On the Flow over High-rise Building for Wind Energy Harvesting: An Experimental Investigation of Wind Speed and Surface Pressure

Featured Application: Flow features on the roof region of an isolated high-rise building for urban wind energy harvesting; wind loading over high-rise building roof. Abstract: The human migration from rural to urban areas has triggered a chain reaction causing the spiking energy demand of cities worldwide. High-rise buildings ﬁlling the urban skyline could potentially provide a means to improve the penetration of renewable wind energy by installing wind turbines at their rooftop. However, the above roof ﬂow region has not received much attention and most results deal with low-rise buildings. This study investigates the ﬂow pattern above the roof of a high-rise building by analysing velocity and pressure measurements performed in an atmospheric boundary layer wind tunnel, including four wind directions and two di ﬀ erent roof shapes. Comparison of the surface pressure patterns on the ﬂat roof with available low-rise building studies shows that the surface pressure contours are consistent for a given wind direction. At 0 ◦ wind direction, a separation bubble is detected, while cone vortices dominate at 30 ◦ and 45 ◦ . The determining factor for the installation of small wind turbines is the vicinity to the roof. Thus, 45 ◦ wind direction shows to be the most desirable angle by bringing the substantial ampliﬁcation of wind and keeping the turbulence intensity low. Decking the roof creates favourable characteristics by overcoming the sensitivity to the wind direction while preserving the speed-up e ﬀ ect.


Introduction
The world has been rapidly urbanising for the past five decades, with~54% of the population currently living in urban areas, expected to reach~66% by 2050 [1]. As a result, cities are transforming with more high-rise buildings being planned to cope with the increasing dwelling demand, living the presented work brings more insight into the flow-specific characteristics, particularly important for urban wind exploitation. It also provides a comparison to the flow pattern observed above flat-roof low-rise buildings, in terms of flow structures and augmented wind speed.

High-Rise Building Models
A 1:300 scale model of a high-rise building, square in plan, has been used in this investigation. Two roof shapes are considered: a flat roof and a deck-type roof as shown in Figure 1a,b, respectively. As shown in Figure 1a, the height of the building is denoted by H (400 mm) and the width by D (133.3 mm).

High-rise Building Models
A 1:300 scale model of a high-rise building, square in plan, has been used in this investigation. Two roof shapes are considered: a flat roof and a deck-type roof as shown in Figure 1a,b, respectively. As shown in Figure 1a, the height of the building is denoted by H (400 mm) and the width by D (133.3 mm).
In the context of the atmospheric boundary layer winds, low-rise buildings are defined as those having two heights smaller than the width (2H < D) and H < 30 m [46]. This matches the definition of squat buildings given by Cook [47], whereas the building is considered tall when 2H > D. This distinction is made based on the behaviour of the wind and its path to overcome the building as an obstacle. When the building is tall, the wind finds it easier to flow around the sides than over the top of the building, except for an area very close to the top, as opposed to squat buildings [47]. In recent literature, related to the flow around buildings, the term high-rise building is used for buildings having a height-to-width ratio of H/D ≥ 2:1, as in [48,49]. Thus, the ratio of H/D = 3:1 is chosen as an example for the geometry of a high-rise building. The inclination of the deck roof in Figure 1b is 30°. The dimensions of the deck roof are the same as those of the flat-roof building with the added cap to the flat roof. The hypothesis here is that the deck-type roof helps to reduce both the separated flow and its turbulence intensity in the above roof flow and hence enhance the quality and mean speed of the wind above the roof for wind energy harvesting. The flow pattern around the two buildings was investigated under four different wind angles, namely 0°, 15°, 30° and 45°.

Wind Tunnel Setup and Incoming Flow
The experiments have been conducted in the atmospheric boundary layer wind tunnel of the Ruhr-University Bochum, Germany. The wind tunnel has a cross-section of 1.6 × 1.8 m 2 and a test section length of 9.4 m. Figure 2 shows the wooden models mounted on a rotating table in the wind tunnel. The reference velocity Uref used to normalise the data was measured using a Prandtl tube mounted at the height of the high-rise building model and placed 1 m upstream of the model. Reynolds number related to all measurements based on the mean velocity at the height of high-rise building and the side of the roof is Re ~ 1.4 × 10 5 . Horizontal buoyancy is found to be absent in the wind tunnel, as very low decay of the static pressure is measured throughout the measurement section [50]. In the context of the atmospheric boundary layer winds, low-rise buildings are defined as those having two heights smaller than the width (2H < D) and H < 30 m [46]. This matches the definition of squat buildings given by Cook [47], whereas the building is considered tall when 2H > D. This distinction is made based on the behaviour of the wind and its path to overcome the building as an obstacle. When the building is tall, the wind finds it easier to flow around the sides than over the top of the building, except for an area very close to the top, as opposed to squat buildings [47]. In recent literature, related to the flow around buildings, the term high-rise building is used for buildings having a height-to-width ratio of H/D ≥ 2:1, as in [48,49]. Thus, the ratio of H/D = 3:1 is chosen as an example for the geometry of a high-rise building.
The inclination of the deck roof in Figure 1b is 30 • . The dimensions of the deck roof are the same as those of the flat-roof building with the added cap to the flat roof. The hypothesis here is that the deck-type roof helps to reduce both the separated flow and its turbulence intensity in the above roof flow and hence enhance the quality and mean speed of the wind above the roof for wind energy harvesting. The flow pattern around the two buildings was investigated under four different wind angles, namely 0 • , 15 • , 30 • and 45 • .

