Coupled radiation-evapotranspiration CFD modelling of an indoor vertical farm: Validation and sensitivity to model complexity

Indoor vertical farming is a promising solution for sustainable urban food production, but its energy demand is strongly affected by the coupled operation of lighting, heating, ventilation, air conditioning and dehumidification systems. Reliable microclimate prediction therefore requires numerical models able to represent the interaction between artificial radiation, airflow and crop transpiration. This study presents a multi-fidelity three-dimensional computational fluid dynamics benchmark of an industrial indoor vertical farming facility, comparing five modelling strategies of increasing complexity, from simplified convection-only approaches to a fully coupled radiation-evapotranspiration model. The proposed framework combines the Discrete Ordinates radiation model with a local Penman-Monteith formulation, in which stomatal resistance depends on the net radiation absorbed by the canopy. The simulations were validated against a 72-h experimental campaign. The results show that simplified models can lead to physically inconsistent predictions. Cases neglecting evapotranspiration reproduced air temperature with apparently acceptable errors, but strongly underestimated relative humidity, with mean absolute errors of 5.98–6.93 percentage points. Conversely, introducing uniform evapotranspiration without radiation reduced relative humidity errors but increased the temperature mean absolute error to 1.26 °C. The coupled radiation-evapotranspiration approaches provided the best overall agreement: the constant-source model achieved mean absolute errors of 0.57 °C and 1.70 percentage points for temperature and relative humidity, respectively, while the fully predictive Penman-Monteith-based model yielded comparable values of 0.64 °C and 1.90 percentage points. Beyond bulk validation, the fully predictive model captured strong local heterogeneity, with coefficients of variation of 61.8% for net absorbed radiation and 83.4% for the moisture source term on a representative vertical slice. These results demonstrate that calibrated source models can support rapid analyses for climate-control design, whereas predictive digital twins and crop-stress mapping require coupled radiation-Penman-Monteith modelling despite the higher computational cost.