Table 1).\nIn relative terms, RMSE corresponded to approximately 18–21% of the mean \ndaily irrigation volume across treatments, supporting the practical interpretability of \nthe error magnitudes for daily scheduling. Because in-sample metrics may overstate \npredictive performance when temporal dependence is present, an out-of-sample \nassessment based on blocked time-series validation is reported in the following \nsection.\n3.4 Baseline Comparison and Blocked Validation\nWithin each treatment, a radiation-only baseline model was compared with an \nEC-adjusted model using blocked 5-fold time-series cross-validation. Daily \nobservations were ordered chronologically and split into five contiguous time blocks; \nin each fold, one block was held out for validation and the remaining blocks were \nused for training. Validation predictions were pooled across folds, and RMSE, MAE, \nand NSE were computed from the pooled out-of-sample predictions (Table 3).\nTable 3. Blocked 5-fold time-series cross-validation performance of the radiation-only \n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted January 24, 2026. ; https://doi.org/10.64898/2026.01.23.701235doi: bioRxiv preprint \n\nbaseline model and the EC-adjusted model (pooled across folds).\nNote: RMSE and MAE are expressed in L d⁻¹ per trough; NSE is dimensionless. Δ indicates \nEC-adjusted minus baseline; negative ΔRMSE/ΔMAE and positive ΔNSE indicate improvement.\nThe cross-validated results indicate that the incremental value of adding EC was \nscenario dependent. Under the low-water high-EC scenario (TK), incorporating EC \nreduced RMSE from 0.829 to 0.815 L d⁻¹ per trough and increased NSE from 0.422 \nto 0.441 (ΔRMSE = −0.014; ΔNSE = +0.019), reflecting a small but consistent \nimprovement. Although the magnitude of improvement was modest, the direction of \nchange (ΔRMSE < 0 and ΔNSE > 0) suggests that EC carries incremental predictive \ninformation only under specific water–salt backgrounds. In contrast, EC adjustment \nslightly degraded performance under CK1 and CK2 (ΔNSE = −0.014 and −0.008, \nrespectively) and produced a larger decline under TC (ΔRMSE = +0.080; ΔNSE = \n−0.066). These comparisons support treating EC as a scenario-specific adjustment \nfactor—most useful under low-water high-EC management—rather than a universal \npredictor that improves accuracy across all treatments.\nCompared with in-sample fitting, NSE decreased under cross-validation, as \nexpected when predicting unseen time blocks, with baseline NSE ranging from 0.422 \nto 0.744 across treatments. Pooled observed–predicted relationships corroborate the \nnumerical comparison: predictions broadly followed the 1:1 reference across \ntreatments, with a modest reduction in scatter for TK under the EC-adjusted model \nand substantial overlap between the two models for CK1, CK2, and TC (Figure 3). \nCollectively, the blocked validation confirms that cumulative radiation captures the \ndominant day-to-day signal, while the additional contribution of EC is scenario \ndependent.\nTreat\nment\nn RMSE \n(baselin\ne)\nMAE \n(baseli\nne)\nNSE \n(baseli\nne)\nRMSE \n(EC-adjus\nted)\nMAE \n(EC-adjus\nted)\nNSE \n(EC-adjusted\n)\nΔRMSE \n(EC − \nbaseline)\nΔMAE (EC − \nbaseline)\nΔNSE \n(EC − \nbaseline\n)\nCK1 179 1.110 0.905 0.744 1.141 0.926 0.730 0.031 0.021 -0.014\nCK2 180 1.020 0.800 0.655 1.032 0.811 0.647 0.012 0.011 -0.008\nTK 180 0.829 0.629 0.422 0.815 0.607 0.441 -0.014 -0.022 0.019\nTC 182 1.313 0.955 0.473 1.393 0.991 0.407 0.080 0.036 -0.066\n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted January 24, 2026. ; https://doi.org/10.64898/2026.01.23.701235doi: bioRxiv preprint \n\nFigure 3. Observed versus predicted daily irrigation volume (I, L d⁻ ¹ per trough) for the blocked \n5-fold time-series cross-validation (pooled predictions) under each treatment. Predictions from \nboth the radiation-only baseline model and the EC-adjusted model are shown. The solid line \nindicates the 1:1 reference line.\n3.5. Radiation-Driven Irrigation and Drainage Responses\nAcross treatments, G maintained a highly significant positive association with I \n(p < 0.001), and the strength of the relationship varied by scenario. The strongest G–I \ncorrelations occurred under CK1 (r = 0.73) and TC (r = 0.72) (Figure 4), reinforcing \nthe consistency of radiation-driven water demand under contrasting management \nbackgrounds.