Green Roof Cooling Models Put to the Test Against Real Rooftop Data

Green roofs have become one of the most visible symbols of urban climate adaptation, promising cooler streets, lower building energy bills, and greener skylines. But when city planners and climate scientists run simulations to estimate how much a city could actually cool down if its rooftops turned green, they rely on computer models that are rarely checked against real measurements. A new study published in Geoscientific Model Development by Maria Martinez Mendoza, Alireza Saeedi, James A. Voogt, and E. Scott Krayenhoff of the University of Western Ontario and the University of Guelph takes on that blind spot, subjecting two widely used green roof models to one of the most rigorous process-level evaluations performed to date.

The team evaluated two configurations of EcoRoof, the green roof module embedded in the building energy simulation program EnergyPlus, alongside a multilayer green roof parameterization coupled to the Weather Research and Forecasting model, known as WRF-MLGR. Rather than simply comparing predicted rooftop temperatures with observations, the researchers dissected the underlying physics: how well each model reproduced latent heat flux, soil heat flux, surface temperature, and substrate water content. This process-based approach matters because a model can sometimes land on the right answer for the wrong reasons, and only by examining the individual energy balance components can scientists identify where the physics itself breaks down.

The observational benchmark came from a green roof test array installed on the rooftop of Talbot College at Western University in London, Ontario. The array consists of elevated modules planted exclusively with Sedum spurium, a hardy succulent common on extensive green roofs, growing in a homogeneous 0.15 meter layer of substrate. Instruments including lysimeters, rain gauges, radiation sensors, and heat flux plates recorded energy balance components and microclimate variables during the summer and fall of 2014, from July through October. The researchers divided the record into two 62-day analysis periods, a summer window and a fall window, and further stratified the data into dry versus rainy days and high versus low net radiation conditions to probe how model performance shifts with environmental context.

The observed energy balance itself provided a valuable reference point. During summer, net radiation peaked around 450 watts per square meter at midday, with roughly half of that energy leaving the surface as sensible heat, about 46 percent converted into latent heat through evapotranspiration, and only about 4 percent conducted into the substrate. Surface temperatures climbed to nearly 30 degrees Celsius. In fall, all components weakened, with midday net radiation near 300 watts per square meter and latent heat accounting for about 41 percent of the partitioning. Notably, the elevated configuration of the modules, which creates an insulating air gap between the growing media and the roof deck, appeared to reduce conductive heat transfer compared with roofs in direct contact with the deck.

When the simulations were compared against these measurements, the two EcoRoof versions, an original configuration and a modified one following refinements proposed by earlier researchers, performed similarly and generally tracked observations well. Across the study periods, their root mean square errors averaged roughly 56 to 60 watts per square meter for latent heat flux, 3 to 4 degrees Celsius for surface temperature, about 22 watts per square meter for soil heat flux, and 0.03 cubic meters per cubic meter for substrate water content. The WRF multilayer green roof scheme matched EcoRoof on latent heat flux but fared far worse elsewhere, with soil heat flux errors exceeding 100 watts per square meter, surface temperature deviations near 5 degrees Celsius, and moisture errors roughly twice as large as EcoRoof’s.

The diurnal patterns of error were especially revealing. EcoRoof tended to underestimate evapotranspiration in the early morning and overestimate it in the afternoon, with positive biases peaking around 50 watts per square meter. WRF-MLGR, by contrast, persistently underestimated latent heat flux through most of the day, with morning biases reaching 100 watts per square meter. For surface temperature, EcoRoof showed a modest but persistent warm bias of roughly 1 to 3 degrees Celsius, while WRF-MLGR swung from morning underestimation to substantial afternoon overestimation. The most dramatic failures appeared in soil heat flux, where WRF-MLGR overestimated midday values by more than 200 watts per square meter under high radiation conditions, at times producing a diurnal pattern that resembled a bare, unvegetated substrate more than the living roof it was meant to simulate.

The physical explanations for these divergences lie in the structural differences between the models. EcoRoof’s overestimation of latent heat traces back to its simplified treatment of vegetation physiology. The model scales transpiration strongly with leaf area index and allows stomata to remain open until vapor pressure deficit becomes high, but Sedum is a CAM plant that conserves water through early stomatal closure, a feed-forward behavior the model does not capture. WRF-MLGR, conversely, represents latent heat solely through transpiration, with no substrate evaporation term at all, and applies strong atmospheric controls on stomatal resistance that suppress transpiration prematurely. Its lack of an explicit vegetation canopy means all incoming radiation is absorbed at a combined substrate-vegetation surface, pushing energy into the high thermal admittance substrate rather than allowing foliage to shade it and buffer the surface.

Moisture dynamics told a similar story. All three model configurations correctly reproduced the timing of rainfall events, showing sharp increases in substrate water content after precipitation, but the magnitudes diverged. The modified EcoRoof, which adds canopy interception and a moisture-dependent runoff formulation, tracked observed wetting and drying cycles most closely. The original version produced larger post-rain peaks and drained too quickly, while WRF-MLGR consistently underpredicted water content because its free-drain bottom boundary and nonlinear hydraulic conductivity allow water to percolate out of the substrate column very efficiently, producing a persistent dry bias. Separating days by moisture regime and radiative forcing showed that dry, high-radiation conditions amplified each model’s characteristic biases, while wet and cloudy conditions reduced mean errors but increased variability.

The implications reach well beyond one rooftop in Ontario. Because EcoRoof systematically overestimates evapotranspiration, earlier studies relying on it may have overstated the evaporative cooling potential of green roofs, particularly under high radiation and low moisture availability, although the model appears more trustworthy for subsurface thermal dynamics and building energy applications. WRF-MLGR’s structural limitations, meanwhile, mean that city-scale simulations of green roof heat mitigation built on this scheme warrant caution, since persistent underestimation of cooling fluxes and overestimation of surface heating could distort adaptation planning. Encouragingly, the research team has already developed an upgraded version of the WRF scheme that addresses the main deficiencies and performs substantially better against field observations.

More broadly, the study delivers a pointed message to the urban climate modeling community: rigorous, process-level evaluation should be a prerequisite for any model deployed to assess heat mitigation strategies. Most prior applications of green roof schemes in WRF and EnergyPlus were never validated against surface energy balance measurements, and those that were evaluated focused narrowly on temperatures. As cities worldwide invest billions in rooftop greening to combat extreme heat, the accuracy of the models guiding those decisions becomes a matter of public consequence. The authors argue that future model development should prioritize vegetation-specific physiology, canopy and substrate heat storage, explicit radiative absorption by foliage, and the distinct pathways of evaporation and transpiration, ensuring that the virtual green roofs guiding real-world policy behave like the living ones above our heads.

Subject of Research: Evaluation of green roof energy balance models in WRF and EnergyPlus against field observations

Article Title: Process-based evaluation of green roof models for assessment of heat mitigation efficacy in WRF (v4.3.1) and EnergyPlus (v8.6.0)

Article References: Martinez Mendoza, M., Saeedi, A., Voogt, J. A., & Krayenhoff, E. S. (2026). Process-based evaluation of green roof models for assessment of heat mitigation efficacy in WRF (v4.3.1) and EnergyPlus (v8.6.0). Geoscientific Model Development, 19(19), 9729-9749. https://doi.org/10.5194/gmd-19-9729-2026

Image Credits: AI Generated

DOI: 10.5194/gmd-19-9729-2026

Keywords: green roofs, urban heat island, WRF, EnergyPlus, EcoRoof, surface energy balance, latent heat flux, model evaluation, evapotranspiration, urban climate modeling, Sedum, Geoscientific Model Development