2.1 APOLLO stations

The APOLLO is a simple and low-cost data logger that records air temperature and pressure. Inspired by the pyranometer network stations used during the HD(CP)² Observational Prototype Experiment (HOPE) campaign (Madhavan et al., 2016), it operates autonomously without an external power source, and its sensor and control units are mounted on a short mast (Fig. 1). The APOLLO is equipped with a fast-response and moisture-resistant thermometer consisting of a small (7 mm×1.2 mm) NTC (negative temperature coefficient) thermistor probe (type: TE Connectivity GA10K3MRBD1; Fig. 2a) placed inside a passively ventilated radiation shield (see sensor specifications in Table 1). The control unit is based on a micro controller board (type: HIMALAYA Matrix-Core ESP32) located inside a small (26 cm × 17 cm × 10 cm) logger box (Fig. 2b). A digital air pressure sensor (type: Bosch BME280 Environmental sensor) is installed on the logger main board, and a counterbalance valve ensures rapid adjustment of barometric pressure inside the logger box.

Table 1Sensor specifications of APOLLO station (TE Connectivity, 2015; Bosch, 2020).

Download Print Version | Download XLSX

Figure 2(a) NTC thermometer with scale for reference and (b) logger main board of APOLLO station. Marked are the micro controller board (1), digital air pressure sensor BME280 (2), micro-SD memory card (3), GPS receiver module (4), long-range (LoRa) module (5), and pressure counterbalance valve (6).

Download

The readings of the temperature and pressure sensors are sampled with a resolution of 1 s and recorded onto a micro-SD memory card. An on-board GPS receiver is used to initially synchronize the internal real-time clock on start of the logger. To minimize power consumption during operation, the GPS module is activated to log the current time only once per hour for about 10 s which allows correction of the logger time stamp for drift in a post-processing step. The logger can establish a local WiFi hotspot for on-site control, data inspection and fast data access via download, and is able to send status messages via the LoRa (long-range) wireless communication protocol for live monitoring.

The APOLLO is powered by a standard USB power bank battery (20 Ah capacity) that corresponds to an autonomous operation time of 10 to 14 d. For easy access during maintenance, the power bank is placed inside a separate plastic tube. All logger units are mounted on a 3 m mast that is anchored in the soil with a ground screw. In the standard configuration, the temperature sensor at the top of the mast and the pressure sensor inside the logger box are situated about 2.9 and 1.7 m above ground, respectively. The material costs for all required parts sum up to around EUR 300 per station.

Figure 3Air temperature perturbation observed by a co-located ultrasonic sensor and APOLLO station during a cold pool event at the Meteorological Observatory Lindenberg – Richard Aßmann Observatory (eastern Germany) on 26 August 2019. The readings of the ultrasonic sensor are 1 s averages of 20 Hz acoustic temperature measurements by an ultrasonic anemometer (type: METEK uSonic-3 Scientific).

Download

The APOLLO station proves to be especially suited for accurately recording sharp thermal boundaries associated with cold pool front passages. In wind tunnel laboratory tests, the shielded NTC temperature sensor exhibits a response time (i.e., e-folding time constant) of $\mathit{\tau }=\mathit{\{}\mathrm{93},\mathrm{14},\mathrm{10}\mathit{\}}$ s ± 10 % at $\mathit{\{}\mathrm{1},\mathrm{3},\mathrm{5}\mathit{\}}$ m s⁻¹ wind speed when being artificially heated and released to room temperature. Since cold pools are usually associated with wind speed increases of 3 m s⁻¹ and more (Kirsch et al., 2021a), the lack of active sensor ventilation is unlikely to impact the accuracy of temperature measurements of cold pool passages. The comparison with an ultrasonic sensor confirms that the NTC thermometer captures well the rapid cooling signature of a cold pool event (Fig. 3). Although the response time of the APOLLO sensor is too large to respond to sub-second temperature fluctuations in the same way as the inertia-free ultrasonic sensor, it does not show any apparent lag in the minute-scale shape of the signal and even captures second-scale fluctuations. Moreover, trial field measurements of two cold pool events show that external site conditions, such as surrounding obstacles or different surface properties, do not systematically impact the strength and shape of a cold pool temperature signal (not shown) and, therefore, do not restrict the choice of measurement sites for APOLLO stations. Since this is not true for observations of wind speed and wind direction, these are only performed by the WXT weather stations at appropriate locations. Similarly, humidity measurements are limited to the WXT stations to avoid the technical and financial effort needed to equip the APOLLO stations with precise sensors.