Wind Tunnel Setup and Incoming Flow
The experiments have been conducted in the atmospheric boundary layer wind tunnel of the Ruhr-University Bochum, Germany. The wind tunnel has a cross-section of 1.6 × 1.8 m 2 and a test section length of 9.4 m. Figure 2 shows the wooden models mounted on a rotating table in the wind tunnel. The reference velocity U ref used to normalise the data was measured using a Prandtl tube mounted at the height of the high-rise building model and placed 1 m upstream of the model. Reynolds number related to all measurements based on the mean velocity at the height of high-rise building and the side of the roof is Re~1.4 × 10 5 . Horizontal buoyancy is found to be absent in the wind tunnel, as very low decay of the static pressure is measured throughout the measurement section [50]. The boundary layer flow is generated in the wind tunnel using both spires at the tunnel inlet and roughness elements as shown in Figure 2a. The mean wind profile matches that of a power law with the exponent of 0.2 as shown in Figure 3a. This wind profile is found to be consistent in the lateral direction of the wind tunnel, covering the central portion of the test section. This spans over approximately 0.9 m at least (±0.45 m from the central position), which is outside of the influence of the walls [51]. This is representative of the terrain category II [52] simulating realistic conditions of the flow around isolated high-rise buildings. Further, measured data can be used as an approximation for the flow pattern in case of an urban area with a dominant high-rise building surrounded by low-rise buildings. This arrangement is common on the outskirts of large cities, university campuses displaced from city centres, etc., all representing potential locations for efficient urban wind energy harvesting. The mean stream-wise wind speed (Uref), the stream-wise turbulence intensity (IU) and the vertical turbulence intensity (IW) at the height of the model are 16 m/s, 13% and 11%, respectively.  The boundary layer flow is generated in the wind tunnel using both spires at the tunnel inlet and roughness elements as shown in Figure 2a. The mean wind profile matches that of a power law with the exponent of 0.2 as shown in Figure 3a. This wind profile is found to be consistent in the lateral direction of the wind tunnel, covering the central portion of the test section. This spans over approximately 0.9 m at least (±0.45 m from the central position), which is outside of the influence of the walls [51]. This is representative of the terrain category II [52] simulating realistic conditions of the flow around isolated high-rise buildings. Further, measured data can be used as an approximation for the flow pattern in case of an urban area with a dominant high-rise building surrounded by low-rise buildings. This arrangement is common on the outskirts of large cities, university campuses displaced from city centres, etc., all representing potential locations for efficient urban wind energy harvesting. The mean stream-wise wind speed (U ref ), the stream-wise turbulence intensity (I U ) and the vertical turbulence intensity (I W ) at the height of the model are 16 m/s, 13% and 11%, respectively. The boundary layer flow is generated in the wind tunnel using both spires at the tunnel inlet and roughness elements as shown in Figure 2a. The mean wind profile matches that of a power law with the exponent of 0.2 as shown in Figure 3a. This wind profile is found to be consistent in the lateral direction of the wind tunnel, covering the central portion of the test section. This spans over approximately 0.9 m at least (±0.45 m from the central position), which is outside of the influence of the walls [51]. This is representative of the terrain category II [52] simulating realistic conditions of the flow around isolated high-rise buildings. Further, measured data can be used as an approximation for the flow pattern in case of an urban area with a dominant high-rise building surrounded by low-rise buildings. This arrangement is common on the outskirts of large cities, university campuses displaced from city centres, etc., all representing potential locations for efficient urban wind energy harvesting. The mean stream-wise wind speed (Uref), the stream-wise turbulence intensity (IU) and the vertical turbulence intensity (IW) at the height of the model are 16 m/s, 13% and 11%, respectively.

Experimental Procedure
In addition to the velocity measurements, the surface pressure has also been obtained at different locations to present a comprehensive database where pressure and wind speed results can be used as a validation test case.
The surface pressure on the roofs of the models was measured using 64 pressure taps, distributed as shown in Figure 4. The pressure sensors are four active element piezoresistive bridges. Two different pressures sensors were used: Honeywell 170 PC (measurement range: ±35 mbar) and AMSYS 5812-0001-D-B (measurement range: ±10.34 mbar). Due to the narrow spaces in the model, the pressure sensors were placed outside of the model and connected to the bores in the wooden deck by brass connectors (inner diameter 1 mm) and plastic pressure tubes (inner diameter 1.5 mm), that are freed from burrs and debris. The length of plastic pressure tubes was about 0.9 m. The calibration of the pressure measurement system included two phases: static and dynamic. The static calibration was performed to establish the pressure-voltage relation for each pressure sensor, while dynamic calibration was performed to correct the dynamic effects of tubes [53]. Surface pressures were acquired with a sampling frequency of 1000 Hz and were scanned in a sample-and-hold modus, enabling the simultaneous sampling of signals. For each measurement, 28 samples of 4.7 s length were collected, where each sample corresponds to a 10 min long sample in full scale. The maximum uncertainty of surface pressures based on five repeated measurements was estimated to be~2.5%. More details about the pressure measurement equipment that consisted of pressure sensors, a tubing system, amplifiers and analog/digital (A/D) converters can be found in [54].

