Evaluating the Energy Performance of Ballasted Roof Systems

Transcription

1 Evaluating the Energy Performance of Ballasted Roof Systems André O. Desjarlais, Thomas W. Petrie and Jerald A. Atchley, Building Envelopes Program, Oak Ridge National Laboratory, Richard Gillenwater, Carlisle SynTec, Inc., and David Roodvoets, SPRI, Inc. Prepared for Single Ply Roofing Industry 77 Rumford Avenue, Suite 3B Waltham, MA ORNL Report Number UF Oak Ridge National Laboratory P.O. Bo 2008 Oak Ridge, Tennessee managed by UT- Battelle LLC. for the U.S. Department of Energy under contract DE-AC05-00OR22725 September 2007

2 Evaluating the Energy Performance of Ballasted Roof Systems Andre Desjarlais, Thomas Petrie, William Miller, ORNL, Richard Gillenwater, Carlisle SynTec, Inc., and David Roodvoets, SPRI, Inc. Abstract It is well known that the mass of a ballasted roof can reduce peak roof temperatures and delay the heat flow into a building. Although they perform similar functions, ballasted roofs do not meet the traditional requirements of high solar reflectance and high thermal emittance set out by the Environmental Protection Agency (EPA) and other organizations regarding being a cool roof. To address whether ballasted roofing systems offer similar energy efficiency benefits as cool roofs, a project to perform side-by-side eperiments was initiated. Different weightings of ballasted roofs were compared to a paver system and black and white membranes. These si test sections were constructed and installed on the Roof Thermal Research Apparatus (RTRA) and were monitored for thermal performance for a twenty-four month period. Data collection includes continuous monitoring of temperatures, heat flows, and weather conditions as well as periodic verification of the surface properties of solar reflectance and thermal emittance. These data will answer what impact a ballasted roof has on heat flow into a building and on roof surface temperature. Furthermore, comparisons between the ballasted and unballasted membranes will allow for an assessment of whether ballasted systems perform as well as white membranes and are deserving of a cool roof status within the codes. This report also describes the modeling of the thermal performance of all si systems with the Simplified Transient Analysis of Roofs (STAR) program. STAR does well for light weight roofs with eposed membranes. Reasonable values of effective heat capacity and thermal conductivity were sought for the ballasted systems that yielded good agreement between the predicted and measured membrane temperatures and insulation heat flues. Goodness of agreement was judged by average differences over si month periods between mid-april and mid-october. Modeling of ballasted roofs was desired so that eperimental results measured in East Tennessee could be generalized to other climates across the United States. Introduction Ballasted systems entered the roofing market in the early 1970 s. The stone used with these systems is different from the traditional ¼-in. or 6mm chip or smaller stone used with built-up and modified bitumen roofing. With these last two systems the small stones are partially imbedded into the topcoat of asphalt to protect the asphalt (same applies to coat tar based systems) from the harmful ultraviolet rays of the sun. The stone used as ballast for single ply systems is large in size, #4 (0.75 to 1.5 inch) in diameter) and larger stone. Ballast comes in other configurations such as concrete or rubber pavers. Ballast is applied in loadings from 10 (the minimum) to over 24 lbs/ft 2. So with the loose-laid ballasted roof system, the contractor places all the components of the roof system, including the thermal barrier - 2 -

3 and insulation, unattached on the roof deck. The membrane is also loose-laid ecept for attachment around the perimeter of the building and at roof penetrations. The ballast is then placed on top of the membrane weighting down all the components to hold them in place. This technique eliminates the use of fasteners that are used to hold the roofing components in place, which in turn minimizes thermal bridging. Ballast is also used with the inverted - protected roof system where the roof system is built upside down. A protective course maybe placed over the deck. The membrane is then laid down, followed by the insulation, a filter fabric and the ballast. The ballast often used in this application is pavers because this system is often used in applications where there will be pedestrian traffic. Plaza decks and roof top terraces are a few eamples. The paver offers a trafficable surface with the insulation acting as a thermal protection layer and a shock absorber for the waterproofing system below it. Another form of ballast is mied soil media and plants to form a roof garden with unique aesthetic appeal and performance characteristics such as storm water management. Early proponents of ballasted roofing systems focused mainly on how to design a ballasted roof system to resist the destructive powers of the wind. This led to a number of wind performance issues toward the end of the 70 s and into the early 80 s. This, in turn, energized the industry to find the answers for designing a ballasted system for specific wind zones. Etensive wind tunnel work was conducted with thorough verification of the modeling through field observations all leading to the development of the ANSI/SPRI RP-4 national standard entitled Wind Design Standard for Ballasted Single-Ply Roofing Systems (ANSI/SPRI, 2001). This standard outlines design procedures for ballasted systems for addressing wind loads on various building designs in locations across the country. This standard has proven its merits with these systems surviving major storm events including the hurricane seasons of 2004 and In recent years, new roofing membranes offering highly reflective surfaces have become the new rage of the industry, government, and code agencies. These membranes are used in fully adhered and mechanically fastened roof systems to take advantage of the reflective property of the membrane. With these systems offering aesthetically pleasing roofs that assist in saving energy for the building owner, ballast systems now seem a little old fashioned and out of step with the times. Is this truly the case or are there hidden attributes to the ballasted system that have not been identified? An eperimental study was initiated to quantify the energy savings of ballasted roofing systems and to compare their thermal performance to that of cool roof membranes. The eperimental design was structured to evaluate how mass of three different stone ballast weights and one paver ballast affected heat flu into the building and the buildup of the membrane surface temperature in comparison to the controls, in this case both a black and a white single-ply membrane. Eperimental work included the initial and subsequent occasional measure of reflectance and initial estimate of the emittance of the test samples, weekly organization of the temperature and heat flu data and the comparison of the ballast with the white and black membrane thermal performance. This work builds on the earlier work completed and published in The Field Performance of High-Reflectance Single-Ply Membranes Eposed to Three Years of Weathering in Various U.S. Climates (Miller et al, 2002), which was prepared by the Oak Ridge National Laboratory (ORNL) for The Single-Ply Roofing Industry (SPRI). This study investigated the reflectivity and thermal performance of single-ply membranes when eposed to the outdoor environment