Table 2Sensor specifications of WXT weather station (Vaisala, 2017).

Download Print Version | Download XLSX

2.2 WXT weather stations

WXT weather stations are employed to complement the APOLLO stations by providing information on other common meteorological parameters at selected locations. The main component of the station is a commercial compact multi-parameter sensor (type: Vaisala Weather Transmitter WXT536) that measures air pressure, temperature, relative humidity, wind speed, wind direction, and precipitation (see sensor specifications in Table 2). Pressure, temperature, and relative humidity measurements are performed by a pressure–temperature–humidity (PTU) module inside a radiation shield that combines a capacitive silicon BAROCAP sensor (pressure), a capacitive ceramic THERMOCAP sensor (temperature) and a capacitive thin film polymer HUMICAP180 sensor (relative humidity). Wind speed and wind direction are detected by an WINDCAP ultrasonic anemometer on top of the radiation shield that consists of three equally-spaced ultrasonic transducers on a horizontal plane. The wind measurements are determined from the transit times of the ultrasound along the three paths with a sampling rate of 4 Hz and are internally averaged. Wind gusts are defined as the highest average wind speed of a 3 s interval which is internally updated every second. The precipitation measurement principle of the WXT536 sensor is based on a RAINCAP piezoelectrical sensor underneath a steel cover that detects the impacts of individual rain drops within a collecting area of 60 cm². Since the individual signals are proportional to the size of the rain drops, the sensor is able to derive the accumulated rainfall amount within the measurement interval. The sensor also distinguishes between impacts of rain drops and hail stones (Vaisala, 2017). Additionally, the WXT station is equipped with an external Pt1000 thermometer (type: TMH cable sensor Pt1000 1/3 DIN Klasse B; size: 40 mm × 3 mm) inside a separate radiation shield that allows for smaller response times of the measured temperature signal ($\mathit{\tau }=\mathit{\{}\mathrm{173},\mathrm{55},\mathrm{39}\mathit{\}}$ s ± 10 % at $\mathit{\{}\mathrm{1},\mathrm{3},\mathrm{5}\mathit{\}}$ m s⁻¹ wind speed) than the internal PTU module ($\mathit{\tau }=\mathit{\{}\mathrm{334},\mathrm{213},\mathrm{155}\mathit{\}}$ s ± 10 % at $\mathit{\{}\mathrm{1},\mathrm{3},\mathrm{5}\mathit{\}}$ m s⁻¹ wind speed).

The WXT data are sampled at 10 s resolution and written onto an SD memory card by a data logger (type: Avisaro M22766) that is synchronized with an integrated GPS module. Each station is powered by a 12 V lead battery. The battery is recharged by a solar panel (67 cm × 41 cm) so that the station can operate autonomously for several months depending on the available sunlight. Similar to the APOLLO design, the station components are mounted on a 3 m mast that is anchored in the soil with a ground screw (Fig. 4).

Figure 4Components of WXT weather station.

Similar to the APOLLO stations, site conditions like surrounding obstacles and surface properties are not expected to have a systematic impact on cold pool signals in temperature, pressure, and humidity that are specific to the WXT sensors. However, wind speed and wind direction are heavily affected by nearby obstacles like buildings or vegetation due to the low installation height of 3 m above ground. Therefore, the individual site properties have to be taken into account when choosing a measurement location and interpreting the wind data of the WXT stations.