Experimental Procedure
In addition to the velocity measurements, the surface pressure has also been obtained at different locations to present a comprehensive database where pressure and wind speed results can be used as a validation test case.
The surface pressure on the roofs of the models was measured using 64 pressure taps, distributed as shown in Figure 4. The pressure sensors are four active element piezoresistive bridges. Two different pressures sensors were used: Honeywell 170 PC (measurement range: ±35 mbar) and AMSYS 5812-0001-D-B (measurement range: ±10.34 mbar). Due to the narrow spaces in the model, the pressure sensors were placed outside of the model and connected to the bores in the wooden deck by brass connectors (inner diameter 1 mm) and plastic pressure tubes (inner diameter 1.5 mm), that are freed from burrs and debris. The length of plastic pressure tubes was about 0.9 m. The calibration of the pressure measurement system included two phases: static and dynamic. The static calibration was performed to establish the pressure-voltage relation for each pressure sensor, while dynamic calibration was performed to correct the dynamic effects of tubes [53]. Surface pressures were acquired with a sampling frequency of 1000 Hz and were scanned in a sample-and-hold modus, enabling the simultaneous sampling of signals. For each measurement, 28 samples of 4.7 s length were collected, where each sample corresponds to a 10 min long sample in full scale. The maximum uncertainty of surface pressures based on five repeated measurements was estimated to be ~2.5%. More details about the pressure measurement equipment that consisted of pressure sensors, a tubing system, amplifiers and analog/digital (A/D) converters can be found in [54]. The velocities above the roof were mainly measured at three different heights (z/D = 0.075, 0.3 and 0.45) above the points marked in red in Figure 4. In addition, above points 36 of the flat roof and 58 of the deck roof, at the centre of the roof, ten heights are considered with the spacing of z/D = 0.075. The hot-wire anemometer consisted of two cross wires allowing to measure both stream-wise and vertical velocity components. All velocity data, including hot-wire anemometer and Prandtl tube, are sampled with the frequency of 2000 Hz. The hot-wire anemometer was calibrated in laminar flow conditions in a calibration tunnel by exposing a probe to the set of known velocities, while corresponding voltages were recorded. The adopted fitting curve is a fourth order polynomial curve with coefficients calculated by fitting the data in the least-squares sense. Uncertainties related to the velocity measurements were calculated following the procedure presented by [55] and [56]. The total uncertainty of the velocity consisted of calibration, linearisation, positioning of the probe, digitalisation and uncertainty due to variation in the experimental conditions (such as temperature or ambient pressure). The maximum uncertainty of the time-averaged stream-wise velocity was 5.6%. The velocities above the roof were mainly measured at three different heights (z/D = 0.075, 0.3 and 0.45) above the points marked in red in Figure 4. In addition, above points 36 of the flat roof and 58 of the deck roof, at the centre of the roof, ten heights are considered with the spacing of z/D = 0.075. The hot-wire anemometer consisted of two cross wires allowing to measure both stream-wise and vertical velocity components. All velocity data, including hot-wire anemometer and Prandtl tube, are sampled with the frequency of 2000 Hz. The hot-wire anemometer was calibrated in laminar flow conditions in a calibration tunnel by exposing a probe to the set of known velocities, while corresponding voltages were recorded. The adopted fitting curve is a fourth order polynomial curve with coefficients calculated by fitting the data in the least-squares sense. Uncertainties related to the velocity measurements were calculated following the procedure presented by [55,56]. The total uncertainty of the velocity consisted of calibration, linearisation, positioning of the probe, digitalisation and uncertainty due to variation in the experimental conditions (such as temperature or ambient pressure). The maximum uncertainty of the time-averaged stream-wise velocity was 5.6%. The maximum uncertainties of the stream-wise and vertical turbulence intensities were estimated to be 9.6% and 9.4%, respectively. These uncertainty estimates correspond to the 95% confidence interval. The manual positioning of the hot-wire anemometer was detected as one of the main contributions to uncertainty, as established based on three repeated measurements at ten different heights above the middle point of the flat roof.
The flow in the roof region is expectably inherently 3D. Some degree of cross-flow is expected in those regions of the flow where vorticity occurs, i.e., closer to the surface. Eventual three-dimensionality in the flow is estimated looking at the performance of the signal measurements and the level of turbulence intensity. In regions where I U > 30%, the flow is likely to be affected by a reversed direction or strong cross-component. It is worth noting that even using 3D probes, reversed flow conditions cannot be detected either. Furthermore, those regions of highly skewed flow are not suitable for urban wind energy. Therefore, the degree of precision given by hot-wire anemometry is considered sufficient for the scope of the study: to understand the evolution of vortex structures and compare them to the surface pressure distribution for the sake of wind energy applications.
No blockage corrections of the measured results were considered, as even in the worst case, the normal-to-wind areas of the testing models were smaller than 1.8% of the wind tunnel cross-section.