\n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted January 24, 2026. ; https://doi.org/10.64898/2026.01.23.701235doi: bioRxiv preprint \n\nFigure 4. Correlation analysis between daily cumulative solar radiation (G, MJ m⁻ ² d⁻¹) and \nirrigation/drainage variables (I and D, L d⁻¹ per trough) under different treatments.\nRelationships between G and drainage volume were more treatment sensitive. \nThe G–D correlation was highest under CK2 (r = 0.78), followed by CK1, while \nweaker responses were observed under TK and TC (r ≈ 0.51–0.52). This pattern \nsuggests that drainage reflects not only radiation-driven irrigation input but also \nscenario-dependent root uptake and substrate buffering effects.\nCorrelations between G and solution properties were generally weaker. \nAssociations with EC irr and pH were mostly small (|r| < 0.35), with significant \nnegative G–ECirr correlations only under specific treatments. Relationships involving \nirrigation and drainage pH were weaker still, indicating that daily variations in EC and \npH were primarily governed by nutrient-solution preparation and substrate buffering \nrather than direct radiation forcing.\n3.6. Yield Dynamics\nFresh fruit yield per harvest (g plot⁻¹) exhibited pronounced seasonal variation \n(Figure 5; S2 Dataset). Low yields in early winter were followed by increasing \nharvest amounts toward spring, with peaks in April–May as radiation and temperature \nimproved.\n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted January 24, 2026. ; https://doi.org/10.64898/2026.01.23.701235doi: bioRxiv preprint \n\nFigure 5. Temporal variation in fresh fruit yield (g plot⁻¹ per harvest) under different treatments.\nFor ‘Jingdan No. 8’ (CK1 and TK), cumulative yield increased steadily \nthroughout the season. For ‘Jingcai No. 8’ (CK2 and TC), yield remained lower early \nin the season and increased later, indicating more stage-dependent responses under the \ncorresponding water–salt regimes.\nLate-season dynamics differed between the two cultivar–scenario groups. \n‘Jingcai No. 8’ displayed greater fluctuations: CK2 peaked in mid-April (1.68 kg d⁻¹), \nwhereas TC peaked in late March (1.43 kg d⁻¹), followed by sharp declines in late \nMay (CK2: 0.23 kg d⁻¹; TC: 0.12 kg d⁻¹). In contrast, ‘Jingdan No. 8’ maintained \ncomparatively better late-season performance under both CK1 and TK.\n3.7. Fruit Nutritional Quality\nFruit quality was assessed using ascorbic acid/vitamin C (AsA/VC, mg 100 g⁻¹ \nFW) and sugar–acid ratio (SAR = TSS/TA) under contrasting seasonal radiation \nconditions (S3 Dataset), representing low-light winter (January) and high-light spring \n(April). Treatment distributions in the AsA/VC–SAR space shifted markedly between \nthese periods (Figure 6), with lower values in January and higher values in April, \nconsistent with improved nutritional quality and flavor balance under increased \nradiation.\n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted January 24, 2026. ; https://doi.org/10.64898/2026.01.23.701235doi: bioRxiv preprint \n\nFigure 6. Fruit quality expressed as ascorbic acid (AsA/VC, mg 100 g⁻ ¹ FW) versus sugar–acid \nratio (SAR = TSS/TA, dimensionless) under different treatments. TSS is in °Brix and titratable \nacidity (TA) is in g 100 g⁻¹ FW (citric acid equivalents). The boundary of “high VC + high SAR” \nis empirical for reference only.\nWithin ‘Jingdan No. 8’, CK1 maintained relatively high AsA/VC but lower SAR \nthan TK. For ‘Jingcai No. 8’, AsA/VC was generally lower across treatments, and TC \nremained consistently low. Treatment separation became more apparent under higher \nradiation, suggesting that light availability modulated the expression of water–salt \nmanagement effects on fruit composition.\n3.8. Seasonal Photosynthetic Responses\nSeasonal changes in photosynthetic traits are summarized in Figure 7 (S4 \nDataset). Net photosynthetic rate (Pn) increased from January to May across \ntreatments. In January, TK showed the highest mean Pn (13.18 μmol CO₂ m⁻² s ⁻¹), \nwhile TC was lowest (7.99 μmol CO₂ m⁻² s⁻ ¹). In May, mean Pn values converged \nacross treatments (13.51–14.14 μmol CO₂ m⁻² s⁻¹), with CK1 slightly higher than the \nothers.