3.1 Experiment area

Hamburg is the second largest city in Germany (population: 1.9 million; 2019) and is located in northern Germany (53.5^∘ N, 10.0^∘ E) about 80 km from the North Sea and the Baltic Sea. The FESST@HH measurement sites cover an area of 50 km × 35 km that includes the urban center of Hamburg and its rural surroundings (Fig. 5). The station network consists of 82 APOLLO sites (Table 3) and 21 WXT sites (Table 4) that are non-uniformly scattered over the domain with a generally higher station density closer to the city center. The arrangement of stations results from the location of private gardens and of public facilities like schoolyards that could be used as measurement sites (see Sect. 3.2). Moreover, the sites for WXT stations were chosen based on a small impact of surrounding obstacles on the local wind field and a roughly uniform distribution over the network. The average nearest-neighbor distance between all 103 measurement sites is 1.85 km with a standard deviation of 1.42 km, whereas it is 1.93±1.41 and 5.49±1.98 km for APOLLO and WXT, respectively. The measurement area is crossed by the Elbe river in southeast–northwest direction and is characterized by relatively flat terrain. While the Elbe valley is situated around sea level, the elevation north and south of it does not exceed about 80 and 110 m, respectively. The altitudes of all measurement sites lie between 1 and 82 m above sea level with an average of 17 m.

To characterize the environmental properties of the measurement sites, we apply the local climate zone (LCZ) framework introduced by Stewart and Oke (2012). This framework classifies the impacts of surface structure, surface cover, and human activity on the local thermal climate with the help of 17 standardized LCZ classes. All 103 FESST@HH sites fall into 15 different LCZ classes, which also includes combinations of classes to describe the heterogeneity within about 500 m around the station (Bechtel et al., 2015). The majority of sites (61) are situated in an open arrangement of low-rise buildings and scattered trees (LCZ 6; mixed in Fig. 5), while 24 sites mostly near the city center are located in the proximity of mid-rise and high-rise buildings either in a compact or open arrangement (LCZ 2, 4, and 5; urban). Each of the other occurring classes contain less than 10 sites which include natural environments (LCZ 9, A, B, and D) and industrial areas with mostly paved surfaces (LCZ 8 and 10). Ten sites are situated less than 50 m away from water bodies like the Elbe river, the Alster Lake, or channels (LCZ G).

Figure 6Examples of instrument installations during FESST@HH: APOLLO station (a) as the standard installation in a private garden, (b) attached to an existing weather station on a schoolyard, (c) on a boat landing stage at a channel, (d) on an air quality observation site at a public road, (e) on the balcony of the building of the Meteorological Institute, and WXT weather station (f) as the standard installation on an agricultural field, (g) on top of a container, and (h) on the roof-deck of the building of the Meteorological Institute.

3.2 Implementation

The measurement period started on 1 June 2020 and ended on 31 August 2020, whereas the installation of all stations was completed in mid-June. The realization of the experiment was enabled by the support of many institutions and private landowners who provided permission at short notice to use their premises as measurement sites. The different groups of landowners are indicated by two-letter acronyms as part of a unique site identifier code that also contains the site number and the installed instrument type (a = APOLLO, w = WXT). The majority of measurement stations were installed in private gardens and backyards (PG; Fig. 6a) and on the grounds of various institutions and clubs (OG). Further groups of sites include existing weather stations and sites of the Meteorological Institute (WS and MI), air quality observation sites of the Free and Hanseatic City of Hamburg (LM), grounds of the University of Hamburg (UH), and schoolyards (KB). A special characteristic of the latter group is that most of the respective APOLLO logger components are mounted at existing weather stations used for educational purposes and, therefore, exhibit lower temperature sensor heights (∼1.8 m) than the standard stations (Fig. 6b). Further deviations from the standard APOLLO installation include four loggers mounted on balconies (040UHa, 092PGa, 111OGa, and 113PGa; Fig. 6e) and two WXT stations installed on top of a container (017LMw; Fig. 6g) and on a roof-deck of the building of the Meteorological Institute (114UHw; Fig. 6h). Due to technical issues, minor changes to the instrumental setup had to be implemented during the measurement period with the replacement of one APOLLO (068OGa) and two WXT stations (048OGw and 114UHw) as well as the replacement of one APOLLO by a WXT station (083PGa).