Wind Velocity
In this subsection, the velocity measurements above the flat-roof building are discussed for the four wind angles of attack, 0 where σ U and σ W are the standard deviations of stream-wise and vertical wind velocity components and U is the mean stream-wise wind speed. Figure Figure 5a shows the velocity profiles above points 18 and 22 for the wind angle of 0 • . A strong separation develops at the upstream edge, as velocity vectors above the height of z/D = 0.3 point upwards. In the vicinity of the roof, a high stream-wise turbulence intensity zone can be detected, reaching up to a value of 55%. In regions of turbulence intensities >30%, hot-wire measurements are expected to be affected by some inaccuracy and, therefore, results cannot be treated as reliable [55]. Nevertheless, these high values of turbulence intensity confirm the existence of separated flow. The separation bubble from the windward edge reattaches in the vicinity of points 50 and 54, as velocity vectors point downwards.
Appl. Sci. 2020, 10, x FOR PEER REVIEW 8 of 22  At 15 • , the velocity profiles above the upstream points 18 and 22 are similar to those found at 0 • . However, profiles above points 38 and 50 show some differences. The velocity vectors above point 38 are no longer parallel to the roof, pointing slightly upwards, suggesting a downwind shift of the average moving vertex core. Besides, vectors above point 50 at heights z/D ≥ 0.3 are nearly parallel to the roof, seemingly out of the separation region, as the flow behaves analogously to points upstream of the separation bubble.
As the wind angle increases to 30 • , a small separation cone generates along the upstream edge of the building, starting at its upstream corner. The down-pointing velocity vectors at the lower measuring points suggest that both points 18 and 50 are located at the tail of the separation cone where the flow tends to attach to the roof. The profiles above the line x/D = 0.8 indicate a slightly smaller separation compared to the 15 • wind angle. This is followed by the reduction in turbulence intensities of the flow above points 38 and 54.
At 45 • , the wind flow is highly turbulent close to the roof above point 50 and less turbulent above point 18. This suggests that a large separation cone occurs at the upstream edge and increases in size along the length of the building side. In this particular case, the lower measurement position above point 50 lies entirely in the separation zone as I U is larger than 50%. In addition, another separation cone formed along the other upstream side of the building is affecting the flow above points 22 and, to a lesser extent, 38. In fact, the lower measurement position above point 38 is pointing slightly downwards, suggesting that it is located at the tail of the separation cone. However, the flow at higher positions above points 38 and 54 seems to lie out of the influence of the separation cone.
One possible way to verify inferences related to the flow patterns is the use of flow visualisation techniques, such as smoke injection. Another way would be to perform and analyse CFD simulations that are validated with the presented experimental results. Thus, future work will focus on justifying and exploring in-depth recognised flow patterns. Figure 5 also shows that the pair of points 18-22 and 50-54 at 0 • and 22-50 at 45 • have a similar configuration, confirming that the flow is symmetric at those angles.

Surface Pressure
The flow pattern above the roof can also be analysed based on surface pressure. In addition, surface pressure can provide more reliable validation as it usually has lower uncertainty levels compared to velocity measurements. Figure 6 shows the contours of the mean surface pressure coefficient of the flat-roof building, at the four wind angles. The pressure coefficient (C p ) is defined as where p ∞ , ρ and U ref are the free-stream pressure, the air density and the reference velocity, respectively. Figure 6a shows a significant reduction in the surface pressure at 0 • close to the upstream edge due to the flow separation, followed by a subsequent increase downstream. A similar distribution of the surface pressure is found at 15 • with a slightly higher reduction in the pressure close to the upstream corner. At higher wind angles of 30 • and 45 • , the contour maps are similar to each other, with a high reduction in surface pressure located at a limited area along both upstream edges. This pattern of pressure contours indicates the existence of two intense conical structures, as observed from the velocity measurements presented in Figure 5. As expected, for the 45 • wind angle, the surface pressure distribution is symmetrical around the diagonal line of the roof. The non-uniform arrangement of the measuring taps might be responsible for the slightly asymmetric pattern in Figure 6d.
Appl. Sci. 2020, 10, x FOR PEER REVIEW 10 of 22 arrangement of the measuring taps might be responsible for the slightly asymmetric pattern in Figure  6d. To provide an improved understanding of the effect of the wind direction on the surface pressure, Figure 7 shows the mean pressure coefficient and its standard deviation along two alignments on the roof, at x/D = 0.35 and x/D = 0.8. For all wind angles in Figure 7a, a characteristic upstream hump shape is observed, which is typical for a flow with a separated region followed by a reattachment [57]. The hump shape is related to the high negative pressure values in the separation region. The largest suction lies directly below the average moving vortex core [23]. The length of the mean recirculation region is related to the peak location of the standard deviation value, considering that the peak occurs just upstream of the mean reattachment position [57]. For wind angles of 0° and 15°, the hump shape is wider than in the case of 30° and 45°. This is the consequence of the large separation bubble generating at the upstream edge, which entails a long recirculation region. Comparing Figure 7a,b, the growth in the size of the conical vortices along the side of the building can be noticed for the 30° and 45° wind angles. To provide an improved understanding of the effect of the wind direction on the surface pressure, Figure 7 shows the mean pressure coefficient and its standard deviation along two alignments on the roof, at x/D = 0.35 and x/D = 0.8. For all wind angles in Figure 7a, a characteristic upstream hump shape is observed, which is typical for a flow with a separated region followed by a reattachment [57]. The hump shape is related to the high negative pressure values in the separation region. The largest suction lies directly below the average moving vortex core [23]. The length of the mean recirculation region is related to the peak location of the standard deviation value, considering that the peak occurs just upstream of the mean reattachment position [57]. For wind angles of 0 • and 15 • , the hump shape is wider than in the case of 30 • and 45 • . This is the consequence of the large separation bubble generating at the upstream edge, which entails a long recirculation region. Comparing Figure 7a Figure 8 gives further insight on the turbulence intensity in the stream-wise (IU) and vertical (IW) directions above all measuring points for the four wind angles. Along with local wind profiles, the inflow turbulence intensity at the same height is also shown as a reference (taken from Figure 3), using a black-filled circle mark.

