Diurnal, weekly, seasonal and spatial variabilities in carbon dioxide flux in different urban landscapes in Sakai, Japan

To evaluate CO2 emissions in urban areas and their temporal and spatial variabilities, continuous measurements of CO2 fluxes were conducted using the eddy covariance method at three locations in Sakai, Osaka, Japan. Based on the flux footprint at the measurement sites, CO2 fluxes from the three sites were partitioned into five datasets representing a dense urban center, a moderately urban area, a suburb, an urban park, and a rural area. Distinct biological uptake of CO2 was observed in the suburb, urban park, and rural areas in the daytime, whereas high emissions were observed at dense and 10 moderate urban areas in daytime. Weekday CO2 emissions in the dense urban center and suburban area were approximately 50% greater than during weekend and holidays, but the other landscapes did not exhibit a clear weekly cycle. Seasonal variations in the urban park, rural area, and suburban area were influenced by vegetation activities, exhibiting the lowest daily emissions or even uptakes during summer months. In contrast, the dense and moderately urban areas exhibited higher emissions in winter and summer months, when emissions significantly increased as air temperature increased in summer and 15 air temperature decreased in winter. Irrespective of the landcover type, all urban landscapes measured in this study acted as net annual CO2 sources, with emissions ranging from 0.5 to 4.9 kg C m yr. The magnitude of the annual CO2 emissions was negatively correlated with green fraction; areas with a smaller green fraction had higher annual CO2 emissions. Upscaled flux estimates based on the green fraction indicated that the emissions for the entire city were 3.3 kg C m yr, which is equivalent to 0.5 Tg C yr or 1.8 Mt CO2 yr based on the area of the city (149.81 km). A network of eddy 20 covariance measurements is a powerful tool to evaluate CO2 emissions from urban areas.


Introduction
Cities emit a considerable amount of carbon dioxide (CO2) that is associated with human activities into the atmosphere (Canadell et al., 2007). Urban areas account for only a samll percentile of the earth's land surface but emit 30-50% of total anthropogenic CO2 (Mills, 2007;Stterthwaite, 2008). Consequently, CO2 emissions from cities are highly heterogeneous at 25 spatial and temporal scales.
Global CO2 emissions have often been estimated using inventories (Oda and Maksyutov, 2011). The major challenge for estimating global CO2 emissions is to understand the spatio-temporal dynamics of CO2 emissions in various cities. Because Atmos. Chem. Phys. Discuss., doi:10.5194/acp-2016-334, 2016 Manuscript under review for journal Atmos. Chem. Phys. Published: 21 June 2016 c Author(s) 2016. CC-BY 3.0 License. emissions data are used in top-down estimates of global CO2 budget (Peters et al., 2007;Schmel et al., 2001), a better estimate of CO2 emissions from cities will improve our understanding of global carbon cycle, including terrestrial and ocean fluxes.
Multi-site eddy covariance towers were used to synthesize the data and showed that vegetation fraction was the major factor controlling spatial variability in annual CO2 emissions Velasco and Roth, 2010;Ward et al., 2015) because vegetation fraction can be correlated with anthropogenic activities. Upscaling can provide a high-resolution map of 15 direct CO2 emissions from cities. Previous studies have examined the relationship between annual CO2 emissions and vegetation fraction at a global scale Velasco and Roth, 2010;Ward et al., 2015). It is unclear whether upscaling CO2 emissions is possible within a city, because multi-site eddy covariance measurements within a city are often unavailable. 20 In this study, we present diurnal, weekly, seasonal, and spatial variabilities in CO2 flux continuously measured at three different locations within 5 km of each other in Sakai, Osaka, Japan. Considering flux footprint, the data represent five urban landscapes, including a dense urban center, a moderately urban area, a suburb, a rural area, and an urban park. Regardless of the landscape type, all landscapes emitted considerable CO2 annually with different temporal metabolisms, providing a useful overview of anthropogenic CO2 emissions. 25