Ensuring the implementation of FESST@HH was compatible with pandemic-related restrictions affected not only the selection of measurement sites but also the maintenance strategy. Instead of a small team servicing the entire network, the maintenance work was shared between nearly 40 people. Private landowners who provided measurement sites in their backyard could also volunteer to regularly change the battery, check the data, and upload it to a FTP server. Other stations located on public grounds, schoolyards, or industrial premises were serviced by colleagues living nearby. The main benefit of this maintenance strategy was that the collective effort kept the individual workload very low and promoted the continuous operation of the instruments, which eventually eased the execution of the experiment under challenging circumstances.

4.1 Level 0 data

The raw measurement data are stored in ASCII files. The APOLLO data logger creates a new file at the start time, which contains the internal logger time, the GPS-synchronized time, the resistance readings of the NTC thermometer, the pressure readings of the BME280 module, and status information at 1 Hz frequency. By contrast, new level 0 WXT data files are created at midnight (UTC) on a daily basis. Each line in the files consists of a GPS-synchronized time stamp and the measurement data of a specific sensor module, whereas each module sends data at 10 s intervals.

4.2 Level 1 data

Measurement data which have passed the first processing step are called level 1 data. At this stage, basic standardization procedures are performed to allow for easy access to the data for preliminary analyses but not quality checks. Most importantly, this includes the correction and homogenization of the data time stamps. For the APOLLO data, the internal logger time is corrected with its deviation from the most recent valid GPS time stamp that is logged once per hour. Data for missing time steps are filled with placeholder values (NaN). Furthermore, daily files are created by splitting multi-day files and merging sub-daily files. In case of the WXT data, all time stamps are moved to a regular 10 s time grid. The measurement data itself are not manipulated except for the APOLLO thermometer readings that are converted from resistance to temperature. For both station types, level 1 data are stored in a standard ASCII format.

4.3 Level 2 data

The purpose of the second processing level is to generate a standardized and quality-controlled data product that is ready to use for scientific analyses. Since the raw APOLLO measurements contain different kinds of erroneous data which originate from technical issues of the logger or characteristics of the measurement site, we apply several quality checks and corrections to clean up the data. This processing step is not required for WXT data due to internal quality checks implemented by the manufacturer of the sensor. In total, less than 1 % of the level 1 data fail the quality criteria and need to be removed. The correction procedure contains the following steps:

• removing erroneous temperature and pressure data outside plausible limits defined by $\mathrm{0}\le T\le \mathrm{40}$ ^∘C and $\mathrm{950}\le p\le \mathrm{1050}$ hPa, respectively;

• removing spikes in temperature (pressure) larger than ΔT=0.5 K (Δp=1 hPa), where ΔT (Δp) denotes the absolute difference from the 30 s (60 s) running median value. For the temperature data, spikes are only removed for phases of active WiFi connection which produces the spikes;

• removing temperature data showing implausibly large differences from the current network mean value defined as $|T-\stackrel{\mathrm{‾}}{T}|>\mathrm{15}$ K. For three stations (037UHa, 040UHa, 068OGa) with erroneous NTC temperature sensors producing unphysical spikes, a criterion of $|T-\stackrel{\mathrm{‾}}{T}|>\mathrm{2}$ K is applied, and all data within a time window of 15 min before and after erroneous data are also removed;