Turbulence Intensity and Flow Acceleration
At 0°, the stream-wise turbulence intensity for all measurement points in the vicinity of the roof is >30%, indicating the presence of a large separation bubble. This is followed by turbulence intensities of ~20-25% in the vertical direction. At z/D = 0.3, the turbulence intensity in the streamwise direction is still higher than the reference profile value, within the limits 15-21%. This is not the case for upstream points (18, 20 and 22), where values are comparable to the reference ones. Turbulence intensities in the vertical direction are comparable to the reference profile within the range 10-13%. Measurements above the height of z/D = 0.3 are closer to the free-stream values, suggesting that the height of the separation region is slightly above 0.3D.
Similarly, to the previous case, turbulence intensity profiles at 15° show high turbulence close to the roof. Values at z/D = 0.075 are mostly over 30% for IU and over 20% for IW, followed by a reduction in turbulence with increasing height. At z/D = 0.3, turbulence intensity lies in the ranges of 12-22% and ~10-15% for the stream-wise and vertical directions, respectively. Moreover, Figure 8 reveals that values above the alignment x/D = 0.2 and x/D = 0.5 show a slight shift in turbulence intensities compared to 0° for both the stream-wise and vertical directions. In contrast, an increase in turbulence intensity is noticeable above the alignment x/D = 0.8. This suggests that the separation bubble moves to the right while maintaining the height of ~0.3D.
Considering the 30° wind angle case, most of the points in the vicinity of the roof, at z/D = 0.075, show IU > 30% and IW > 15%. Exceptions are points above the main diagonal over the roof (36 and 54) with lower values, ~23% for IU and ~13% for IW. It leads to the conclusion that these points are near the border of the cone vortex formed at the upstream edge. Measurements from height z/D = 0.3 are analogous to the reference profile values.
For a wind angle of 45°, higher values of turbulence intensities, reaching up to 50% in streamwise and 25% in the vertical direction, are observed at the lowest measured heights, in regions affected by the two separation cones. Turbulence levels are increasing as the cone is increasing in size (e.g., above points 20 and 22). Above the main diagonal on the roof (points 18, 36 and 54), measurements at all heights are characterised by relatively low turbulence intensities. These turbulence levels are comparable or even lower, as in the case of vertical turbulence, than the reference inflow. Namely, the wind separates from the two upstream sides of the building, forming two cone vortices, while the flow at the centre of the roof might be fully attached. Again, as at 30°, turbulence intensities from height z/D = 0.3 for all points match free-stream values.  Figure 8 gives further insight on the turbulence intensity in the stream-wise (I U ) and vertical (I W ) directions above all measuring points for the four wind angles. Along with local wind profiles, the inflow turbulence intensity at the same height is also shown as a reference (taken from Figure 3), using a black-filled circle mark.

Turbulence Intensity and Flow Acceleration
At 0 • , the stream-wise turbulence intensity for all measurement points in the vicinity of the roof is >30%, indicating the presence of a large separation bubble. This is followed by turbulence intensities of 20-25% in the vertical direction. At z/D = 0.3, the turbulence intensity in the stream-wise direction is still higher than the reference profile value, within the limits 15-21%. This is not the case for upstream points (18, 20 and 22), where values are comparable to the reference ones. Turbulence intensities in the vertical direction are comparable to the reference profile within the range 10-13%. Measurements above the height of z/D = 0.3 are closer to the free-stream values, suggesting that the height of the separation region is slightly above 0.3D.
Similarly, to the previous case, turbulence intensity profiles at 15 • show high turbulence close to the roof. Values at z/D = 0.075 are mostly over 30% for I U and over 20% for I W , followed by a reduction in turbulence with increasing height. At z/D = 0.3, turbulence intensity lies in the ranges of 12-22% and~10-15% for the stream-wise and vertical directions, respectively. Moreover, Figure 8 reveals that values above the alignment x/D = 0.2 and x/D = 0.5 show a slight shift in turbulence intensities compared to 0 • for both the stream-wise and vertical directions. In contrast, an increase in turbulence intensity is noticeable above the alignment x/D = 0.8. This suggests that the separation bubble moves to the right while maintaining the height of~0.3D.
Considering the 30 • wind angle case, most of the points in the vicinity of the roof, at z/D = 0.075, show I U > 30% and I W > 15%. Exceptions are points above the main diagonal over the roof (36 and 54) with lower values,~23% for I U and~13% for I W . It leads to the conclusion that these points are near the border of the cone vortex formed at the upstream edge. Measurements from height z/D = 0.3 are analogous to the reference profile values.
For a wind angle of 45 • , higher values of turbulence intensities, reaching up to 50% in stream-wise and 25% in the vertical direction, are observed at the lowest measured heights, in regions affected by the two separation cones. Turbulence levels are increasing as the cone is increasing in size (e.g., above points 20 and 22). Above the main diagonal on the roof (points 18, 36 and 54), measurements at all heights are characterised by relatively low turbulence intensities. These turbulence levels are comparable or even lower, as in the case of vertical turbulence, than the reference inflow. Namely, the wind separates from the two upstream sides of the building, forming two cone vortices, while the flow at the centre of the roof might be fully attached. Again, as at 30 • , turbulence intensities from height z/D = 0.3 for all points match free-stream values. Thus, it can be concluded that for wind angles of 30° and 45°, turbulence assumes characteristics similar to the free-stream values above the height z/D = 0.3 at all positions on the roof. Arguably, the wind direction of 45° provides a suitable flow pattern for wind energy, as the turbulence intensities above the main diagonal of the roof are low, even at positions closer to the roof.
To compare the speed-up effect above the roof, Figure 8 additionally illustrates the normalised velocity profiles above the measuring points for the different wind angles. These profiles are normalised using the reference velocity (Uref). A similar normalisation is also shown in Figure 5.
At 0°, the zones of high turbulence intensity close to the roof are characterised by low values of normalised wind speed. Above, at heights z/D = 0.3 and z/D = 0.45, an increase in wind speed in the To compare the speed-up effect above the roof, Figure 8 additionally illustrates the normalised velocity profiles above the measuring points for the different wind angles. These profiles are normalised using the reference velocity (U ref ). A similar normalisation is also shown in Figure 5.
At 0 • , the zones of high turbulence intensity close to the roof are characterised by low values of normalised wind speed. Above, at heights z/D = 0.3 and z/D = 0.45, an increase in wind speed in the range of 20-30% can be observed. The highest value of~30% occurs above point 36 at z/D = 0.3, suggesting this is the upper limit of the separation zone.
Similar to the previous case, at 15 • , a decrease in wind speed occurs due to the separation in the vicinity of the roof. In this case, an increase in the normalised velocity of~15-20% is identified above several measuring points from the height z/D = 0.3.
At 30 • , normalised velocities at the lowest measurement height are mostly <1. At height z/D = 0.3, they are increased by the range of 10-20%, reaching 20% in the zone above the cone vortex development.
For a wind angle of 45 • , even at the vicinity of the roof, all points at the main diagonal are experiencing an increase in wind speed of more than 15%. Yet, the maximum velocity increase of 20% occurs above the downstream point 54. These positions close to the roof are good candidates for wind harvesting due to their location and the fact that turbulence intensity levels are rather low for this wind direction. When all the cases are considered, the most favourable result regarding flow acceleration is detected above point 36 at a height of z/D = 0.3 at 0 • . Still, this region seems to be affected by slightly higher turbulence intensity levels, implying that for the adequate positioning of the wind turbine, an even higher location, just by a bit, should be considered.