\n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted January 24, 2026. ; https://doi.org/10.64898/2026.01.23.701235doi: bioRxiv preprint \n\nFigure 7. Temporal variation of photosynthetic parameters (Pn, Gs, Ci, and Tr; units shown on \naxes) in tomato under different treatments.\nStomatal conductance (Gs) generally decreased from January to May, \naccompanied by a decline in intercellular CO₂ concentration (Ci). Transpiration rate \n(Tr) increased seasonally in CK1, CK2, and TC, whereas TK maintained relatively \nhigher Pn with comparatively lower Tr, indicating scenario-specific decoupling \nbetween carbon assimilation and water loss.\n4. Discussion\nLimitation (study design): treatments were cultivar-bound (CK1-TK for 'Jingdan \nNo. 8'; CK2-TC for 'Jingcai No. 8'), so treatment contrasts are interpreted within each \ncultivar-specific case study and we avoid cross-cultivar causal claims.\n4.1. Radiation as the Primary Driver of Daily Irrigation Demand\nDaily cumulative solar radiation (G) showed consistently strong and highly \nsignificant positive correlations with irrigation volume across all treatments, \nconfirming radiation-driven transpiration demand as the dominant determinant of \nshort-term water requirements in soilless greenhouse tomatoes [13,33]. This finding \naligns with previous studies demonstrating that, under protected cultivation, daily \nwater uptake is more tightly coupled to incident radiation than to air temperature or \nhumidity alone, particularly when canopy development and root-zone moisture are \nmaintained within suitable ranges [5,11–14]. Radiation governs stomatal opening, leaf \nenergy balance, and transpiration flux, thereby directly translating light variability \ninto irrigation demand at the daily scale [30–32].\nCompared with approaches based on stage-based coefficients or \nevapotranspiration estimates requiring multiple environmental inputs, the \nradiation-driven relationship observed here provides a simplified yet robust basis for \nirrigation scheduling [10,15]. Within each cultivar-specific treatment set (CK1–TK \n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted January 24, 2026. ; https://doi.org/10.64898/2026.01.23.701235doi: bioRxiv preprint \n\nfor ‘Jingdan No. 8’ and CK2–TC for ‘Jingcai No. 8’), the G–I relationship remained \nstable, suggesting that radiation is the primary driver under the management \nbackgrounds examined. Accordingly, G can be used as the core predictor for daily \nirrigation decision-making, while EC and irrigation volume are incorporated as \nscenario-specific adjustments rather than as cross-cultivar general rules.\n4.2. Scenario-Dependent Effects of Electrical Conductivity on Irrigation Demand\nAlthough radiation dominated daily irrigation variability, the role of irrigation \nsolution electrical conductivity (EC) exhibited clear scenario dependence. Under \nconventional water supply (CK1), EC showed a significant negative association with \nirrigation demand, suggesting that higher salinity increased osmotic resistance in the \nroot zone, thereby suppressing plant water uptake [41]. In contrast, under the \nlow-water high-EC scenario (TK), EC exerted a positive effect on irrigation demand, \nindicating that moderate salinity stress may have enhanced transpiration or induced \ncompensatory water uptake responses under limited water supply [27–29].\nThese contrasting responses highlight that EC effects should not be interpreted \nindependently of irrigation background. Under sufficient water availability, elevated \nEC primarily increases osmotic stress and reduces effective water absorption [21,22]. \nUnder restricted irrigation, however, higher EC may alter root hydraulic conductivity, \nstomatal regulation, or assimilate partitioning in ways that partially offset water \nlimitation [20, 23]. Similar interaction effects between salinity and irrigation regime \nhave been reported in greenhouse tomatoes, where moderate salinity improved certain \nphysiological or quality traits without proportionally reducing water uptake [42].