4 The eperimental work is ongoing and it is anticipated that three years of eperimental data will subsequently be compiled. To date, two research papers have been written and presented for this research project. Gillenwater et al. (2005) gave details on the construction and instrumentation of the systems. They presented and discussed the measured membrane temperatures and insulation heat flues during the first year of monitoring before deployment of the reflective pavers. Desjarlais et al. (2006) reviewed the behavior of the membrane temperatures and insulation heat flues through two years of monitoring. Modeling the stone for its thermal performance is one of the goals of this project. The paver ballasts were included to aid in developing the model. If successful, this model could eventually allow ballast to be entered as a roof component in the DOE Cool Roof Calculator (Petrie et al. 2001, Petrie et al. 2004). Such a generalization of the eperimental results would permit the annual heating and cooling loads to be estimated for specific ballast configurations on roofs with various insulation levels located in different regions of the country. This report combines a summary of the eperimental work and describes the results of an effort to model the thermal performance of the ballasted systems through the first 24 months of the three-year project. The thermal and physical properties of the various configurations are used as input to the onedimensional transient heat conduction equation that is programmed in finite difference form in Simplified Transient Analysis of Roofs (STAR) (Wilkes 1989). STAR is the modeling tool used in the development of the DOE Cool Roof Calculator. The purpose of the modeling is to achieve good agreement between two predicted and measured quantities. One is the temperature of the control membranes and of the black membranes under the ballasts. The other is the heat flu at the interface between the two pieces of fiberboard that comprise the insulation in each system. Eperimental Facilities The Roof Thermal Research Apparatus (RTRA) at the Oak Ridge National Laboratory located in Oak Ridge, TN was constructed in the late 1980 s for documenting the effects of long-term eposure of small, low-slope roof test sections to the East Tennessee climate. The RTRA has four 4 ft by 8 ft openings in its roof to receive different instrumented low-slope roof test sections. Each test section may be divided into multiple areas. The original use of the RTRA showed in-service aging effects with CFC and alternative blowing agents for polyisocyanurate foam insulation boards in roofs covered by black and white membranes. In the late 1990 s, the RTRA was used to document the thermal performance of low-slope roofs coated with reflective coatings. Each test section was divided into 2 ft by 2 ft areas with as many as eight different surfaces on a test section. For this project, three of the four test sections are used. Each test section is divided into two 4 ft by 4 ft areas. One contains the ballast systems for the 10 lbs/ft 2 and 16.8 lbs/ft 2 tests. The second contains the 23.5 lbs/ft 2 tests, both stone and paver. The third contains the control systems, one with a black membrane and the other with a white membrane. Figure 1 is a photograph of the RTRA that shows the entire building including the weather station

5 Figure 1. Roof Thermal Research Apparatus with weather station A dedicated data acquisition system is housed inside the RTRA. It acquires the outside temperature and relative humidity and the wind speed and direction 10 ft above the roof of the RTRA. The total horizontal solar insolation and the total horizontal infrared radiation are measured at the top of the railing in Figure 1. There are also many dedicated input channels for thermocouples and for millivolt signals, such as those produced by heat flu transducers. Jack panels are conveniently located under the test sections on the inside of the RTRA walls to make for short lead wires from the test sections to the jack panels. Data are acquired under control of a database that is specific to each eperiment. The database instructs the data acquisition program as to what data to acquire and how often. Most channels are polled every minute. Data are stored in a compressed historical record. For ongoing eperiments, averages every 15 minutes of all variables are written weekly to a spreadsheet. Special reports can be generated for further detail on time dependency down to the frequency in the historical record. Test Sections Figure 2 is a photograph taken on top of the RTRA that shows the three test sections being used for the ballast systems project. The controls are in the foreground and the ballast systems are in the background beyond an uninstrumented area for unmonitored eposure of materials. To begin construction of the ballast systems, pavers 2-in. thick and 2 ft square were weighed on a scale to determine their weight per unit area. Three of the four pavers required for a 4 ft square test section were sawed in half in order not to have any seams at the center of the paver test section. A whole paver occupies the center and halves complete it. The required weight of stone in the test area to achieve the same loading as the pavers was determined for the heaviest stone. The lightest stone ballast loading was set at 10 lbs/ft 2, which is the minimum allowed for a ballast system, and it did supply 100 percent coverage of the membrane. The third ballast was set at the average of the heaviest and lightest. Buckets were used to carry the #4 stone from the scale to the roof of the RTRA where it was distributed inside frames to confine the ballast to its assigned area. Eactly enough stone was used to achieve the 10, 16.8, and 23.5 lbs/ft 2 loadings. Separate determinations were made of the weight of stone to eactly fill a bucket and the volume of the bucket. This yielded a density of 92.4 lb/ft 3 for the stone. Dividing each loading by the density of the stone yielded average thicknesses of 1.3, 2.2 and 3.1 in., respectively, for the three stone ballast systems