Flow Pattern Comparison with Other Flat-Roof Buildings
In order to compare the flow above high-rise buildings with available results for low-rise buildings, a comparison with the databases AIJ and CEDVAL is proposed. The CEDVAL database reports flow measurements around a cube (H/D = 1:1) obtained through laser Doppler anemometry (LDA two-component system) for various wind directions. At 0 • , a large separation over the top of the roof is detected. However, the actual size of the separation length is not available due to the lack of measurement data near the roof, as the closest measurement positions are located at z/D = 0.2. The maximum velocity increase of 21% occurs above the central point of the cube at z/D = 0.6. This is followed by slightly higher turbulence intensity levels than the reference ones [39].
The CEDVAL database also considers an incident angle of 45 • . Close to the roof, i.e., at a height of z/D = 0.2, the normalised velocity increase is in the range 12-15%. The maximum value is reached at a similar location as in the present study, i.e., close to the downstream point 54. As also found in the present study, at 45 • , turbulence intensities I U and I W are comparable to or even lower than the reference inflow values.
A significantly smaller separation bubble, as compared to the current case, is instead reported in the AIJ dataset [37], which considers a square isolated flat-roof building, at 0 • , with a ratio of H/D = 2:1. Based on the reversed flow detected, the reattachment length can be estimated at~1/4D, which is lower than in the present case. This can be attributed to the higher turbulence intensity in the inflow. In addition, a slightly lower flow acceleration~24% is found above the location at 1/4D at height z/D = 0.375, compared to the current high-rise case, where similar turbulence intensities to the reference one are recorded.
Comparing the surface pressure patterns on the flat roof from the low-rise buildings [37,39] and the current high-rise building, it can be concluded that the surface pressure contours are consistent for a given wind direction. These patterns reveal the existence of analogous flow structures, i.e., separation bubble at α = 0 • and conical vortices for the oblique wind direction, as observed in Figure 6. The Architectural Institute of Japan published two databases on pressure measurement on low-rise and high-rise buildings. Those experimental campaigns were mainly focused on the sides of the buildings and only measurements on the low-rise buildings included measurement positions at the flat roof. Figure 9 compares the mean pressure coefficient over the centre line of two cubes having respectively H/D = 1:2 and H/D = 1:1, taken from the AIJ database, for the two wind directions 0 • and 45 • . As mentioned above, the size of the hump shape of the pressure coefficient curve, which is related to the high negative pressure in the separation region, can be used as an estimate of the size of the separation bubble. Figure 9 shows a possible trend for the hump shape, which is increasing alongside the height of the building. A smaller increase in size seems to occur for 45 • , yet more windward measurement points would be required for a more detailed analysis. The increase in the size of the separation bubble is combined with an increase in the acceleration effect above the roof. The velocity measurements over buildings with different aspect ratios, namely H/D = 1:1 from the CEDVAL database; H/D = 2:1 from the AIJ database; and H/D = 3:1 from the current study, reveal that the wind acceleration increases together with the building height. Further, the CFD study by Abohela et al. [34] shows the same conclusion regarding the acceleration effect over the barrel-vaulted roof of a high-rise building with three different heights.
A comprehensive study considering the same inflow conditions and measurement locations above the flat roof would provide more general conclusions on the effect of building height to identify favourable locations suitable for wind energy exploitation.