\nThis scenario dependence also helps interpret why EC improved blocked \ncross-validation only under TK. One likely reason is that EC irr was operationally \nadjusted in response to radiation (significant negative G–EC irr correlations occurred \nonly under specific treatments), which can introduce collinearity and temporal \nnon-stationarity that weakens out-of-sample generalization in CK1/CK2/TC. Under \nTK, where irrigation supply is restricted and salinity management is more tightly \ncoupled to plant water status, EC irr may better reflect osmotic constraints affecting \nwater uptake and thus provides incremental predictive signal beyond radiation.\nCollinearity between G and ECirr was mild in all treatments (pairwise r = -0.26 to \n-0.52; VIF = 1.07–1.36), suggesting that the scenario-dependent gain/loss is more \nlikely driven by how EC was operationally adjusted over time than by severe \nmulticollinearity.\nConsistent with this interpretation, under the high-water moderate-EC scenario \n(TC), EC effects on irrigation demand were weak and non-significant, indicating that \nabundant water supply can buffer salinity-related impacts on short-term water demand \n[8,9]. Together, these results emphasize that EC should be treated as a \nscenario-adjustment parameter rather than a universal modifier in radiation-driven \nirrigation models.\n4.3. Performance and Practical Applicability of Radiation-Driven Empirical Models\nThe empirical regression models based on daily radiation and EC achieved \n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted January 24, 2026. ; https://doi.org/10.64898/2026.01.23.701235doi: bioRxiv preprint \n\nmoderate to high explanatory power (R² = 0.52–0.79) across treatments, with RMSE \nvalues of 0.75–1.13 L d⁻¹ and NSE values closely aligned with R² [16–18,35,36]. \nThese performance metrics indicate that the models effectively captured major \nday-to-day fluctuations in irrigation demand, while prediction errors were primarily \nassociated with occasional extreme deviations rather than systematic bias [13,17].\nIt is important to emphasize that the proposed models are intended as applied \nscheduling tools rather than mechanistic simulations of plant or root-zone processes \n[10,34]. By relying on easily measurable inputs and treatment-specific coefficients, \nthe models balance accuracy with operational feasibility, which is a key consideration \nfor greenhouse management [11,15]. Similar empirical or semi-empirical approaches \nhave been widely adopted in protected horticulture to support real-time \ndecision-making, particularly where full physiological modeling is impractical \n[12,16,18].\nThe blocked cross-validation results further confirm that daily cumulative \nradiation is the dominant predictor of day-to-day irrigation demand, as the \nradiation-only baseline already provided robust accuracy across treatments. The \nadditional EC term did not uniformly improve cross-validated errors, but its \ncontribution was clearly scenario-dependent, consistent with the direction and \nsignificance of EC effects observed in the fitted models. Therefore, EC is better \ninterpreted as a scenario-adjustment factor for water–salt management rather than a \nuniversal accuracy booster.\n4.4. Yield and Quality Responses under Integrated Water–Salt Management\nSeasonal yield dynamics reflected the combined influences of radiation \navailability, temperature, and water–salt management [2,3,33]. Low yields during \nwinter coincided with limited radiation, while yield peaks in spring corresponded to \nimproved light conditions [1,13]. Under the low-water high-EC treatment (TK), yield \naccumulation was initially comparable to or slightly higher than the conventional \ntreatment but declined later in the season, indicating stage-specific sensitivity to \nwater–salt stress [27,28].\nQuality responses showed clearer differentiation among treatments. Higher \nascorbic acid content and sugar–acid ratios under TK, particularly during \nhigh-radiation periods, suggest that moderate water and salinity stress promoted \nassimilate concentration and quality enhancement [43]. This trade-off between yield \nstability and quality improvement is consistent with previous findings that controlled \nstress can enhance fruit quality attributes in greenhouse tomatoes while maintaining \nacceptable yield levels [22,23,29,44].