6 Due to the nature of the stone, the thicknesses vary over the area of each stone test section. Figure 2. Test sections configured for the ballast tests The instrumentation for each test section is shown in Figure 3. The metal decks are eposed to the conditions inside the RTRA, which is maintained year round between 70 and 75 F by an electric resistance heater and a small through-the-wall air conditioner. The membranes, in the case of the unballasted controls, or the top surfaces of the ballast, for the other test sections, are eposed to climactic conditions. Thermocouples on the decks and at the top of the test sections monitor the direct response to the imposed conditions. Additional thermocouples are at the internal interfaces. Wood fiberboard insulation 1.5 in. thickness is used to maimize sensitivity to differences among the test sections. At the interface between 1 in. thick and 0.5 in. thick pieces of insulation, a heat flu transducer (HFT) is embedded in the top of the thicker insulation board. Each HFT was especially calibrated in the same configuration. Thermocouples are deployed at the level of each HFT, 6 and 12 in. from its center to monitor if there is any significant heat flow in the horizontal direction. Thermocouples at the other levels are 6 in. from the center of the test section

7 Ballast (if present) Membrane Wood Fiberboard (0.5 and 1.0 in. thick) Metal Deck CLCL 6 in. CLCL 6 in. 6 in. 1 ft 6 in. 1 ft Thermocouples Heat Flu Transducer Figure 3. Thermocouple and heat flu transducer placement relative to the center of each test section The irregular upper surface of the stone-ballasted test sections presents a special challenge for monitoring surface temperature. Figure 4 shows the scheme that was adopted. Aluminum wire is strung across the middle of the frame from side to side in both directions. Thermocouples are attached to the wires with plastic wire ties. The lead wire between each measuring junction and its nearest wire tie is bent to hold the measuring junction against a stone at the top of each test section. At the top of the paver-ballasted test section a shallow hole was drilled into the top of the central paver, about 6 in. from its center, and the thermocouple was epoied in place with its measuring junction touching the bottom of the hole. Figure 4. Thermocouple measuring junctions placed against pieces of stone at the top of the stone-ballasted test sections - 7 -

8 A property of primary interest for understanding the thermal performance of roofs is the solar reflectance of the roof surface. It was measured for the surfaces of the test sections with two different techniques. For the smooth surfaced controls and the relatively smooth surfaced pavers, a Devices & Services solar spectrum reflectometer was taken onto the RTRA and used according to ASTM C , Standard Test Method for Determination of Solar Reflectance Near Ambient Temperature Using a Portable Solar Reflectometer (ASTM, 2004, Petrie, 2000, and Childs, 2001). Solar reflectance for the white TPO membrane at five locations on its surface averaged 0.78 for measurements on 12 March 2004 and decreased to 0.67 on 27 September On 18 April 2005, the solar reflectance was measured at For the black EPDM membrane, the average solar reflectance at five locations was 0.06 on 12 March 2004 and 0.09 on 27 September 2004 and 18 April Measurements on the central full-sized paver yielded 0.52 on 12 March 2004, 0.55 on 27 September 2004, and 0.50 on 18 April For the stone-covered test sections, a Davis Energy Group roof surface albedometer was taken onto the RTRA and used with guidance from ASTM E , Standard Test Method for Measuring Solar Reflectance of Horizontal and Low-Sloped Surfaces in the Field (ASTM, 1997). The albedometer measures the solar reflectance of a surface as the ratio of the output of a solar spectrum pyranometer when inverted (facing downward toward the surface) and facing upward during an interval of constant solar irradiance. The area of the ballasted test sections is only 4 ft by 4 ft, not the 13 ft by 13 ft recommended in E 1918 for use of the instrument. In order to minimize the effect of shadows from the assembly on the test section during use of the albedometer, a standard 20 in. height of the sensor above test sections is specified. It is achieved by the support stand that is part of the assembly. Because of the relatively small size of the ballasted test sections, the standard height was relaed. A special guide was made to achieve heights of 10 in., 15 in., and 20 in. above the surfaces while manually holding and leveling the pyranometer and its support arm long enough for a steady response from the millivolt meter that monitors the output of the pyranometer. Apparent solar reflectance was measured at these three heights. Shape factor algebra yielded the fraction of the pyranometer s view taken up by the stone. The remainder is surroundings at some constant but unknown reflectance. The reflectance of the surroundings was varied by trial-and-error until the reflectance of the stone was constant with height of the pyranometer above the surface. It was concluded that the solar reflectance for the stone ballast is 0.21 ± 0.01 on 19 March Repeat measurements on 29 March 2005 indicated that the average ballast solar reflectance was A property of secondary interest for the thermal performance of roofs is the infrared emittance of the roof surface. It is difficult to measure for thermally massive systems, especially the irregular surfaces of the stone-ballasted systems. In general, non-metallic surfaces have infrared emittance near 0.9. This value is assumed to apply to all the test sections in the ballast system study and has been verified often in the SPRI study for single-ply white and black membranes. Eperimental Results The ballast study went live on March 12, 2004 with the start of the data collection that has continued uninterrupted for one hundred weeks through April Figure 5 depicts the membrane temperature and heat flues through the si roof assemblies for a twenty-four hour period on 5 April These data start at midnight (Hour 0) and were compiled approimately one month after the eperiments started