Wind Velocity
The following subsection presents the results of the velocity measurements above the deck-roof building, obtained for wind angles of 0° and 45°. The mean velocity and turbulence intensity have been measured above two points on the roof top, i.e., points 50 and 58, marked in red in Figure 4b. The velocity has been measured at ten heights above the roof centre point (point 58 in Figure 4b) and five heights above point 50 at 0°, and four heights above both points at 45°. The normalised velocity profiles, based on the stream-wise and vertical velocity components above these points, are shown in Figure 10, together with the longitudinal (IU) and vertical (IW) turbulence intensities calculated using Equation (1). Figure 10 also shows the percentage increase in the mean value of the streamwise velocity component as in the previous section. In order to compare the behaviour of the deckroof to that of the flat-roof building, Figure 10 also shows the wind profiles above x/D = 0.5 (points 20 and 36) for the same wind angles.
At 0°, results in Figure 10a show that velocity vectors above point 50 point upwards, which suggests that the flow separates at the windward edge of the deck-roof building. However, the velocity vectors above point 58 are almost parallel to the roof, indicating that the size of the separation bubble is much smaller than that above the flat-roof building presented in Figure 10b. It is interesting The increase in the size of the separation bubble is combined with an increase in the acceleration effect above the roof. The velocity measurements over buildings with different aspect ratios, namely H/D = 1:1 from the CEDVAL database; H/D = 2:1 from the AIJ database; and H/D = 3:1 from the current study, reveal that the wind acceleration increases together with the building height. Further, the CFD study by Abohela et al. [34] shows the same conclusion regarding the acceleration effect over the barrel-vaulted roof of a high-rise building with three different heights.
A comprehensive study considering the same inflow conditions and measurement locations above the flat roof would provide more general conclusions on the effect of building height to identify favourable locations suitable for wind energy exploitation.

Wind Velocity
The following subsection presents the results of the velocity measurements above the deck-roof building, obtained for wind angles of 0 • and 45 • . The mean velocity and turbulence intensity have been measured above two points on the roof top, i.e., points 50 and 58, marked in red in Figure 4b. The velocity has been measured at ten heights above the roof centre point (point 58 in Figure 4b) and five heights above point 50 at 0 • , and four heights above both points at 45 • . The normalised velocity profiles, based on the stream-wise and vertical velocity components above these points, are shown in Figure 10, together with the longitudinal (I U ) and vertical (I W ) turbulence intensities calculated using Equation (1). Figure 10 also shows the percentage increase in the mean value of the stream-wise velocity component as in the previous section. In order to compare the behaviour of the deck-roof to that of the flat-roof building, Figure 10 also shows the wind profiles above x/D = 0.5 (points 20 and 36) for the same wind angles.
the shape of the roof, thus preventing the development of the large separation bubble at the upstream edge. This fact is supported by the flow acceleration noticeable in the vicinity of the roof.
At 45°, a similar flow pattern can be observed above the central points (point 58 in Figure 10c and point 36 in Figure 10d) for both roof shapes. The lack of a vertical velocity component at the lowest measurement positions as well as low turbulence levels indicate that the flow is attached to the roof. The difference is noticeable in the flow above point 50 for the deck-roof building case and point 20 for the flat roof case. While point 20 is in the separation zone at flat-height z/D = 0.075, point 50 seems to be above the separation zone due to the low turbulence intensities and the increase in the mean value of the wind speed.  At 0 • , results in Figure 10a show that velocity vectors above point 50 point upwards, which suggests that the flow separates at the windward edge of the deck-roof building. However, the velocity vectors above point 58 are almost parallel to the roof, indicating that the size of the separation bubble is much smaller than that above the flat-roof building presented in Figure 10b. It is interesting to note that turbulence intensities are lower close to the top of the deck-roof building, in contrast to high-turbulence intensity zones detected in the same zone in Figure 10b. This may be explained by the effect of the shape of the deck roof. As the incoming air overcomes the obstacle provided by the high-rise building, lifting over the roof of the building, the streamlines of the velocity vectors follow the shape of the roof, thus preventing the development of the large separation bubble at the upstream edge. This fact is supported by the flow acceleration noticeable in the vicinity of the roof.
At 45 • , a similar flow pattern can be observed above the central points (point 58 in Figure 10c and point 36 in Figure 10d) for both roof shapes. The lack of a vertical velocity component at the lowest measurement positions as well as low turbulence levels indicate that the flow is attached to the roof. The difference is noticeable in the flow above point 50 for the deck-roof building case and point 20 for the flat roof case. While point 20 is in the separation zone at flat-height z/D = 0.075, point 50 seems to be above the separation zone due to the low turbulence intensities and the increase in the mean value of the wind speed.

Surface Pressure
The plot of the distribution of the mean surface pressure coefficient on the deck roof is shown in Figure 11 for wind angles of 0 • , 15 • , 30 • and 45 • .
At 0 • , at the central flat portion of the deck roof, a reduction in the surface pressure close to the leading edge is limited to a more confined area compared to the flat roof case ( Figure 6). As the approaching wind angle increases, the separated area linked to the leading edge decreases. At 30 • and 45 • , the shape of the pressure contours still shows slight similarities to the flat roof case (Figure 6), as the contours become denser close to the windward corner. Nevertheless, the number of measurement points in this confined area is limited and it is not clear from the pressure plots if small conical vortices exist. However, the results presented in Figure 10c confirm the existence. Thus, for all wind angles, the shape of the deck roof suppresses considerably the separation area. In addition, at all plots, it can be observed that downstream of this confined high-suction area the surface pressure distribution is rather uniform. This uniform pressure indicates that the flow is generally attached to the roof. Looking at the distribution over the decks of the roof, a similar pattern can be observed. The windward sides have a high-suction zone close to the upwind edges, while on the leeward sides the pressure distribution is mostly uniform.