\nNotably, quality differences among treatments became more pronounced under \nhigher radiation, indicating that light availability modulates the expression of water–\nsalt management effects on fruit composition [30–32]. These results underscore the \nimportance of explicitly integrating radiation conditions when evaluating irrigation \nand salinity strategies aimed at balancing yield and quality.\n4.5. Variety-Specific Responses, Limitations, and Management Implications\n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted January 24, 2026. ; https://doi.org/10.64898/2026.01.23.701235doi: bioRxiv preprint \n\nVarietal responses differed under comparable management frameworks. ‘Jingdan \nNo. 8’ exhibited greater yield stability and quality potential under both conventional \nand low-water high-EC conditions, whereas ‘Jingcai No. 8’ showed weaker \nlate-season performance under high-water moderate-EC management [2,3,45]. These \ndifferences likely reflect varietal variation in root architecture, salinity tolerance, and \nphotosynthetic regulation, reinforcing the need for variety-specific optimization rather \nthan universal irrigation or EC targets [8,9].\nSeveral limitations should be acknowledged. Treatments were variety-bound, \nwhich restricts direct cross-variety comparisons of EC effects. In addition, pH \nvariation among treatments was limited, preventing robust evaluation of pH–water–\nsalt interactions [7,39]. Future studies should incorporate factorial designs with \nclearer gradients of EC and pH within the same variety, as well as mixed-effects \nmodeling approaches to better address heterogeneity [16,17].\nFrom a management perspective, the results suggest that radiation-driven \nirrigation prediction can serve as the core component of daily scheduling, with EC \nused to adjust strategies according to water availability and varietal characteristics \n[10,11,15]. Such an approach provides a practical pathway toward integrated water–\nsalt management in soilless greenhouse tomato production.\n5. Conclusions\n(1) Daily irrigation demand in soilless greenhouse tomato was primarily driven \nby cumulative solar radiation (G), showing a strong positive association with \nirrigation amount (I) and drainage (D). A radiation-only linear baseline therefore \nprovides a parsimonious predictor for day-to-day irrigation forecasting.\n(2) In 5-fold blocked time-series cross-validation, the radiation-driven models \nachieved RMSE of 0.815–1.393 L d⁻¹ per trough and NSE of 0.407–0.730 across \ntreatments. Adding irrigation-solution electrical conductivity (EC irr) yielded a small \nimprovement under the low-water high-EC scenario (TK), but changes were \nnegligible or negative under CK1, CK2, and the high-water moderate-EC scenario \n(TC). Thus, EC should be treated as a scenario-specific adjustment rather than a \nuniversal driver.\n(3) Differences in yield, fruit quality, and photosynthetic traits were consistent \nwith the contrasting water–salt management backgrounds, supporting that irrigation \nforecasting should be interpreted together with production targets. However, inference \nremains limited to the cultivar–management combinations tested (CK1–TK in \n‘Jingdan No. 8’; CK2–TC in ‘Jingcai No. 8’).\n(4) For practice, we recommend a two-step scheduling workflow: estimate daily \nirrigation from G using the radiation-only baseline, then optionally apply an EC-based \ncorrection where validated (e.g., low-water high-EC conditions), and distribute the \ndaily total into irrigation events according to greenhouse control settings. This \nprovides a low-input pathway toward integrated water–salt management for soilless \ngreenhouse tomato production.\nAuthor Contributions: Conceptualization, Y.Y. and Y.M.; Methodology, L.X., Y.M. \n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted January 24, 2026. ; https://doi.org/10.64898/2026.01.23.701235doi: bioRxiv preprint \n\nand Y.Y.; Formal analysis, L.X. and Q.F.; Investigation, X.G., H.S. and X.L.; Data \ncuration, L.X. and X.L.; Writing—original draft, L.X.; Writing—review and editing, \nY.Y., Y.M. and Q.F.; Visualization, L.X.; Supervision, Y.Y.; Project administration, \nY.M. and Y.Y.; Funding acquisition, Y.Y.