Study sites
Sakai is the second largest city in Osaka Prefecture, located in western Japan. The population was approximately 842,000 in 2015. Because the city is located on the eastern shore of Osaka Bay, sea breeze circulation is evident throughout the year except when seasonal winds are not strong. The area is on a uniformly flat plane; the north-south and the east-west slopes are 30 0.0030° and 0.0024°, respectively. The climate of Sakai is temperate; the mean annual air temperature is 15.9°C, the maximum monthly mean air temperature is 28.0°C in August, the minimum monthly mean air temperature is 5.2°C in Atmos. Chem. Phys. Discuss., doi:10.5194/acp-2016-334, 2016 Manuscript under review for journal Atmos. Chem. Phys. Published: 21 June 2016 c Author(s) 2016. CC-BY 3.0 License. January, and the mean annual precipitation is 1187 mm yr -1 between 1981 and 2010 according to the apanese Meteorological Agency.
The Sakai city center (SAC) site ( Fig. 1; Table 1) is located on a tower at the top of a city office building (34°34′25′′N, 135°28′80′′E). The population density around the city center is approximately 12150 km -2 , based on the Japanese 5 Government Statistics. The area is a densely built-up urban area with a mean building height of 10.7 ± 3.1 m. Many arterial roads and two highways with heavy traffic are present within the flux footprint. Because industrial and commercial areas are located in the western and northern parts of the city, those areas are expected to show higher rates of human activity than locations where residential areas are dominant.

10
The Oizumi Ryokuchi urban park (IZM) site ( Fig. 1; Table 1) is located at the northern end of the city (34°33′48′′N, 135°32′1′′E), and was established in 1972. The measurements were conducted at a tower located at the east edge of the park.
Because of the consistent presence of a sea breeze, the tower is mostly located downwind direction of the park during the daytime. The landcover of the park consists of 51% trees, 15% grassland, and 34% other. Measurements using a plant canopy analyzer (LAI-2000, LI-COR, Lincoln, Nebraska, USA) showed that the leaf area index of trees ranged from 3.2 to 15 5.7 m 2 m -2 with a mean of 4.3 m 2 m -2 in the summer months. The area surrounding the IZM is a mixed landscape of residential area and agricultural fields and is characterized as a rural area.
The Osaka Prefecture University (OPU) site ( Fig. 1; Table 1) is located at the west edge of Osaka Prefecture University (34°32′50′′N, 135°30′10′′E). Because the measurements were conducted at the roof top of a building, the flux footprint 20 represents only a small suburban area. The western part of the site contains a protected forest on an ancient tumulus, Mozu Kofungun. The area is characterized as a suburb, consisting of a university, a residential area, small streets, a graveyard, and trees.

Observations 25
We measured CO2 fluxes using the eddy covariance method at the three sites. For SAC, a sonic anemometer (SAT550, Sonic using an open-path infrared gas analyzer (LI-7500, LI-COR). Another eddy covariance system using a sonic anemometer (DA600,Sonic Corp.) and an open-path infrared gas analyzer (LI-7500, LI-COR) was installed on a different edge of the building in November 2015. This additional measurement system increased data acquisition, because we eliminated the data coming from the roof. Consequently, CO2 fluxes were calculated based on the different systems with relevant corrections. 10 We confirmed that there was no significant difference between open-path and closed-path systems through an intercomparison, but these flux measurements have higher uncertainties than those from the other sites.
Meteorological and environmental variables were measured at each site. Air temperature, relative humidity, and incoming solar radiation were measured at the three sites. Rainfall, atmospheric pressure, incoming longwave radiation, and ground 15 heat fluxes at the top of the building were measured at OPU. Leaf area index was manually measured approximately once a month using a plant canopy analyzer (LAI-2000, LI-COR) at ten forested sectors in IZM.