Turbulence Intensity and Flow Acceleration
The measurements of turbulence intensities in the stream-wise (I U ) and vertical (I W ) directions over the deck roof are shown in Figure 12 for two wind angles of 0 • and 45 • , at four heights above the measurement points on the alignment x/D = 0.5, together with the free-stream values marked with black circles. The figure also includes the normalised velocity profiles of the mean value of the stream-wise component in order to examine the speed-up effect above the roof.
Considering the results for both wind angles, the turbulence intensity profiles present favourable conditions for the installation of wind turbines over the deck roof. Namely, measurements of the stream-wise turbulence intensity at the majority of heights are comparable to the free-stream ones, not exceeding 13%. The only exception is the value at height z/D = 0.075 above point 58 at 0 • , which reaches 19%, indicating that it is affected by upstream separation. Vertical turbulence intensity values are even lower than the reference ones for all cases, particularly close to the roof, as they are in the range of 5-10%.

Turbulence Intensity and Flow Acceleration
The measurements of turbulence intensities in the stream-wise (IU) and vertical (IW) directions over the deck roof are shown in Figure 12 for two wind angles of 0° and 45°, at four heights above the measurement points on the alignment x/D = 0.5, together with the free-stream values marked with black circles. The figure also includes the normalised velocity profiles of the mean value of the streamwise component in order to examine the speed-up effect above the roof. Considering the results for both wind angles, the turbulence intensity profiles present favourable conditions for the installation of wind turbines over the deck roof. Namely, measurements of the stream-wise turbulence intensity at the majority of heights are comparable to the free-stream ones, not exceeding 13%. The only exception is the value at height z/D = 0.075 above point 58 at 0°, which reaches 19%, indicating that it is affected by upstream separation. Vertical turbulence intensity values are even lower than the reference ones for all cases, particularly close to the roof, as they are in the range of 5-10%. In the vicinity of the roof, for both wind angles, the wind speed accelerates with a rate of more than 20%. Moreover, a larger area seems to be affected at 45 • with this favourable flow condition. As the height increases, the speed-up effect gradually decreases.
Presented results for the deck-roof building show that the wind direction has a small effect on both the mean wind velocity and turbulence intensity above the roof. In addition, although the deck roof generates a velocity amplification of similar magnitude compared to the flat roof at 0 • , this occurs at notably lower positions (closer to the roof surface). Moreover, the turbulence intensity in that zone is equal or even lower compared to the free-stream at most measurement points. Similar flow conditions are observed over the deck roof at 45 • compared to the flat roof case.

Conclusions
A series of experimental tests has been carried out in a boundary layer wind tunnel to measure both the velocity field and the surface pressure above the roof of a high-rise building. Besides providing validation test cases of future numerical investigations, the study aims at improving the understanding of the flow pattern of the roof region for wind energy harvesting. The effects of wind direction and roof shape have been investigated in terms of flow pattern, turbulence intensity and accelerated wind velocity.
For all wind angles, the flow separates at the leading edge of the flat roof. Nevertheless, different separation patterns take place, depending on the angle of the wind, which are visible in both the surface pressure and wind speed profiles. At 0 • , the separation bubble extends to about 2/3 of the building width, with a height of about 0.3D. In contrast to the separation bubble, cone vortices dominate the flow pattern at 30 • and 45 • . The cone vortices grow in size as they evolve along the side of the building. The largest size for cone vortices is found at 30 • .
As a broad region is affected by the separation, the 0 • configuration is the most turbulent one. In general, as the wind angle increases, the region of high turbulence intensity reduces gradually, and 45 • is found to be the most preferable wind direction, minimising turbulence. As regards the flow acceleration, the 0 • configuration provides the highest increase in wind speed, up to U~1.3U ref at z~0.3H at the centre of the roof. However, this position is still influenced by relatively high turbulence intensities. On the other hand, the 45 • wind direction provides a substantial amplification of wind speed up to 20%, at significantly lower positions close to the roof with a lower turbulence intensity. This vicinity to the roof is the determining factor for the installation of small wind turbines.
A similar behaviour is observed in comparison with the flow over low-rise buildings. Namely, similar flow patterns are recognised for different flow angles and besides, similar areas are detected as the most favourable locations for the possible installation of wind turbines. In addition, an increase in building height seems to be followed by an increase in wind acceleration. Yet, more detailed studies are needed to provide more definitive conclusions on the relation between building features and wind resource.
Decking the roof significantly improves the flow for wind energy purposes. Among the most important observations is the insensitivity of the flow above the deck roof to changes in wind direction, unlike in the flat roof case. Results also show a far smaller separation zone for all wind directions, which is beneficial as wind turbines can be installed closer to the surface. Turbulence intensities are equal to or lower than the reference value of the free-stream wind profile for all configurations, highlighting the advantage of this shape. Besides the favourable effect on turbulence intensities at lower heights, the deck-roof shape performs well in terms of enhanced wind acceleration, reaching a maximum increase in wind speed of around 20%.