9 Bare Black EPDM Bare White TPO Under 10 lb/ft² Stone Under 17 lb/ft² Stone Under 24 lb/ft² Stone Under Paver Membrane Temperature ( F) Hours into 4/5/ Hours into 4/5/2004 Heat Flu through Insulation [Btu/(h ft²)] Figure 5. Membrane temperatures and roof heat flues one month into eperimental program. In Figure 5, the black membrane shows the maimum temperature, peaking at approimately 146ºF. The white membrane shows the lowest peak temperature with a maimum temperature of 86ºF. Between the two etremes are the three ballast systems and the paver system. The three ballast systems have peak temperatures of 103, 95, and 90ºF respectively. The paver system has approimately the same peak temperature as the ballast system having the same areal density. The ballast systems show a delay in the peak temperature and that delay ranges from 30 minutes to two hours. The delay increases as the areal density increases. The thermal inertia of the ballast and paver systems can also be seen in their nighttime behavior keeping the massive systems warmer throughout the nighttime hours. The heat flu data in Figure 5 corroborates the temperature data described above. Peak heat flues are arranged in the identical order as the peak temperatures shown above. After one month of performance, the white membrane has a lower peak temperature and heat flu suggesting that it is outperforming the other roof systems in terms of reducing cooling energy loads and peak demand. Figures 6, 7 and 8 depict the membrane temperature and heat flues through the si roof assemblies for a twenty-four hour period on 4 October 2004, 21 March 2005, and 5 October 2005, respectively. These data start at midnight (Hour 0) and were compiled approimately seven months after the eperiments started. After seven months in service, the black membrane continues to be the warmest membrane reaching a peak temperature of 150ºF. Unlike the initial results, the white membrane no longer is the coolest surface. After only seven months of service the 23.5 lbs/ft 2 ballast loading as well as the paver now have peak temperatures that are lower than the white membrane. The 16.8 lbs/ft 2 ballast loading has almost an equal peak temperature to the white membrane. Similar to what was seen after the roofs were commissioned, the ballast loadings and pavers delay the peak temperatures and heat flues. Delays from one to three hours are measured with the delays increasing with increasing levels of mass. It is interesting to note that the heaviest ballast system and the paver have identical areal densities but - 9 -

10 substantially different solar reflectances. The ballast and paver have solar reflectances of 0.22 and 0.55, respectively after seven months. These observations strongly suggest that the controlling parameter is the mass and not the solar reflectance. Bare Black EPDM Bare White TPO Under 10 lb/ft² Stone Under 17 lb/ft² Stone Under 24 lb/ft² Stone Under Paver Membrane Temperature ( F) Hours into 10/4/ Hours into 10/4/2004 Heat Flu through Insulation [Btu/(h ft²)] Figure 6. Membrane temperatures and roof heat flues seven months into eperimental program. From a heat flu perspective, the temperature trends noted above are duplicated. The 23.5 lbs/ft 2 ballast loading as well as the paver now have peak heat flues that are lower than the white membrane (4.5 vs. 7.0 Btu/hr ft 2 ). After seven months, the black membrane has the highest peak heat flu of 17.0 Btu/hr ft 2. After one year of eposure in East Tennessee (see Figure 7), the black membrane continues to be the warmest membrane reaching a peak temperature of 146ºF. Similar to the seven month results, the 23.5 lbs/ft 2 ballast loading as well as the paver have peak temperatures that are 9ºF lower than the white membrane. In addition, the 16.8 lbs/ft 2 ballast loading has a slightly lower peak temperature than the white membrane (approimately 1ºF cooler). After twelve months, the black membrane continues to ehibit the highest peak heat flu of 16.8 Btu/hr ft 2. The 23.5 lbs/ft 2 ballast loading as well as the paver have peak heat flues that are 4.2 and 4.6 Btu/hr ft 2. These peaks are lower than the white membrane (7.3 Btu/hr ft 2 ) and approimately equal to the 16.8 lbs/ft 2 ballast loading. After nineteen months of eposure in East Tennessee (see Figure 8), all of the conclusions drawn after one year continue to be true. We see the heaviest ballast system and pavers outperforming the white membrane (from a cool roof perspective) and the medium weight ballast system having very similar performance as the white membrane. As the white membrane continues to age, its solar reflectance should continue to drop (Miller et al, 2002) while little change is anticipated in any of the ballast systems. Since these roofing systems are epected to have a service life well in ecess of ten years, the

