Advertisement

  • Loading metrics

Open Access

Peer-reviewed

Research Article

Finding the sea level – Investigation of rural references for estimation of the urban heat island

Abstract

The urban heat island (UHI) effect is one of the most studied phenomena in urban climatology. Numerous studies have revealed the heterogeneous nature of air temperature within cities, manifesting as an urban heat “archipelago” with small-scale air-temperature differences and multiple hot and cold spots rather than a single, uniform hot spot in the city core. With the introduction of the local climate zones (LCZs) scheme, close attention has been paid to the definition and description of “urban” and “rural” sites. However, what remains understudied and inconsistent across studies is the “sea level” of the “archipelago,” i.e., defining the air-temperature conditions of the surroundings, unaffected by the city. Here, we compare definitions and requirements of that “sea level,” and investigate multiple possible data sets for UHI calculation. Most typically, single weather stations, often at airports, are used as a “rural” reference. However, these stations are not ubiquitously available and typically suffer from effects such as a high fraction of impervious surfaces or urban heat advection when located downwind of the city. Crowd weather stations (CWS), which have gained attention in recent years in urban climate studies due to their abundance, are often affected by nearby buildings and are mostly located in urban areas. Besides station-based data, reanalysis products such as ERA5-Land could provide an independent reference, as they are available globally and often do not consider urban areas. In this study, all three data sources were compared against an ideal case of multiple professional weather stations placed around the city. We investigate the two temperate European cities of Paris (France) and Berlin (Germany) during six years (2019–2024), focusing on crowdsourced data from CWS. We find that ERA5-Land air-temperature data is a consistent and ubiquitously available reference for the definition of “rural,” providing UHI-calculations most comparable to the ideal case of having multiple professional weather stations. Using it as a universal rural reference allows for comparison between cities and further enables exploiting the potential of CWS, even in regions with few stations and a lack of professionally-operated rural weather stations. Establishing a consistent “sea level” for urban air-temperature and UHI-studies enables comparison between various cities globally and allows for the integration of different data sources, such as local networks of weather stations or CWS, on a larger scale.

Citation: Kittner J, Fenner D, Demuzere M, Bechtel B (2026) Finding the sea level – Investigation of rural references for estimation of the urban heat island. PLOS Clim 5(7): e0000955. https://doi.org/10.1371/journal.pclm.0000955

Editor: Jingyu Wang, Nanyang Technological University, SINGAPORE

Received: December 22, 2025; Accepted: May 12, 2026; Published: July 7, 2026

Copyright: © 2026 Kittner et al. This is an open access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.

Data Availability: The underlying CWS data can be accessed via the Netatmo public /getpublicdata (https://dev.netatmo.com/apidocumentation/weather#getpublicdata) and /getmeasure (https://dev.netatmo.com/apidocumentation/weather#getmeasure) API. The global map of Local Climate Zones by Demuzere et al. 2022 is publicly available for download from zenodo via: https://doi.org/10.5281/zenodo.6364593. ERA5-Land data (Muñoz-Sabater et al. 2021) is available via Google Earth Engine: https://developers.google.com/earth-engine/datasets/catalog/ECMWF_ERA5_LAND_HOURLY or alternatively the CDS: https://cds.climate.copernicus.eu/cdsapp#!/dataset/reanalysis-era5-land?tab=overview.

Funding: This work was partially carried out in the context of the ENLIGHT project (https://www.geographie.rub.de/klima/enlight.html.en), funded by the German Research Foundation (DFG) under grant No. 437467569 with additional funding received from the Institute of Geography, Faculty of Geosciences, Ruhr University Bochum by JK. In addition, MD was partially supported by the European Union’s HORIZON Research and Innovation Actions under grant agreement No. 101137851, project CARMINE (Climate-Resilient Development Pathways in Metropolitan Regions of Europe, https://carmine-project.eu). DF was partially funded by the European Research Council (ERC) under the European Union’s HORIZON 2020 research and innovation program under grant agreement No. 855005 (urbisphere, https://urbisphere.eu) and partially by the Technische Universität Berlin. JK and BB received a salary from the Institute of Geography, Faculty of Geosciences, Ruhr University Bochum. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.

Competing interests: The authors have declared that no competing interests exist.

Introduction

The urban heat island (UHI) is probably the most studied and taught phenomenon in urban climate sciences [1–3] since it is an easy-to-grasp concept and description of the multitude of complex processes driving the heterogeneous climate of cities. However, it has since been widely acknowledged that the altered urban climate cannot be characterized as an ideal “island” but instead an “archipelago” with many localized hot and cool spots caused by the different forms and functions of a city [4–6].