\nFunding: This research was supported by the National Key R&D Program of China \n(Grant No. 2023YFD2000602); the Major Science and Technology Projects of \nXinjiang Uygur Autonomous Region (Grant Nos. 2022A02005-1, 2022A02005-5); \nthe Key R&D Program of Xinjiang Uygur Autonomous Region (Grant Nos. \n2023B02024-2, 2023B02024-2-1); and the Agricultural Science and Technology \nInnovation Steady Support Program of Xinjiang Academy of Agricultural Sciences \n(Grant No. xjnkywdzc-2025003-02,xjnkywdzc-2025002-10). The funders had no role \nin the study design, data collection and analysis, decision to publish, or preparation of \nthe manuscript.\nData Availability Statement:All relevant data underlying the findings of this study \nare available within the manuscript and its Supporting Information files. Specifically:\nS1 Dataset: Daily environmental parameters (radiation, temperature), irrigation \nvolumes, and drainage records used for modeling.\nS2 Dataset: Detailed yield and fruit number data sorted by harvest date.\nS3 Dataset: Fruit nutritional quality measurements (Vitamin C, soluble sugar, \ntitratable acidity).\nS4 Dataset: Photosynthetic characteristics (Pn, Gs, Ci, Tr) measured at different \ngrowth stages.\nS1 Code: Python script used for analyzing fruit quality and generating Figure 6.\nThere are no restrictions on data access.\nEthics statement: Not applicable.\nConflicts of Interest: The authors declare no conflict of interest.\nReferences\n1. Tang, M.; Lin, R.; Song, J.; Yu, J.; Zhou, Y. Advances in Physiological and Molecular \nMechanisms of Tomato Responses to Light and Temperature Stress. Acta Horticulturae Sinica \n2022, 49, 2174–2188. \nhttps://doi.org/10.16420/j.issn.0513-353x.2022-0600.\n2. Ma, X.; Chai, X.; Liu, Y.; Li, J. Evaluating the Canopy Light Environment, Photosynthesis, and \nFruit Comprehensive Performance of Greenhouse Tomato under Different Mechanized \nPlanting Layouts. Horticultural Plant Journal 2025.\n3. Zhang, Y.; Li, Y.; Sun, Z.; Li, W.; Liu, X.; Li, T. Estimating the Light Interception and \nPhotosynthesis of Greenhouse-Cultivated Tomato Crops under Different Canopy \nConfigurations. Agronomy 2024, 14, 249.\n4. Dirlik, I.F.; Kaya, C. Sensor-Guided Smart Irrigation for Tomato Production: Comparing Low \nand Optimum Soil Moisture in Greenhouse Environments. Food Energy Security 2025, 14, \ne70082.\n5. Liu, C.; Gao, H.; Lv, G.; Wu, X.; Zhao, F.; Huo, Z.; Zhang, X. Estimation and Experiment of \n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted January 24, 2026. ; https://doi.org/10.64898/2026.01.23.701235doi: bioRxiv preprint \n\nthe Suitable Irrigation Amount of Potted Tomatoes with Coconut Bran Using Pan Evaporation. \nTransactions of the Chinese Society of Agricultural Engineering 2022, 38, (issue not \nprovided).\n6. Wang, W.; Zhang, X. Integrated Effects of Irrigation and Fertilization on Yield, Quality and \nWater–Fertilizer Use Efficiency of Substrate-Cultured Tomato. Journal of Irrigation and \nDrainage 2023, 42, 51–56. https://doi.org/10.13522/j.cnki.ggps.2022218.\n7. Song, S.; Lim, R.B.H.; Gaw, L.Y.F.; Richards, D.R.; Tan, H.T.W. Comparison of Vegetable \nProduction, Resource-Use Efficiency and Environmental Performance of High-Technology \nand Conventional Farming Systems for Urban Agriculture. Science of The Total Environment \n2022, 807, 150621.\n8. Yin, Z.; Cai, J. Combined Effect of Irrigation and Fertilization on Soil Water, Nutrient \nTransport, Yield and Water Use Efficiency of Greenhouse Tomato. Journal of Irrigation and \nDrainage 2023, 42, 33–44. https://doi.org/10.13522/j.cnki.ggps.2022421.\n9. Lu, T.; Wang, H.; Zhang, T.; Shi, C.; Jiang, W. Influence of Electrical Conductivity of Nutrient \nSolution in Different Phenological Stages on the Growth and Yield of Cherry Tomato. \nHorticulturae 2022, 8, 378.\n10. Chen, W.-H.; Yang, F. Semiclosed Greenhouse Climate Control under Uncertainty via \nMachine Learning and Data-Driven Robust Model Predictive Control. IEEE Transactions on \nControl Systems Technology 2021, 30, 1186–1197.