11 performance will continue to eceed that of the white membrane system for over 90% of their service lives. Membrane Temperature ( F) Bare Black EPDM Under 10 lb/ft² Stone Under 24 lb/ft² Stone Bare White TPO Under 17 lb/ft² Stone Under Paver Hours into 3/21/2005 Hours into 3/21/2005 Figure 7. Membrane temperatures and roof heat flues twelve months into eperimental program Heat Flu through Insulation [Btu/(h ft²)] Membrane Temperature ( F) Bare Black EPDM Under 10 lb/ft² Stone Under 24 lb/ft² Stone Bare White TPO Under 17 lb/ft² Stone Under Paver Hours into 10/5/2005 Hours into 10/5/ Heat Flu through Insulation [Btu/(h ft²)] Figure 8. Membrane temperatures and roof heat flues nineteen months into eperimental program

12 Measured Seasonal Average Membrane Temperatures and Insulation Heat Flues Another way of looking at the eperimental data is to sum it into seasonal periods. Although the daily analyses clearly demonstrates that the peak energy demand of ballasted systems is equal to or lower than a cool roof membrane, shifts in the loads make it difficult to assess whether any energy savings are obtained with the ballasted systems. To determine whether there were any energy savings, the hourly averages of membrane temperature and insulation heat flu were then averaged over seasons. The seasons are defined as the intervals from 4/20/2004 through 10/19/2004 (summer 2004), 10/20/2004 through 4/20/2005 (winter 2004) and 4/21/2005 through 10/20/2005 (summer 2005), and 10/21/2005 through 4/20/2006 (winter 2005). Table 1 contains the seasonal average measured temperature for the membrane of each system, the average measured heat flu directed inward through the insulation and the average measured heat flu directed outward through the insulation. Average membrane temperatures in summer 2004 show that the eposed white membrane was coolest and the eposed black membrane was warmest. The membrane temperatures under the ballasted systems are between these etremes in order of the loading. The 10# stone is warmest and the paver is the coolest. The paver has the same loading as the 24# stone but it has higher solar reflectance than the stone, leading to cooler temperatures. The membrane temperature is warmer for the paver than the 24# stone in summer 2005 but not in summer The solar reflectance of the uncoated paver decreased from summer 2004 to summer 2005 but remained higher than that of the stone. The white membrane shows temperature changes due to its decreasing solar reflectance. In summer 2005 the white membrane is over 3 F warmer on average than it was in summer This difference is affected by the different climatic conditions in summer 2004 and summer However, the black membrane is only about 1.4 F warmer in summer 2005 than in summer 2004 so the average climatic conditions were only slightly different. Table 1. Measured membrane temperatures and inward and outward heat flues averaged over summer 2004, winter 2004, summer 2005 and winter Measurements T membrane, F q inward, Btu/(h ft² ) q outward, Btu/(h ft² ) Average summer 2004 White Black # # # Paver Average winter 2004 White Black # # # Paver

13 Average summer 2005 White Black # # # Paver Average winter 2005 White Black # # # Paver The membrane temperatures in the winter period, comprising late fall, winter and early spring in East Tennessee, are not ordered quite like they are in the summer periods. The eposed white membrane is cooler and the eposed black membrane is warmer than the black membranes under the stone ballasts. However, the membrane under the paver and the eposed white membrane have about the same average temperature. The membrane of the 10# system is cooler than the membranes of the 17# and 24# systems. The purpose of separating the heat flues into those directed inward and those directed outward is to focus on solar effects. Inward heat flues are driven by high membrane temperatures due to solar loads. Outward heat flues occur generally in the absence of solar loads ecept during cold clear winter days. The inward heat flues during the summer periods show that the heavy ballasted systems outperform the white system on average. The benefit of their thermal mass outweighs their lower solar reflectance. This ordering is the same in the winter period when inward heat flues might be beneficial for decreasing heating loads. The lower solar reflectance of the white membrane in summer 2005 compared to summer 2004 allows the 17# system to have lower inward heat flu than the white system in summer 2005 but not in summer The outward-directed heat flues for the eposed white and black membranes are essentially equal for each period but the values vary with the season and year. They are slightly larger than any of the averages for the ballasted systems. This is consistent with the effect of eposed thermal mass, which retains daytime solar and releases it slowly at night. The heat flu transducers in the insulation do not see this release to the outside, but are affected the same as the membrane temperatures by solar effects that penetrate through the ballasts. Properties Needed to Predict Membrane Temperatures and Insulation Heat Flues The program STAR requires data for the boundary conditions at the inside and outside surfaces of a low-slope roof system. The temperature measured at the bottom of the insulation was used as the inside boundary condition. Modeling of the deck and the heat transfer under each system was not done. A weather station on the eperimental facility provided local climatic conditions. They were acquired continuously along with data for the thermal performance of each system. The climatic conditions were

