Modelling green roof physics has mainly involved developing complex numerical models to simulate physical processes that occur between the many surfaces and materials that define a green roof system. However, a recent review of these models declares that (1) increasing model complexity may not necessarily translate into better predictability of key thermal performance metrics (e.g., interior temperature), and (2) researchers should consider developing parsimonious models and alternate modelling techniques that can predict variables or processes that are more indicative of green roof thermal performance. In this paper, two inverse models - a resistor-capacitor (RC) thermal network model and an implicit finite difference (FD) model - are developed. The models are calibrated with multi-year sensor data from a green roof in Ottawa, Canada by employing the genetic algorithm. The calibrated models are then evaluated based on their ability to predict hourly rates of heat flux. Our results demonstrate that characterization of green roof thermal properties is affected by differences in spatial resolution between the models. Predictability of hourly heat flux by the RC and FD models resulted in a root -mean-squared error that ranged between 0.51 and 1.04 W/m2 and 0.42-0.81 W/m2, respectively, across five separate months: May through September 2016. Percent reductions in total monthly heat exchange relative to a conventional roof were better predicted by the FD model each month. Validation of each model using five continuous months of data from 2017 demonstrates the inverse models generated realistic thermophysical green roof properties.
展开▼