• manually removing single periods of erroneous temperature and pressure measurements that are not filtered by the previous criteria. In case of station 039UHa, temperature measurements on more than 40 d are affected by the warm air outlet of a nearby air-conditioning facility and are manually removed;

• applying a 10 s running mean smoothing procedure on the temperature measurements of station 113PGa to account for an anomalously high noise level of the NTC sensor; and

• correcting the individual biases of NTC temperature sensors. The biases are determined from one-week-long calibrations with respect to a reference WXT sensor (114UHw) between January and June 2020. The pressure readings of the BME280 sensors are not calibrated since test measurements of the absolute values showed an oscillating drift pattern with an amplitude and period of about 0.5 hPa and 10 d, respectively. The WXT536 and Pt1000 sensors are calibrated by the manufacturer.

The level 2 data (Kirsch et al., 2021b) are stored in the NetCDF4 data format (Eaton et al., 2017) while the naming of files, variables, and attributes complies with the SAMD (Standardized Atmospheric Measurement Data) product standard (Lammert et al., 2018). Accordingly, all measurement data of the same variable and from the same type of instrument are merged and stored in daily files. Note that the time variable expressed as seconds since 1 January 1970, 00:00:00 UTC ignores leap seconds (i.e., occasional adjustments of UTC time to variations in Earth's rotation velocity). Also note that the time drift correction and quality control procedures cause irregular and isolated data gaps that may affect only single seconds, which has to be taken into account when using the data set at its full time resolution. The file header contains all relevant metadata describing the individual measurement sites, namely, the station identifier, the station name, the geographic coordinates, the altitude above sea level, the sensor height above ground, and the LCZ. Attached to the data set are also images of all measurement stations and panoramic views of the surroundings of all WXT stations to support the interpretation of the corresponding observation data.

Figure 7FESST@HH level 2 data from 1 June to 31 August 2020: (a) daily mean fraction of valid temperature observations over all APOLLO and WXT stations; (b) daily median (filled dots), 5th percentile and 95th percentile (whiskers), and minimum and maximum (open circles) of all WXT air temperature observations; and (c) individual accumulated rainfall amount at all WXT stations (open circles). Filled dots mark the daily median. Days are defined in UTC.

Download

5.1 Data availability

The level 2 data set of FESST@HH covers the period from 1 June to 31 August 2020 (Fig. 7a). The average availability of valid temperature observations is 83.2 % and 87.6 % for all APOLLO and WXT stations, respectively. Since the three-month period also includes the installation phase of the instruments during the first half of June, these numbers increase to 90.0 % and 94.3 % if only the period after 15 June is considered. During this period, the daily availability of APOLLO and WXT measurements is always larger than 82.6 % and 90.0 %, respectively. Apart from the removed measurements affected by erroneous sensors and local site conditions, most of the data gaps in the APOLLO data are due to missing power supply of the loggers either caused by technical issues related to the power bank batteries or insufficient maintenance. In only one case was a battery stolen. The greater stability of the WXT power supply is also the reason for the generally greater availability of WXT data compared to APOLLO data.

5.2 Weather conditions

The weather conditions during the FESST@HH period covered the full range of a typical mid-latitude summer. The air temperature measured by the station network ranges between a minimum of 5.3 ^∘C on 13 July and a maximum of 35.6 ^∘C on 8 August (Fig. 7b; days defined in UTC). During the entire 92 d period, the maximum temperature exceeded 25 ^∘C on 41 d and 30 ^∘C on 16 d, including an exceptionally long period of 12 consecutive days with maximum daily temperature exceeding 30 ^∘C between 6 August and 17 August. In contrast, July was characterized by relatively cold temperatures with a maximum of above 25 ^∘C on only 6 d. Measurable rainfall, defined as a daily rainfall accumulation of more than 1 mm at at least one WXT station of the network, was observed on approximately half of the days from 1 June to 31 August (Fig. 7c). On 11 d, mostly between 27 June and 10 July, the median rainfall amount exceeded 5 mm. A maximum daily accumulation of 28.2 mm was observed on 27 June in association with a strong convective event.