14 used in STAR to impose convection and thermal radiation as the boundary condition at the outside of each system. STAR also requires a layer by layer description of the physical and thermal properties of a low-slope roof system. Figure 3 is a sketch of a typical ballasted system. Table 1 lists the properties for initial runs of STAR. Data are listed for the three loadings of stone (10#, 17# and 24#), the paver, the eposed white and black membranes, and the two layers of wood fiberboard insulation that are used in each system. Direct measurements were made of the thickness and density of the various components of the systems. The weight of several pavers was measured by a scale and divided by the measured volume to yield density. A 5-gallon bucket was weighed, filled with stone, weighed again, and emptied. The volume of the bucket was then determined by measuring the weight of water to fill it. The weight of stone divided by this volume yields the average density of the stone including the effect of the air spaces around the individual stones. Table 2 includes ranges of solar reflectance for all surfaces. Table 2a gives seasonal variation for the eposed smooth surfaces (paver and white and black membranes). These solar reflectances were measured at about si month intervals during the project according to ASTM C , Standard Test Method for Determination of Solar Reflectance Near Ambient Temperature Using a Portable Solar Reflectometer (ASTM 2002). The solar reflectance of the stone was measured at the beginning of each year of the project according to ASTM E , Standard Test Method for Measuring Solar Reflectance of Horizontal and Low-Sloped Surfaces in the Field (ASTM 1997). All the eposed surfaces are non-metallic solids for which the infrared emittance is taken to be 0.9 from previous measurements and eperience (Petrie et al. 2001). The thermal conductivity and heat capacity of the white and black membranes and the fiberboard were obtained from the literature and our own measurements. For the stone and pavers, the program Properties Oak Ridge (PROPOR) was used as part of the ongoing analysis of the eperimental data to estimate thermal conductivity and Table 2: Properties input to STAR for initial modeling of the ballasted and control systems 10# 17# 24# Paver White Black Fiberboard Loading, lb/ft² negl. negl. n.a. Thickness, in ,1.0 Thermal conductivity, **a+b T Btu in./(h ft² F) ±6 ±7 ±2 ±4 Density, lb/ft Heat capacity, Btu/(lb F) ±0.2 ±0.3 ±0.1 ±0.04 Infrared emittance, % not needed *Solar reflectance, % to to 62 8 to 9 not needed * Ranges, if given, span observed variation in the first 18 months of the project ** From guarded hot plate measurements: k fiberboard = T( F)

15 TABLE 2a: Variation of solar reflectance for the smooth surfaces in the project Solar reflectance, % Summer 2004 Winter 2004 Summer 2005 White control Black control Paver volumetric heat capacity (the product of density and heat capacity). PROPOR compares the temperatures and heat flues that are measured inside a system to those predicted by the transient heat conduction equation. Temperatures measured at the outside and inside surfaces are boundary conditions. Thermal conductivity and volumetric heat capacity are considered parameters. Values of the parameters are adjusted by an automated iteration procedure until best agreement is obtained. Best agreement is defined as the minimum of the squares of the differences between measured and predicted temperatures and heat flues inside the systems. An estimate of the confidence in the final parameter values is included as part of the output from the program (Beck et al. 1991). PROPOR had difficulty converging to estimates of the thermal conductivity and volumetric heat capacity for the 10# and 17# loadings of stone ecept for several weeks during winter in East Tennessee. Even then the estimates were not acceptably precise. Convergence for the 24# loading was less difficult. Convergence was obtained for the paver no matter what the weather conditions. PROPOR needs the temperature at the outside surface as a boundary condition in order to estimate parameters for the ballasts. Measurements of the temperature at the outside surface were done for the ballasted systems by placing thermocouple measuring junctions against two stones on each stone surface and by burying a measuring junction slightly below the outside surface of the paver (Gillenwater et al. 2005). This procedure is not likely to yield a reliable surface temperature for the light loadings of stone when the sun is high in the sky. At these times sunlight will penetrate to the black membrane and cause its temperature to generally eceed the average for two stones on the surface. The thermal conductivity and heat capacity for the stone and paver in Table 1 are the averages of estimates from PROPOR for weeks when it converged. To illustrate the difficulty PROPOR had with the data from the stone, the uncertainty is appended to these estimates. Heat capacity is obtained by dividing the volumetric heat capacity by the measured density. Only the volumetric heat capacity is used by PROPOR and by STAR. The uncertainty in the estimates for the stone is of the order of 50% to 150% of the estimates themselves. This indicates that it is difficult for PROPOR to estimate parameters in the thermal conduction equation to describe the thermal behavior of the stone ballasts. The difficulty is judged to be caused by the inaccuracy of measured temperature at the top of the ballast and convection effects in the stone ballast when solar loads are high. Handbook values for the thermal conductivity of ballast vary greatly. Assuming limestone is typical of the composition, thermal conductivity varies from 6.5 Btu in./(h ft² F) (ASHRAE 2005) to 13.2 Btu in./(h ft² F) (CRC 1973). These values are the same magnitude as the values for the stone and paver in Table 1. As a limit in the air around the stones, the thermal conductivity of still air at 75 F (24 C) and 1 atm is 0.18 Btu in./(h ft² F). The ASHRAE handbook value of 0.19 to 0.24 Btu/(lb F) for the heat capacity of heavyweight concrete and the generally accepted heat capacity of 0.24 Btu/(lb F) for air compare well to the values for the ballast in Table

