Adjusting soil parameters to improve green roof winter energy performance based on neural-network modeling

Green roofs contribute notably to green building and sustainable city goals. Most studies focused on summer cooling and energy-saving benefits, with little attention on winter warming effect. This study evaluated contributions of the soil component to winter warming with reference to the thickness and moisture content. In subtropical Wuyishan city, Fujian Province, China, we built three houses as experimental green-roof plots. We adjusted soil thickness, measured moisture content, and monitored their impacts on roof outer surface temperature (ROST) in winter. We designated measured outdoor temperature, relative humidity, solar radiation, wind speed, soil thickness and soil moisture content as independent variables, and ROST as dependent variable. The empirical data were enlisted to establish a neural-network prediction model in conjunction with a genetic algorithm to optimize parameter thresholds and weights. The numerical model successfully predicted winter ROST at different combinations of soil thickness and moisture content and identified the best value for energy-saving performance. The model predicted that the maximum ROST of 13.5 °C was achieved at 20 cm soil thickness and 3.9 % moisture content. The choice of deciduous or evergreen plants could fine-tune soil moisture requirement and improve warming service. The study demonstrated the feasibility of optimizing the green-roof design by adjusting controllable soil variables. The neural-network model could reliably predict ROST under different soil conditions to inform design and management.