6.1 Cold pools

During the three-month measurement period, the station network recorded 37 cold pool events. We define events as periods with at least five stations per hour satisfying the cold pool detection criterion described in Kirsch et al. (2021a), i.e., a local temperature drop of at least 2 K within 20 min. This rather conservative threshold was chosen to robustly discriminate the cold pool signals from other potential sources of spatial temperature variability. The plausibility of cold pool detections is checked based on the visual inspection of corresponding rain radar imagery by an X-band radar of the University of Hamburg (Burgemeister et al., 2022) and the operational C-band radar network of the German Weather Service. The number of approximately 12 events per month is larger than the long-term average at the Hamburg weather mast site (Bruemmer et al., 2012) of about seven events per month (Kirsch et al., 2021a) as expected from the larger areal coverage compared to a point measurement. Figure 8 shows an overview of all 37 identified cold pool events and illustrates the spatial variability of maximum temperature perturbations detected within the station network. Most of the events (28) exhibit a median temperature signal of −4 K or weaker, however, the variability in signal strength and number of affected stations largely vary for different events. To demonstrate the ability of the station network to capture the characteristics of an individual cold pool and to shed light on its internal structure, we present one case study.

Figure 9Perturbation in air temperature since 12:30 UTC from (a) 13:05 UTC to (d) 14:05 UTC observed by APOLLO (circles) and WXT stations (crosses) during a cold pool event on 10 August 2020. Bluish colors mark stations inside the cold pool defined by a temperature perturbation stronger than −2 K (color map by Crameri, 2018; Crameri et al., 2020). Black arrows indicate instantaneous wind speed and direction observed by WXT stations. The length of the reference arrow corresponds to 5 m s⁻¹ wind speed. The four highlighted stations refer to the respective data illustrated in Fig. 10.

On 10 August 2020 around 12:45 UTC, a strong and nearly stationary convective precipitation cell developed southeast of the city center. About 10 min later, the formation of a cold pool became evident from a rapid cooling of the surface-layer air. Figure 9 illustrates the evolution of the cold pool with snapshots of the temperature perturbations relative to the pre-event state observed by the station network. About 20 min after initiation of the convective cell, the cold pool exhibited a temperature perturbation of up to −8 K within an area of less than 10 km in diameter (Fig. 9a). After another 20 min, the temperature perturbation strengthened to about −11 K and the cold pool expanded to a size of roughly 10 km × 20 km (Fig. 9b). This process continued until around 14:00 UTC when the cold pool reached its maximum temperature perturbation of about −12 K and a diameter of nearly 30 km when assuming a roughly elliptical shape of the cold pool, whereas its northeastern parts were outside of the network (Fig. 9d). The associated near-surface wind field observed by the WXT weather stations indicated a strong divergent flow at the cold pool center especially during the early stages of the cold pool's life cycle (Fig. 9a and b). Consistent with expected characteristics of a cold pool, the radial expansion of the cold-air region during the later stages was also present in the wind observations predominantly southwest of the cold pool center (Fig. 9c and d).