16 For the boundary conditions needed by STAR and for direct comparison to its predictions, hourly averages were generated from data acquired continuously at one-minute intervals during the project. For this report, to judge the accuracy of the predictions, the hourly averages of membrane temperature and insulation heat flu were then averaged over seasons. The seasons are defined as the intervals from 4/20/2004 through 10/19/2004 (summer 2004), 10/20/2004 through 4/20/2005 (winter 2004) and 4/21/2005 through 10/20/2005 (summer 2005). Table 1 contains the seasonal average measured temperature for the membrane of each system, the average measured heat flu directed inward through the insulation and the average measured h eat flu directed outward through the insulation. Seasonal Average Differences Between Predictions and Measurements Using Initial Estimates of Properties STAR was eecuted with the properties in Table 2 and the solar reflectances in Table 2a in order to predict membrane temperatures and insulation heat flues for summer 2004, winter 2004 and summer In light of the large uncertainty in thermal conductivity and heat capacity for the stone, the initial task was to select appropriate values for them. Hourly predicted membrane temperatures and insulation heat flues were entered in a spreadsheet that contained the hourly averages of the measurements. The spreadsheet was organized into worksheets for each si month period to facilitate calculation of seasonal averages. Table 3 lists the differences between the average predicted temperatures and heat flues for these three seasons using the properties in Table 2 and Table 2a and the average measured temperatures and heat flues in Table 1. Ideally, all differences should be zero. This is not a reasonable epectation, even for the simple control systems. The solar intensity input to STAR is diminished by cloudiness, but effects of rain are not modeled. For several hours after rain events water ponded on the stone-ballasted membranes and the eposed black membrane. East Tennessee has a typical mied climate. Heavy dew/frost occurs during many nights and usually takes until mid-morning to disappear from eposed surfaces. No moisture effects were modeled. Because of the small differences in Table 3 between predicted and measured membrane temperatures and inward heat flues for the white system, the conclusion is that STAR does an ecellent job of predicting the thermal performance of this simple system. Membrane temperature and inward heat flu differences for the black system are consistently positive and larger than for the white system. The white and black systems shared the same test panel, with the black system at the low end. Differences for the paver system are about the same as those for the black control. STAR predicts average membrane temperatures for the stone-ballasted systems that are 6 to 10 F warmer than the measurements. Inward heat flues are 2 to 4 Btu/(h ft²) larger than the corresponding measurements. This is not surprising in light of the large uncertainty in the thermal conductivity and heat capacity of the stone systems. Additional predictions with different properties are warranted. Differences between predicted and measured average outward heat flues are shown in the last column of Table 3. The black control shows the largest magnitude of differences in all periods. STAR appears capable of doing an adequate job for all systems in the absence of solar effects, if it is provided with suitable values of system properties. TABLE 3: Differences between predicted and measured membrane temperatures, inward heat flues and outward heat flues for summer 2004, winter 2004 and summer

17 Predicted - Measured: T membrane, F q inward, Btu/(h ft² ) q outward, Btu/(h ft² ) Average summer 2004 White Black # # # Paver Average winter 2004 White Black # # # Paver Average summer 2005 White Black # # # Paver Averages over si month periods do not show how well STAR models the daily peak membrane temperatures and insulation heat flues. Moreover, averages do not retain any details that show how well STAR models the diurnal variations in the temperatures and heat flues. To show such features, graphs were prepared for several clear days during the first 18 months of the project. Clear days show maimum solar effect and have smooth curves through the hourly temperatures and heat flues with few deviations caused by cloudiness that make it difficult to visually compare the data. The relatively simple thermal behavior of eposed white and black membranes over a low-slope roof with low thermal mass is well understood from previous eperience with test sections used to validate STAR for the DOE Cool Roof Calculator (Petrie 2001). On the several clear days the hourly predictions for the eposed white membrane were in good agreement with the measurements and consistent with our understanding. The hourly behavior of the eposed black membrane, when compared to that from previous eperience, indicated that the measured temperatures were accurate but the measured heat flues might be low. Temperatures and heat flues were measured independently with thermocouples and small heat flu transducers, respectively. More uncertainty in measured heat flues is consistent with our eperience. It occurs despite calibration of the heat flu transducers in the test materials according to ASTM C , Standard Test Method for Steady-State Heat Flu Measurements and Thermal Transmission Properties by Means of the Heat Flow Meter Apparatus (ASTM 1998)

18 The shape of the predicted curves on the clear days was correct for the control systems, with low thermal mass and either an eposed white or black membrane. Predicted peak times coincided with the measured peak times. The nighttime predictions were generally low for both these controls as outward heat flues in Table 3 show. This is likely due to the effects of condensation and no attempt was made to model its effect. Regarding diurnal behavior of predictions for the ballasted systems with properties in Table 1, the 10# system was the only one for which peak times coincided with the corresponding measured peak times. Predicted peaks for the 17# and 24# systems occurred later and those for the paver occurred earlier than measured peaks. Predicted peak values for all ballasted systems were higher than the corresponding peak measurements. Trials, in which density was held at the measured values, indicated that peak times are most sensitive to heat capacity. If heat capacity is increased, peak time is delayed. Peak values are most sensitive to thermal conductivity. If thermal conductivity is decreased, the peak membrane temperature and insulation heat flu also decrease. However, changes in heat capacity affect peak values and changes in thermal conductivity affect peak times to some etent. Seasonal Average Differences between Predictions and Measurements Using Final Estimates of Properties STAR was eecuted with thermal conductivity values for the stone and paver that were varied as a percentage of the values in Table 1 in 5% increments. Best agreement between predictions and measurements was judged to occur for thermal conductivity corresponding to 10%, 15%, 20% and 20% of Table 1 values for the 10#, 17#, 24# and paver systems, respectively. The ASHRAE Handbook of Fundamentals (ASHRAE 2005) lists thermal conductivity of 140 lb/ft 3 concrete to range between 9-18 Btu in./(h ft² F). Heat capacity for the stone was set to 0.10 Btu/(lb m F) by trial-and-error. For the paver it was set to 0.21 Btu/(lb m F). The ASHRAE Handbook of Fundamentals (ASHRAE 2005) lists 0.19 to 0.24 Btu/(lb m F) as the range for heavyweight concretes, yielding a geometric mean of 0.21 Btu/(lb m F). Table 4 lists the complete set of property values. Table 2a was used for the seasonal variation of solar reflectance. Table 5 shows the differences between predictions and measurements using these final values for properties. The differences for the eposed white and black membranes are repeated from Table 3. In the summer seasons, the differences between predictions and measurements of membrane temperatures under the stone-ballasted systems are smaller in Table 5 than corresponding differences in Table 3. The differences in the membrane temperatures for the paver are about the same in both tables. All eceed the differences for the eposed black and white membranes. In Table 5 the differences in the inward heat flues for all the ballasted systems are smaller than the differences for the eposed black membrane and about the same as those for the eposed white membrane. In Tables 3 and 5 differences for the outward heat flues for all the ballasted systems are about the same magnitude but opposite sign. Table 4: Properties input to STAR for final modeling of the ballasted and control systems 10# 17# 24# Paver White Black Fiberboard Loading, lb/ft² negl. negl. n.a. Thickness, in ,1.0 Thermal conductivity, Btu in./(h ft² F) **a+b T