\n11. Liu, T.; Han, X.; Hai, Y.; Li, F.; Wang, F.; Xu, W. Construction and Application of a \nWeighing Feedback-Based Irrigation System for Facility Tomatoes. Transactions of the \nChinese Society of Agricultural Engineering 2024, 40, 85–96.\n12. Xu, D. Effects of Different Aerobic Drip Irrigation Methods on Yield, Quality and \nPhotosynthesis of Cherry Tomato in Greenhouse. Jiangsu Journal of Agricultural Sciences \n2020, 36, 152–157.\n13. Shang, C.; Han, Y.; Guo, W. Water Requirement Rule of Tomato in Glasshouse under Rock \nWool Nutrient Solution Cultivation. Journal of Agricultural Science and Technology 2019, 21, \n109–117. https://doi.org/10.13304/j.nykjdb.2018.0540.\n14. Lozano-Castellanos, L.F.; Lozano-Castellanos, I.C.; Correa-Guimaraes, A. Technologies \nApplied to Artificial Lighting in Indoor Agriculture: A Review. Sustainability 2025, 17, 3196.\n15. Zhang, Z.; Liu, J.; Kong, T.; Xie, J.; Zhang, M. Design and Experiment of Greenhouse Melon \nIrrigation System Based on Environmental VPD Decision. Transactions of the Chinese Society \nfor Agricultural Machinery 2022, 53, 371–378.\n16. Xiang, L.; Li, J. Effects of Irrigation Amount and Frequency on Root Growth, Yield and \nNutrient Absorption of Tomato. Journal of Agricultural Science and Technology of Northwest \nA&F University 2024, 52, 80–92.\n17. Zhu, C.; Zhang, J.; Xiang, L.; Li, J. Validation of Advanced Decision Irrigation Model of \nGreenhouse Tomato Soil Cultivation Based on Environmental Factors. Journal of China \nAgricultural University 2023, 28, 60–73.\n18. Wang, Z.; Yuan, M.; Yin, W.; Zhang, C.; Zhang, Z.; Hu, X. A Machine Learning-Based \nIrrigation Prediction Model for Cherry Tomatoes in Greenhouses. Computers and Electronics \nin Agriculture 2025, 237, 110558.\n19. Collado, C.; Hernandez, S.; Hernandez, R. Supplemental Greenhouse Lighting Increased the \nWater Use Efficiency, Crop Growth, and Cutting Production in Cannabis sativa. Frontiers in \n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted January 24, 2026. ; https://doi.org/10.64898/2026.01.23.701235doi: bioRxiv preprint \n\nPlant Science 2024, 15, 1371702.\n20. Karaca, C.; Buyuktas, D.; Kurunc, A.; Bastug, R.; Navarro, A. Effects of Salinity Stress on \nDrip-Irrigated Tomatoes Grown under Mediterranean-Type Greenhouse Conditions. \nAgronomy 2023, 13, 36.\n21. Alshami, A.K.; El-Shirbeny, A.; Al-Omran, A.M.; Alghamdi, A.G.; Louki, I.; Alkhasha, A. \nResponses of Tomato Crop and Water Productivity to Deficit Irrigation Strategies and Salinity \nStress in Greenhouse. Agronomy 2023, 13, 3016.\n22. Pei, S.; Zhang, H.; Zhang, Z.; Zhao, F.; Li, Z. Physiological Response of Potted Tomatoes to \nNaCl and Na₂SO₄ Brackish Water Irrigation. Scientia Agricultura Sinica 2024, 57, 570–583.\n23. Chen, P.; Dong, X.; Tian, L.; Zhang, X.; Liu, X.; Sun, H. Regulation Effects of Fulvic Acid on \nTomato Yield and Quality under Saline Water Irrigation. Chinese Journal of Eco-Agriculture \n2023, 31, 452–462. \nhttps://doi.org/10.12357/cjea.20220178.\n24. Linker, R.; Kirzhner, I. Model-Based Simulation–Optimization of Irrigation Scheduling: A \nField Evaluation with Processing Tomatoes. Smart Agricultural Technology 2023, 4, 100234.\n25. Liu, H.L., H.; Ning, H.; Zhang, X.; Li, S.; Pang, J.; Wang, G.; Sun, J. Optimizing irrigation \nfrequency and amount to balance yield, fruit quality and water use efficiency of greenhouse \ntomato. Agricultural Water Management, 2019. 226: p. 105787.\n26. Incrocci, L.; Malorgio, F.; Pardossi, A. New Trends in the Fertigation Management of \nIrrigated Vegetable Crops. Horticulturae 2017, 3, 37.\n27. Song, Y.; Zhang, J.; Xing, J.; Wang, L.; Wang, X.; Hu, C.; Li, W.; Tan, Z.; Cheng, Y. \nOptimizing Water–Fertilizer Coupling Across Different Growth Stages of Tomato in Yellow \nSand Substrate: Toward Enhanced Yield, Quality and Resource Use Efficiency. Plants 2025, \n14, 936.\n28. Zhang, P.; Dai, Y. Effects of Salinity Stress at Different Growth Stages on Tomato Growth, \nYield and Water-Use Efficiency. Communications in Soil Science and Plant Analysis 2017, \n48, 624–634.