The time series of temperature measurements at selected stations of the network provide further insights into the spatial structure and life cycle of the cold pool (Fig. 10a). APOLLO station Luxweg (104PGa) near the cold pool center experienced an initial temperature drop of approximately 8 K within 5 min, that continued at a slower rate afterwards and reached a maximum perturbation of −12 K after 1 h. This value is on the order of strongest temperature perturbations expected during cold pool passages in Hamburg (Kirsch et al., 2021a). The readings of the stations Ochsenwerder Norderdeich (082PGa) and Obsthof Lehmbeck (054OGa), located 4 and 12 km further south, respectively, reveal that the maximum amount of cooling decreases by about two-thirds away from the center, which suggests a highly heterogeneous temperature structure inside the cold pool. Since the site Luftmessnetz Habichtstraße (018LMa), situated northwest of the cold pool center, experienced no cooling at all, the cold pool produced a gradient in temperature perturbation of 12 K within a distance of 11 km. This result is in line with the findings of van den Heever et al. (2021), who reports a variation of near-surface cold pool temperature perturbations of order 1 to 2 K on scales between 𝒪(100) m and 𝒪(1) km. The time lags between the cooling signals observed on the southern side of the cold pool also allow us to estimate its propagation velocity. Based on time lags of about 15 and 50 min for travel distances of 3.9 km (104PGa–082PGa) and 11.9 km (104PGa–054PGa), respectively, the velocity is about 4 m s⁻¹. This value is lower than the propagation velocity of 6.7 m s⁻¹ found for a cold pool event in north-central Oklahoma (Borque et al., 2020), but is plausible considering that the parent precipitation cell was almost stationary and, therefore, the observed propagation of the cold outflow air was not biased by the movement of the cell itself.

Figure 10Perturbation in (a) air temperature and (b) pressure since 12:30 UTC observed by the four selected APOLLO stations Luftmessnetz Habichtstraße (018LMa), Obsthof Lehmbeck (054OGa), Ochsenwerder Norderdeich (082PGa), and Luxweg (104PGa) during a cold pool event on 10 August 2020. Locations of the stations are highlighted in Fig. 9.

Download

The accompanying air pressure measurements at the four selected sites confirm that the negative temperature perturbation of the cold pool is associated with a typical hydrostatic pressure rise (Fig. 10b). However, the positive pressure perturbation was well-pronounced only at station 104PGa while the amplitudes were not proportional to the strength of the corresponding cooling signals. The pressure signal at station 104PGa also experienced rapid fluctuations of about 1 hPa that were not apparent at the three other sites. Although an in-depth investigation of such effects lies beyond the scope of this study, this may indicate that the spatial structure of a cold pool differs between temperature and pressure, and that mechanisms other than hydrostatic cooling control the pressure signal near the center, like dynamic effects already described in early studies (i.e., Wakimoto, 1982; Mueller and Carbone, 1987).

Figure 11(a) Mean diurnal cycle of urban temperature modification ΔT_city and (b) nighttime (21:00 to 03:00 UTC) mean of ΔT_city at all APOLLO and WXT stations from 15 June to 31 August 2020. Highlighted in orange and green are the urban station Wetterstation HafenCity (010WSa) and the rural station Obsthof Schuback (055OGw), respectively. © OpenStreetMap contributors 2021. Distributed under the Open Data Commons Open Database License (ODbL) v1.0.

6.2 Urban heat island

The FESST@HH data set is well suited for studying urban climate effects as the station network covers both the city center and its rural surroundings. Stations near the city center record weather conditions with urban modifications whereas rural stations provide undisturbed references. Focusing on air temperature, we define the urban modification ΔT_city as the difference between the locally observed air temperature and the undisturbed, natural temperature T_nat determined exclusively by large-scale weather conditions. On scales of 𝒪(10) km without orography, it is reasonable to assume that T_nat varies only linearly in space. We estimate the mean and the slope of this linear temperature field with a regression using all stations in weakly sealed (i.e., mostly natural) environments defined by a maximum sealing up to open low-rise (LCZ 6). The mean diurnal cycle of ΔT_city features a step function at almost all stations (Fig. 11a). During daytime, the urban effects are small and ΔT_city is close to zero everywhere. In contrast, during nighttime, predominantly between 21:00 and 03:00 UTC, ΔT_city reaches a constant level at all stations, which largely reflects the well-known nocturnal urban heat island effect. For example, the most central station Wetterstation HafenCity (010WSa; orange in Fig. 11a) is almost 2.5 K warmer than estimated T_nat and the most rural site Obsthof Schuback (055OGw) is even 1.5 K colder than the reference. The spatial pattern of the mean nocturnal urban heat excess is in agreement with the typical heat island structure featuring largest values close to the city center (Fig. 11b). Further analysis of individual days (not shown) provides indication that the advection of the heat island effect causes downwind heat plumes to form which affect specific outskirts depending on the prevailing wind direction.