Data analysis
Turbulent fluxes were calculated using the eddy covariance method using the Flux Calculator program (Ueyama et al., 2012).
Before the half-hourly covariance of vertical wind velocity and scalar quantities was calculated, spike data were removed 20 from the raw data. No trend removal was applied. The artificial fluctuations of sonic air temperature associated with water vapor were corrected. Vertical wind velocity was coordinated as mean vertical wind velocity was zero using the double rotation method. The angle-of-attack errors for the Gill Instruments and R. M. Young anemometers were corrected based on Nakai and Shimoyama (2012) and Kochendorfer et al. (2012), respectively. The high-frequency loss for line averaging and sensor separation was corrected using theoretical transfer functions for the open-path systems (Massman, 2000) and 25 empirical transfer functions for the closed-path system (Moore, 1986). Air density fluctuations were corrected based on Webb et al (1980). Filtering nighttime data using friction velocity (u*) threshold was not applied in this study. This was because (1) no clear threshold was obtained in nighttime data, (2) data coverage at night was small due to the limited flux footprint, and (3) 30 sensible heat fluxes in summer months often showed positive values even at night except IZM. Our handling of nighttime data was the same in previous studies in urban areas (e.g., Liu et al., 2012), but a potential underestimate of nighttime fluxes may have occurred.
Flux data were selected for each landscape after a quality test and footprint analysis. First, we applied the quality test to remove half-hourly flux data that included noise based on a criterion (Appendix B.1 in Ueyama et al., 2012). A stationary test, integral turbulence test, and higher moment test were applied. The fluxes, measured when winds came from the tower directions, were also removed. For OPU, the fluxes, measured when winds came from the directions of the building, were 5 removed as well. For SAC, based on a footprint model (Kormann and Meixner, 2000), we rejected a data, when the source area contributing 80% of the flux footprint contained sea and mountains. Similarly, for IZM, we rejected flux data, when the source area contributing 50% of the flux footprint exceeded the boundary of the urban park.
Depending on wind directions, flux data at IZM and SAC were divided into two data series. For IZA, flux data from the west 10 represented the urban park, whereas data from other directions represented the rural area consisting of mixed residential and agricultural areas (Fig. 1). For SAC, flux data from the west represented the densely built-up urban center, whereas data from other directions represented the moderate urban to residential area (Fig. 1). Consequently, we formed five flux datasets from measurements at the three sites: a dense urban center (west SAC), a moderately urban area (east SAC), a suburb (OPU), an urban park (west IZM), and a rural area (east IZM). Data coverage was 11% in west SAC, 21% in east SAC, 31% in OPU, 15 16% in west IZM, and 13% in west IZM.
Partitioning CO2 fluxes into assimilatory and respiratory fluxes was conducted only for the west and east IZM and OPU datasets, because biological signals were present in CO2 fluxes. The flux partitioning was conducted using the Flux Analysis Tool program (Ueyama et al., 2012). First, the relationship between nighttime CO2 fluxes and air temperature was 20 established based on a model (Lloyd and Taylor, 1994). The relationship was determined daily with a 49-day moving window. Assimilatory fluxes were calculated as the difference between the estimated respiratory flux and the measured CO2 flux. Because the estimated respiratory fluxes consisted of biological fluxes and nighttime anthropogenic fluxes, it is important to note that the estimated assimilatory fluxes did not truly represent gross primary productivity, which is often used in ecosystem studies (e.g., Baldocchi, 2008). 25 Gaps in the five datasets were filled using the Flux Analysis Tool program. First, small data gaps for periods less than 2.0 h were filled by linear interpolation. Second, for the west and east IZM datasets, gaps were filled using a combination of lookup-table and non-linear regression methods (Ueyama et al., 2012), an approach well established for use in natural ecosystems (Ueyama et al., 2013).For data gaps from west and east SAC and OPU, mean diurnal variations were applied, in which a 30 mean diurnal pattern was created daily using a 51-day moving window. Two mean diurnal patterns were created, one for weekdays and once for weekends and holidays according to the weekly cycle.
The potential mitigation of CO2 uptake by vegetation (RM) was inferred for west and east IZM and OPU using the followings: where FA represents the partitioned assimilatory flux, FNUE represents CO2 emissions measured by the eddy covariance, and ε represents the conversion efficiency of gross assimilation into net assimilation (0.06; Baldocchi, 2008). The value of ε was determined based on a global synthesis of disturbed ecosystems, although there is considerable uncertainty in its use with urban vegetation. Furthermore, the estimated FA may be biased due to nighttime anthropogenic activities.

Upscaling using GIS data 10
The annual CO2 flux was upscaled according to the relationship between annual fluxes and green fraction. The green fraction was estimated using a green census data developed by government of Sakai City. The green census data were created using high-resolution aerial photographs from August 2001 consisting of polygons of approximately 5-m spatial resolution. Based on the high-resolution polygon data, the green fraction was evaluated at a 500-m spatial resolution. Because the green census data often classified water area as green area, we masked water area using landcover data based on a geographical 15 information system (Digital Map 5000 for the Kinki region in 2008 by the Geospatial Information Authority of Japan).