\n29. Li, J.; Qu, Z.; Wang, S.; He, P.; Zhang, N. Effects of Alternating Irrigation with Fresh and \nSaline Water on Soil Salt, Soil Nutrients, and Yield of Tomatoes. Water 2019, 11, 1693.\n30. Xiao, L.; Ma, Y.; Wang, X.; Xing, J.; Zheng, J.; Ma, Y. Effects of Supplemental Light Quality \nand Substrate Moisture Content on Morphological Regulation and Photosynthetic \nCharacteristics of Cucumber Seedlings. Journal of Agricultural Science and Technology 2025, \n27, 68–77. \nhttps://doi.org/10.13304/j.nykjdb.2023.0733.\n31. Marie, T. Growing Tomato in Controlled Environments under Continuous Light Requires \nDynamic LEDs to Entrap the Circadian Rhythm, Adjust Canopy Architecture, and Balance \nPhotostasis. Ph.D. Thesis, University of Guelph, Guelph, Canada, 2024.\n32. He, W.; Li, Z.; Su, W.; Gan, L.; Xu, Z. Effect of Different Light Intensities on the \nPhotosynthate Distribution in Cherry Tomato Seedlings. The Journal of Horticultural Science \nand Biotechnology 2019, 94, 611–619.\n33. Hong, M.; Zhang, Z.; Fu, Q.; Liu, Y. Water Requirement of Solar Greenhouse Tomatoes with \nDrip Irrigation under Mulch in the Southwest of the Taklimakan Desert. Water 2022, 14, 3050.\n34. Sun, L.; Yao, M.; Mao, L.; Zhao, M.; Niu, H.; Xu, Z.; Wang, T.; Wang, J. Simulation of Soil \nWater Movement and Root Uptake under Mulched Drip Irrigation of Greenhouse Tomatoes. \nWater 2023, 15, 1282.\n35. Ge, J.; Gong, X.; Ping, Y.; Luo, J.; Li, Y. Evaluation of Irrigation Modes for Greenhouse Drip \n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted January 24, 2026. ; https://doi.org/10.64898/2026.01.23.701235doi: bioRxiv preprint \n\nIrrigation Tomatoes Based on AquaCrop and DSSAT Models. Plants 2023, 12, 3863.\n36. Shan, Z.; Zhang, X.; Si, Z.; Yi, R.; Fan, H. Optimizing Irrigation and Nitrogen Application for \nGreenhouse Tomato Using the DSSAT–CROPGRO–Tomato Model. Water 2025, 17, 426.\n37. Hu, J.; Li, Y.; Hou, J.; Sun, Z.; Wang, H.; He, D. Analysis and Prospect of the Environmental \nControl Systems for Greenhouse. Transactions of the Chinese Society of Agricultural \nEngineering 2024, 40, 112–128.\n38. Liu, X.; Shu, S.; Zhang, A.; Sun, J. Potential Advantages and Theoretical Basis of \nLight-Associated Temperature Regulation in Protected Cultivation. Journal of Nanjing \nAgricultural University 2023, 46, 823–832.\n39. Zhao, W.; Yu, H.; Wang, Z.; Ma, H. Effects and Driving Factors of Water and Fertilizer \nCoupling on the Yield and Quality of Tomato Cultivated in Carbon Substrate. Transactions of \nthe Chinese Society of Agricultural Engineering 2025, 41, 199–207.\n40. Savvas, D.; Gruda., N. Application of soilless culture technologies in the modern greenhouse \nindustry—A review. Eur. J. Hortic. Sci, 2018. 83(5): p. 280-293.\n41. Moya, C.; Verdugo, G.; Flores, M.F.; Urrestarazu, M.; Álvaro, J. E.Increased Electrical \nConductivity in Nutrient Solution Management Enhances Dietary and Organoleptic Qualities \nin Soilless Culture Tomato. HortScience 2017, 52, 868–872.\n42. Bonachela, S.F., M. D.; Cabrera-Corral, F. J.; Granados, M. R. Salt and irrigation management \nof soil-grown Mediterranean greenhouse tomato crops drip-irrigated with moderately saline \nwater. Agricultural Water Management, 2022. 262: p. 107433.\n43. Wu, Z.; Qiu, Y.; Hao, X.; Li, S.; Kang, S. Response of Yield and Quality of Greenhouse \nTomatoes to Water and Salt Stresses and Biochar Addition in Northwest China. Agricultural \nWater Management 2022, 270, 107736.\n44. Roșca, M.M.; Stoleru, V. Tomato Responses to Salinity Stress: From Morphological Traits to \nGenetic Changes. Frontiers in Plant Science 2023, 14, 1118383.\n45. Shin, J.-Y. Analyses of Canopy Light Interception and Photosynthesis of Greenhouse Tomato \nPlants under Diffuse Films Using 3D Plant Model and Ray-Tracing Simulation. Ph.D. Thesis, \nSeoul National University, Seoul, Korea, 2020.\n.CC-BY 4.0 International licenseperpetuity. It is made available under a \npreprint (which was not certified by peer review) is the author/funder, who has granted bioRxiv a license to display the preprint in \nThe copyright holder for thisthis version posted January 24, 2026. ; https://doi.org/10.64898/2026.01.23.701235doi: bioRxiv preprint","source_license":"CC-BY-4.0","license_restricted":false}