UHI studies specifically focus on urban-rural temperature differences. When based on air temperature, these differences are often derived from a single station pair or multiple stations. In the past, it has been criticized that the definition of what kind of urban and what kind of rural was often imprecise, making it impossible to compare methods and studies [2]. With the introduction and, as of today, wide adoption of the local climate zone (LCZ) scheme [7], the challenge of systematically characterizing urban areas has been addressed and numerous studies have shown the validity of the concept [8–13]. The scheme provides a precise, standardized, and basically globally-valid description of urban areas. Moreover, with the introduction of the LCZ Generator, an easy way to create city-specific maps is publicly available [14].

The second part of the UHI definition, i.e., the definition of the rural reference, has received much less attention, despite the potential application of the LCZ scheme also to natural surfaces. However, the rural reference is equally important and a frequent source of error or uncertainty [13,15–20]. World Meteorological Organization (WMO) guidelines [19] give precise specifications for the ideal placement and setup of a measurement site (land station). The station’s location should be representative of an area of 2 000 km2 to 10 000 km2, and the plot of land should be approximately 1 ha in size. It must be protected from the influence of industry and situated well away from constructions or woods, specifically at least 10–20 times the height of the closest obstruction. In addition, it should be at least 100 m away from bodies of water. The observing area itself should be ideally no smaller than 25 m x 25 m and have a north-south orientation [19]. Areas that meet all criteria and are accessible for the setup of a weather station rarely exist. Adding to this complexity, studies often fail to precisely describe the rural reference used [21], introducing additional uncertainties.

Most commonly, UHI studies use a single rural weather station [15,22–35], sometimes only paired with a single urban station. Less commonly, likely due to limited availability, studies utilize multiple rural weather stations [16–18,36–42]. Such networks of rural stations are often only available for large and developed cities such as Berlin (Germany) [43], Paris (France) [39], Seoul (South Korea) [17], and Phoenix, US [16] (see: S1 Appendix Table A for a curated overview of selected studies following [2]). The combination of multiple weather stations forms a more robust reference in which local effects are less pronounced, resulting in an “artificial” meso-scale-climatic baseline (the “sea level”) that is representative across a broader spatial scale and optimized for consistency.

Weather stations are commonly available at airports, which are usually situated close to but outside of cities. However, weather stations at airports are often surrounded and influenced by extensive impervious areas, increased anthropogenic heat flux and overall large-scale urban-like developments, and may vary regionally depending on the national standard setup. [20] stresses that such stations must be carefully assessed before relying on their measurements. While these limitations are well known and documented, a single weather station at the local airport is still a commonly used rural reference, most often due to the lack of alternatives.

[44] already states that the commonly used two-station approach of a single urban and single rural station for UHI analysis is flawed due to a multitude of influencing local and mesoscale effects such as urban heat advection (UHA), which has proven to have a non-negligible effect [45–48]. Looking further into these effects, previous studies have explored the effect of the choice of a rural reference on the overall results [13,16,28,29]. [16] state that the average UHI intensity may range from 9.4 ºC to 12.9 ºC depending on the reference chosen. [13] report differences of at least 1.5 K, and [28,29] of around 1 K.

With recent advancements in the field of crowdsourcing, crowd weather stations (CWS) have become a commonly used data source for urban climate studies, overcoming the low spatial density of typical urban weather station networks [26,49,50]. CWS have also become an integral part of the emerging research field of urban climate informatics [51]. Large collections of CWS now enable urban climate studies without setting up an extensive measurement network, which is costly and difficult to maintain long-term. CWS were also already used to provide supplementary data for extensive high-end measurement campaigns [52], providing high-density air-temperature measurements. Studies focusing on land-surface temperature also utilized the large spatial coverage of CWS [53,54].

With these recent developments the question of a “correct” and widely available, truly “rural” reference remains and becomes especially challenging when trying to utilize CWS data in areas without any nearby professional weather station (PWS). [55] state that they had to reject available urban stations due to all rural reference stations being too far away from the study area, hence reducing the newly gained potential of crowdsourced weather station data.

Summarizing the common problems that may arise when selecting a (rural) reference for UHI studies:

  • no station may be available at all
  • the only available station is located at an airport where measurements suffer from the general urban-like characteristics of airports
  • a station may be available but is located too close to the city and hence is temporarily exposed to urban effects such as UHA
  • the available station may be too far away from the city, hence is no longer representative of the regional climate the city is located in
  • any single station only covers a specific point with unique micro- to local-climatic effects in an overall also heterogeneous rural surrounding of the city
  • based on the unique choice of a single station, results, specifically absolute values, cannot be compared between studies
  • there may not be any “rural” area at all, e.g., in large cities such as Tokyo (Japan) or the Ruhr-Area (Germany)

Multiple approaches exist to address some of these limitations. [56], e.g., introduced the concept of an “inverse UHI” where, instead of a rural station, a central urban station is chosen as the reference location. CWS can also be used as a rural reference. A single station may be selected that exhibits the coolest air-temperature conditions [26]. Alternatively, multiple CWS around the city together with LCZ information can be combined for a more robust estimate. Also, the UHI definition can be avoided entirely, only comparing intra-urban differences based on LCZs [12].