Diurnal variations
CO2 fluxes showed distinct diurnal variation (Fig. 2). At the IZM, in the urban park and rural area, distinct biological signals were observed, representing apparent daytime uptake. The magnitude of the daytime uptake was stronger in the urban park 20 than in the rural area. A daytime uptake was also observed at OPU in the summer months from April to August. In contrast to the apparent biological signals, diurnal variations at SAC showed greater CO2 emissions during the daytime than at night.
Emissions were greater in the dense urban center (west SAC) than in the moderately urban area (east SAC). Emissions from the urban areas tended to be high in the daytime, particularly in the morning rush hours. Such diurnal variations were similar to those for traffic counts measured by highway exits within the flux footprint (Fig. 2b). Note that the traffic counts at the 25 exits peaked in the evening, whereas those at the entries could peaked in the morning (data are not shown).

Seasonal variations
Different urban landscapes showed different seasonal variations in CO2 flux (Fig. 3) showed CO2 emissions throughout the seasons, but the emissions rate tended to be lower in the spring than in other months.
The SAC site showed high CO2 emissions throughout the seasons, and higher emissions were observed in the dense urban center than in the moderately urban area. The seasonal variations in SAC exhibited two distinct peaks during the summer and winter periods.

5
The seasonal variations in the daily CO2 flux were dependent on daily mean air temperature and exhibited different patterns in different landscapes (Fig. 4). For the urban site of SAC, CO2 emissions increased as temperatures decreased (0.46-0.27 g C m -2 d -1 °C -1 ; p < 0.1) when mean daily temperature less than 10°C. Higher CO2 emissions were also observed at higher temperatures at SAC. An increase in CO2 emissions at higher temperatures was also observed at OPU. Gas consumption by university buildings within a footprint of OPU was consistent with the two seasonal peaks revealing higher consumption in 10 the summer and winter months (Fig. A1). In the urban park and rural area, CO2 emissions decreased as temperatures increased above 20°C.
Assimilatory fluxes were greater in summer months than in winter months (Fig. 3). Surprisingly, the assimilatory fluxes in the urban park and OPU were comparable, probably due to contributions of trees around the university and from the tomb at 15 OPU. The assimilatory fluxes for the rural area were approximately half those for the urban park and OPU. The assimilatory fluxes increased as temperatures increased higher than 20°C at 0.15-0.38 g C m -2 d -1 °C -1 (p < 0.01).

Weekly variations
Among the five landscapes, distinct weekly cycles of CO2 emissions were only observed at the west SAC and OPU sites (Fig.   5). On average, CO2 emissions on weekdays were approximately 50% greater than emissions on weekends and holidays, 20 even though the weekday CO2 flux at the east SAC tended to be 10% higher than the fluxes on holidays. The greater emissions on weekdays were consistently observed throughout all seasons, and were consistent with the traffic counts from the highway exits, where traffic was approximately 23% higher on weekdays than on weekends and holidays (Fig. 2b).

Annual CO2 balance and its spatial variations
All urban landscapes measured in this study acted as net CO2 emissions sources on the annual timescale ( Fig. 6; Table 2). 25 The strength of the annual CO2 emissions was negatively correlated with the green fraction (R 2 = 0.96; p < 0.01); areas of a smaller green fraction had higher annual CO2 emissions. Based on the annual assimilatory fluxes and annual CO2 emissions measured using Equation 1, the potential mitigation of CO2 uptake by urban vegetation ranged from 5% to 9% (Table 2).
This pattern indicates that direct mitigation by vegetation is limited and that the green fraction explains the magnitude of anthropogenic activities rather than the direct contributions from vegetative uptake. The annual CO2 emissions estimated in 30 this study were lower than those examined using a global synthesis by Nordbo et al. (2012) (Fig. 6). Based on the significant relationship between the green fraction and the annual CO2 flux, annual CO2 fluxes were upscaled to the city scale (Fig. 7). Because the green fraction of Sakai was low in the north and high in the south (Fig. 7a), annual CO2 emissions were greater in the north than the south (Fig. 7b). The annual CO2 fluxes from the entire city were 3.3 kg C m -2 yr -1 , which corresponds 0.5 Tg C yr -1 or 1.8 Mt CO2 yr -1 based on the area of the city (149.81 km 2 ). The estimated emissions were lower than an inventory-based estimate from 2000 to 2012 (8.0 ± 0.6 Mt CO2 yr -1 ) published by the government. 5