Figure 12(a) Mean diurnal cycle of standard deviation of 1 min temperature fluctuations σ_T at all APOLLO stations (gray lines), and at the highlighted stations Stadtteilschule Blankenese (yellow; 033KBa), Schule Redder (orange; 036KBa), Luftmessnetz Rothenburgsort (blue; 020LMa), and Uni Sternwarte (green; 043UHa). Map data © Google Earth 2021. (b) Net radiation at the Hamburg weather mast site dependent on σ_T at station Wetterstation Zollamt both averaged over 10 min. Blue boxes and orange bars indicate the interquartile range and median for the respective σ_T bin. The analyses in both (a) and (b) are valid for sunny days from 1 June to 31 August 2020, whereas sunny days are defined as days on which the daily averaged global radiation at Hamburg weather mast is larger than at least one-third of the daily averaged theoretical maximum of clear-sky insolation.

6.3 Turbulent temperature fluctuations

The NTC temperature sensor of the APOLLO station responds rapidly to air temperature fluctuations. The recorded time series with a sampling rate of 1 Hz partly resolves temperature eddies related to turbulent sensible heat fluxes. The high frequency temperature fluctuation expressed by the standard deviation of temperature readings within 1 min intervals, σ_T, features a well-pronounced diurnal cycle (Fig. 12a). During nighttime, σ_T is close to 0.05 K at all stations, which most likely resembles the noise level of the instruments. However, at daytime, σ_T is well above that noise level and follows the common diurnal cycle of net radiation or sensible heat flux at the surface. Turbulent fluxes are strongly determined by local surface conditions like albedo, sealing, or vegetation cover. Likewise, the differences in the diurnal cycle of σ_T can be explained by local surface conditions: the station Stadtteilschule Blankenese (033KBa) at a schoolyard next to a sports ground (low albedo, no evapotranspiration) with free insolation exhibits the highest temperature variability. High values also occur at the schoolyard of station Schule Redder (036KBa) but are shifted and limited to morning hours due to a building west of the station casting a shadow and reflecting sunlight. Even more extreme, buildings shade the station Luftmessnetz Rothenburgsort (020LMa) almost the whole day except during the late afternoon. The lowest temperature variance is observed at station Uni Sternwarte (043UHa), the only station below tall trees. The heat exchange within the tree canopy effectively reduces the energy exchange at the ground.

The relation between surface heat exchange and temperature variability is also directly confirmed by correlating the net radiation observed at the Hamburg weather mast reference site with σ_T at the neighboring APOLLO station Wetterstation Zollamt (011WSa) located about 750 m away (Fig. 12b). Values of σ_T lower than 0.75 K are strongly correlated to situations where the net radiation is close to zero. Increasing variance beyond this threshold is mostly explained by increasing net radiation (r²=0.77). Instead of the sensible heat flux, we use the net radiation for this analysis to minimize the influence of land surface differences between the two sites (pasture at Wetterstation Zollamt versus sealed area at Hamburg weather mast). Additional measurements on a test site show that σ_T does not change if the sensor shield is protected from direct solar radiation (not shown). This finding supports our reasoning that the APOLLO stations capture turbulent temperature fluctuations caused by the surface heat exchange in the near environment of the observation site rather than artificially generated by the heating of the sensor shield. Direct measurements of turbulent heat fluxes and surface properties like temperature and moisture added in the FESSTVaL follow-up experiment will help to better understand the land–atmosphere interactions for different land classes.