In this study, we explore if there is a suitable (rural) reference for utilizing CWS data that is (a) available globally, (b) fully covers the region of interest, (c) is consistent and allows comparison between cities and across studies.

We use six years of quality-controlled CWS data combined with LCZ information, observations from multiple professional weather stations as the best available benchmark, and ERA5-Land reanalysis data [57]. We investigate the effect of the selected rural reference and thus UHI calculation on the detected UHI intensity to identify an “ideal” rural reference for future UHI studies, not only with CWS data.

Materials and methods

Study areas and period

This study focuses on Paris (France) and Berlin (Germany) (Fig 1), chosen for their large networks of CWS and PWS [48]. After quality control (QC), 17,802 individual CWS in Paris, and 5,729 in Berlin were retained covering the six-year period between 2019-01-01 00:00 UTC and 2025-01-01 00:00 UTC. Extensive PWS coverage in both urban and rural areas allows us to compare city observations with multiple rural references. Both cities are located in Köppen Geiger zone Cfb, i.e., temperate oceanic climate [58], and are dominated by westerly winds [59–61].

thumbnail
Download:
Fig 1. Regions of interest and weather stations used as rural reference stations in Paris (a) and Berlin (b) and local climate zones (LCZs).

The region of interest (ROI) shows the maximum bounding box defined by the spatial extent of crowd weather stations (CWS). The grid and ERA5-Land pixel centroids indicate the spatial resolution of this dataset. All professional weather stations (PWS) with their corresponding LCZ used as rural references with the airport station highlighted are shown as squares. All CWS used as rural references are indicated as dots with their corresponding LCZ. The coordinates mark the SW and NE corners of the ROI, respectively. Projection: EPSG:3035 ETRS89-extended / LAEA Europe, background map: [62].

https://doi.org/10.1371/journal.pclm.0000955.g001

Datasets

Crowd weather station data.

The same dataset as in [48] was used for this study, only adding one additional year of data (2024). The dataset consists of Netatmo (https://netatmo.com/smart-weather-station) CWS data from weather stations consisting of identical sensors forming a uniform network [26]. The stations usually consist of an indoor module (private data) and an outdoor module (public data), with additional potential modules for wind and precipitation. Evaluations of the Netatmo outdoor sensors have shown that the manufacturer-provided accuracy of 0.3 K is correct and stable over time [50,63–65]. However, studies have shown a relatively slow sensor response with a time lag () ranging between = 12.7 minutes [65] and = 22.46 minutes [66]. In addition, the sensor placement might introduce further errors, which can be addressed by a QC, filtering out erroneous measurements, improving the overall dataset quality.

The underlying crowd database of this research [25] contains hourly mean values of air temperature, relative humidity, precipitation, wind, and atmospheric pressure from CWS located in over 500 cities worldwide. Data was continuously collected from 2019 onward via the company’s API, while at the same time keeping track and ensuring consistency in metadata of the stations, i.e., their position by (a) adding new stations, (b) updating the metadata of existing stations and restarting time series of stations if necessary, and (c) only downloading data when a station’s metadata was continuously observed (see also: [48]).

All data were quality controlled using the CrowdQC+ (https://github.com/dafenner/CrowdQCplus) R-package [25] with default settings. Here, we selected quality level o2 (> 80% daily data availability) and in addition added an annual data-availability requirement of at least 80% per station per year in at least one of the six years investigated, to avoid a seasonal sampling bias. Stations not fulfilling these criteria were excluded. Even after applying the QC and strict data availability criteria, a large number of CWS remains in the ROIs (S1 Appendix Fig A).

Professional weather station data.

One-hourly averaged 2 m air-temperature data from PWS were acquired to match the temporal resolution of CWS data and ERA5-Land. For the Paris region, data from 32 PWS operated by METEO France were acquired [67] (Fig 1). For Berlin, data from the Deutscher Wetterdienst (German Weather Service) (DWD) (10 stations) [68], the Technische Universität Berlin (TU) (11 stations) [69] and the Freie Universität Berlin (FU) (28 stations) [70–81] were combined, resulting in a total of 49 PWS. To fully cover the rural surroundings in Berlin in all directions, two PWS had to be selected that were slightly outside of the ROI defined by the CWS (Fig 1b). In this study, PWS are defined as institutionally operated stations (e.g., by national meteorological services or universities) that use high-quality sensors and follow standardized practices for siting, maintenance, and data processing.