Discussion
Annual CO2 emissions from Sakai City ranged in those measured in other studies, but tended to be at the lower end of the range (Fig. 6). For the same fraction of green area (in this case, the green fraction was less than 20%), urban emissions ranged from 4 to 18 kg CO2 m -2 yr -1 for other cities Velasco and Roth, 2010). The low CO2 emissions rate in Sakai City was evident in daytime peaks during winter months (Fig. 2) compared with London (e.g., more than 50 10 μmol m -2 s -1 , Ward et al., 2015) and Beijing (30 μmol m -2 s -1 , Liu et al., 2012). Warmer winter temperatures may contribute to lower emissions as a result of reduced building heating and thus lower annual emissions in Sakai City compared with other northern cities. The annual emissions rate in our urban center was comparable to that of the densely populated residential area in Tokyo (4.3 kg CO2 m -2 yr -1 , Hirano et al., 2015).

15
The sensitivity of the CO2 emissions to cold temperatures was comparable to that described in the previous studies (Bergeron and Strachan, 2011;Liu et al., 2012;Pawlak et al., 2011). The effect of building heating has often been estimated as a slope between air temperature and the CO2 emissions rate: -2.02 g C m -2 d -1 °C -1 in London (Ward et al., 2015), -0.21 g C m -2 d -1 °C -1 in Łódź (Pawlak et al., 2011), and -0.35 g C m -2 d -1 °C -1 in Beijing (Liu et al., 2012). These values are comparable to those obtained in our city: -0.46 to -0.27 g C m -2 d -1 °C -1 (Fig. 4). No sensitivities to cold temperatures were found in the 20 urban park (west IZM), rural area (east IZM), or residential area (OPU), could be due to the mixed effects of biological and anthropogenic signals.
CO2 emissions in urban landscapes (SAC and OPU) also increased as temperatures increased in the summer months (Fig. 4): 0.22 g C m -2 d -1 °C -1 in west SAC (p = 0.01), 0.24 g C m -2 d -1 °C -1 in east SAC (p = 0.02), and 0.13 g C m -2 d -1 °C -1 in OPU 25 (p = 0.26). Because traffic did not show a clear seasonal variation (Fig. 2b), the reason for this increase is unclear, but one possibility is the contribution of emissions from gas-powered air conditioners (Fig. A1). The prevalence rate of gas-powered air conditioners is approximately 20% of non-residential buildings, based on an assessment by the Japan Gas Association. A weaker dependence in OPU probably occured because emissions from gas-powered air conditioners from the university building were canceled out by an increase in biological uptake (Fig. 3b). The sensitivity of assimilatory fluxes to warming 30 temperatures was 0.38 g C m -2 d -1 °C -1 in OPU (p < 0.01).
Weekly cycles of CO2 emissions were only observed at urban sites (Fig. 5), representing the strength of human activities.
Previous urban CO2 flux studies reported that major contributors to anthropogenic emissions were vehicle emissions and gas consumption (Gioli et al., 2012;Hirano et al., 2015;Velasco et al., 2005;Ward et al., 2013). Velasco and Roth (2010) indicated that weekly cycles were primarily related to vehicle emissions. Traffic count was high on weekdays at SAC (Fig.   2b), and business offices including the university are often more active on weekdays than weekends and holidays. In contrast, 5 there were no clear weekly cycle in east SAC, urban parks, and rural areas. Lager differences between weekday and holiday in west SAC and OPU suggest the greater contributions of emissions from vehicles and business offices compared with other landscapes. This underscore the importance of temporal variations in CO2 emissions by land use.
The role of the urban park in migrating urban CO2 emissions was limited. Several factors explain the annual emissions from 10 the urban park. First, the urban park frequently suffered from various management activities, such as harvesting and weeding.
Such frequent disturbances could decrease sink and increase source (Gough et al., 2007;Latty et al., 2004). Warmer climates in the urban area may induce higher respiration (Awal et al., 2010). A limited footprint might influence CO2 fluxes arising from emissions from surrounding areas. We re-checked the data selection using stricter criteria according to which we rejected data when 80% of the flux footprint exceeded boundary of the urban park, but the results were almost the same. A 15 positive CO2 budget of 2400 g CO2 m -2 yr -1 was previously measured at an urban park in Germany (Kordowski and Kuttler, 2010).
Partitioning flux data measured at a single site with distinct landscapes is a useful approach in urban flux studies. CO2 fluxes at different landscapes measured at a single site showed considerably different behaviors (Fig. 2, 3, 6). The approach 20 previously used for clarifying variations in fluxes in different landscapes involved single flux measurements Kordowski and Kuttler, 2010;Hirose et al., 2015). The partitioning concurrently contained limitations that data availability decreases with partitioning. In the study area, sea-breeze circulation was dominant in the summer months, resulting in a large data gap from certain wind directions (shown in section 2-3). Accumulating long-term data could be useful for filling the data gap. 25 The green fraction can be useful for upscaling annual CO2 flux in urban area (Fig. 6). Our study indicated that direct mitigation by vegetation was limited in terms of partitioned assimilatory flux (Fig. 3b); thus, the green fraction was an index of human activities. The applicability of the green fraction was previously reported based on a global synthesis based on the eddy covariance measurements in urban areas Velasco and Roth, 2010;Ward et al., 2015). The 30 relationship between the annual CO2 flux and the green fraction in Sakai City tended to be lower than the relationship revealed by the global synthesis   (Fig. 6). This difference might indicate that the relationship differs in each city or country. Consequently, to quantify the effects of the green fraction on CO2 emissions in various cities, further direct measurements of CO2 flux at various urban sites are required.
Upscaled annual CO2 fluxes for the city (Fig. 7) were lower than estimated using the inventory published by the government.
According to the inventory, approximately 57% of CO2 emissions were associated with the industrial sector, but there was no eddy covariance site in the coastal industrial region. Part of the discrepancy occurred because our upscaling estimated the net flux of urban emissions and vegetative uptake, whereas the inventory quantified the emissions. Hirano et al. (1996) 5 estimated that vegetation in Sakai, primarily in southern sectors, absorbed 0.87 Mt CO2 yr -1 of CO2 based on an inventorybased estimates. Another reason for the discrepancy was that our estimate did not include hot spot emissions, such as power plants and incineration facilities, or non-CO2 gas emissions. Oda and Maksyutov (2011) estimated that approximately half of annual total CO2 emissions were from point sources in most countries. Because our upscaled CO2 flux did not include such point sources, CO2 emissions from point sources could be more rigorously quantified using the governmental inventory 10 than non-point sources (Oda and Maksyutov, 2011). Thus, upscaled CO2 flux could be useful as an additional constraint, providing more information regarding CO2 emissions from non-point sources. This method could be useful for estimating regional CO2 emissions when combined with inventory estimates.