19 Density, lb/ft Heat capacity, Btu/(lb F) Infrared emittance, % not needed *Solar reflectance, % to to 62 8 to 9 not needed * Ranges, if given, span observed variation in the first 18 months of the project ** From guarded hot plate measurements: k fiberboard = T( F) Table 5: Differences between predicted and measured membrane temperatures, inward heat flues and outward heat flues for summer 2004, winter 2004, summer 2005 and winter 2005 using properties in Table 4 and Table 2a Predicted Measured: T membrane, F q inward, Btu/(h ft²) q outward, Btu/(h ft²) Average summer 2004 White Black #: 0.1 Btu/(lb m F) #: 0.1 Btu/(lb m F) #: 0.1 Btu/(lb m F) Paver: 0.21 Btu/(lb m F) Average winter 2004 White Black #: 0.1 Btu/(lb m F) #: 0.1 Btu/(lb m F) #: 0.1 Btu/(lb m F) Paver: 0.21 Btu/(lb m F) Average summer 2005 White Black #: 0.1 Btu/(lb m F) #: 0.1 Btu/(lb m F) #: 0.1 Btu/(lb m F) Paver: 0.21 Btu/(lb m F) In the winter season for the ballasted systems, differences for membrane temperatures and outward heat flues are larger in Table 5 than in Table 3. However, differences for inward heat flues are improved. It is concluded that the ballasted systems, especially the stone-ballasted systems, are modeled well enough by the transient heat conduction equation in winter using the same properties as in summer. Since the purpose of the project is to compare ballasted systems to cool roofs, modeling behavior under summer conditions is most important. The generally small differences for inward heat flues in the summer seasons for the ballasted systems indicate successful modeling of the most important thermal behavior. Several clear days were again selected to show the peak behavior and diurnal variation of the predicted

20 and measured membrane temperatures and insulation heat flues. Fig. 9, for July 21, 2004, displays the best agreement that was found. In terms of both peak values and diurnal variation, predicted membrane temperatures agree very well with the measurements for all systems. The predicted peak insulation heat flu for the eposed black membrane is larger than the measured peak, consistent with the comment above about our eperience with heat flu transducers. Predicted peak times for the stone-ballasted systems are slightly later than the measurements but nighttime agreement is good. Generally, for all days and all ballasts, there are anomalies in the measurements that a model like STAR, with relatively few parameters, cannot duplicate. However, overall agreement between predictions and measurements, as indicated by Table 5 and Fig. 9, is good enough to conclude that STAR, with appropriate property values for ballasted systems, can model their thermal behavior relative to that of a system with an eposed white membrane. This is essential for generating comparisons between ballasted systems and conventional cool systems under conditions different from test nditions. Discussion The fall 2004 readings shown in Figure 6 now show the heaviest ballast and paver assemblies peaking in Membrane Temperature ( F) Under 10# Under 24# Bare White + STAR 10# STAR 24# STAR White Under 17# Under Paver Bare Black STAR 17# STAR Paver STAR Black Hours into 7/21/2004 Hours into 7/21/2004 Fig. 9: Comparison of membrane temperatures and insulation heat flues using ballast properties in Table 4 for July 21, 2004 Heat Flu through Insulation [Btu/(h ft²)] temperature at lower levels than the white membrane. The intermediate weight assembly is peaking at a temperature that is basically the same as the white membrane. At this point in the roof systems lives, the white membrane is still above the ENERGY STAR minimum reflectance value of 0.65 indicating that the ballast systems are performing as a cool roof. After eight months of eposure in East Tennessee s climate, the white single-ply membrane control lost 15 percent of its reflectance, a similar rate as seen in the earlier study referenced above that showed the eposed single-ply membranes lost thirty to fifty percent of their reflectance after three years of outdoor eposure. Cool roofs do have a market benefit in that they can be advertised as complying with the Environmental Protection Agency s (EPA s) ENERGY STAR Roofing Products. In addition, California s