Conclusion
Based on continuous measurements using the eddy covariance method at three different urban sites, diurnal, weekly, 15 seasonal, and spatial variabilities in CO2 flux were evaluated in Sakai, Osaka, Japan. The urban center and university sites acted as CO2 sources throughout seasons. A clear weekday/holiday cycle of CO2 emissions was observed at those sites. A diurnal pattern in the urban center was correlated with those for traffic count. High emissions were observed in the urban site in both winter and summer months, even though the traffic did not change seasonally, suggesting that changes in gas consumption influenced the seasonal variabilities. The urban park and rural area exhibited CO2 uptake during summer 20 months, with distinct daytime uptake. Regardless of the green fraction, all landscapes considered in this study acted as an annual CO2 source. Previous studies of the relationship between annual CO2 emissions and the green fraction for a wide spectrum of global cities Velasco and Roth, 2010) revealed slightly differing patterns from that observed in our city. Consequently, the relationship based on eddy covariance data within a single city could be useful to evaluate CO2 emissions at the city scale. The network of eddy covariance measurements within a city is a very powerful tool 25 to evaluate CO2 emissions in urban areas.

Acknowledgments
We thank Dr. Hiroyuki Kaga of Osaka Prefecture University for supporting the GIS analysis. We thank the people of Sakai City Office for supporting measurements at SAC. The measurements at IZM were supported by the Sumitomo Foundation 30 (143205). The measurements at SAC were partly supported by Nissei Foundation grants for Environmental Problems, H21.
Traffic data regarding Hanshin Expressway were provided by the Hanshin Expressway Company. Data on gas consumption by Osaka Prefecture University were provided by the university.