A quality control consisting of a range check (), and a persistency check (240 minutes), was applied to the PWS air-temperature data, following procedures suggested by WMO [19], adapting absolute values. Values not fulfilling these criteria were excluded. Multiple PWS in Berlin were completely excluded due to quality concerns or incomplete time-series (S1 Appendix Fig D), reducing the number of PWS to 44 (Fig 1 and S1 Appendix Fig E).

ERA5-land data.

ERA5-Land data [57,82] were downloaded via Google Earth Engine [83]. For each city, the maximum extent of all available CWS forms the respective ROI. ERA5-Land pixels were then selected where the pixel intersected with the defined ROI. Data for the 2 m-air temperature (“t2m”) were downloaded for Paris and Berlin, using the native spatial resolution of at one-hourly temporal resolution. In addition, also a per-pixel mean altitude was derived from the Multi-Error-Removed Improved-Terrain Digital Elevation Model (MERITDEM) [84] to be later used for height-correcting of the air-temperature values.

Urban heat Island calculations

Four different datasets of rural reference air-temperature data were used to calculate spatial air-temperature differences () between urban locations and the rural environment, i.e., the UHI intensity, using . In the following subsections, the different datasets and respective methods to obtain are detailed further. While for Berlin height differences between weather stations are moderate and height correction could be neglected, for Paris, topographic differences are larger and height correction is inevitable (S1 Appendix Fig B). All individual station data (CWS, PWS) and ERA5-Land pixel data were initially height-corrected to the closest integer value of the median height of all available CWS in the ROIs (74 m amsl for Paris, 46 m amsl for Berlin S1 Appendix Fig B). An environmental lapse rate (ELR) of -0.0065 K m-1 was used for the correction. For each ERA5-Land pixel, the mean pixel height was derived from the MERITDEM (90 m resolution at the equator) [84].

Single professional airport reference station.

Most commonly in UHI studies, only one PWS is used as a rural reference station (). In many cities of the world, these stations are located at airports. Hence for benchmarking, a single PWS located at an airport was selected for Paris and Berlin, respectively, to consider this case.

For Berlin, Berlin-Brandenburg-Airport (formerly Schönefeld) was chosen as an airport reference station operated by DWD. The airport PWS is located approximately 17 km southeast of the city core (see: Fig 1). For Paris, Charles de Gaulle Airport was selected, which is approximately 22 km northeast of the city core and part of the METEO France network (see: Fig 1).

Multiple professional reference stations.

To account for air-temperature variability in rural areas (e.g., [16]) and to avoid urban effects on a single rural reference station, such as urban heat advection when located downwind [45,56], data from multiple PWS were selected. Multiple, evenly distributed PWS around the city are better suited to reflect the rural spatial heterogeneity than using only one or two stations. This will serve as the “best possible case” in this study. However, this dataset may still suffer from residual uncertainties such as land-use changes over time or varying levels of maintenance. To select the rural PWS from the full set of stations, all PWS located in “urban” LCZs (1–7 & 10) were excluded. LCZ information at the stations’ locations was extracted from the global LCZ map - version 2 [62]. Some stations in LCZ 8 (large lowrise) were added because they were located at airports outside the cities (e.g., in Paris). Other PWS were excluded when located either within a few hundred meters of water or still within the city (e.g., inside parks).

For Paris, 13 stations remained as professional rural references, for Berlin 12 stations. For both cities, the rural reference PWS are evenly distributed around the city (see: Fig 1). All air-temperature data from these stations were averaged per hour to a combined .

Crowd weather stations.

The UHI intensity was also calculated using solely CWS data, combined with the global LCZ map [62].

The CWS-based rural reference was defined as CWS located in any of LCZ 9 (sparsely built), B (scattered trees), C (brush, scrub), D (low plants), or F (bare soil or sand). Similar to [12,37], the predominant LCZ class for each station inside a circular buffer with 250 m radius was calculated. All LCZ map pixels intersecting with the circular buffer were selected. If there was a tie between two LCZ classes, no predominant class was assigned. A CWS was only selected when the fraction of the most common LCZ class was , and of the pixels in the circular buffer were valid (not missing or LCZ G (water)). Additionally, a weighted fraction was calculated similar to [37] and [12], using weights defined by [85] and [86], taking the similarity of different LCZs into account. The weighted fraction of the predominant LCZ class had to be . Even after applying this rigorous filtering, many of the selected stations will likely still be placed close to buildings (see section Crowd weather stations).

All CWS satisfying the above criteria were then averaged for each timestamp resulting in an hourly rural reference temperature for the entire ROI.

Spatially-aggregated ERA5-land.

Lastly, ERA5-Land, a reanalysis product [57] was used to characterize the rural background conditions since it does not consider the effects of cities. ERA5-Land uses the Carbon Hydrology-Tiled ECMWF Scheme for Surface Exchanges over Land (CHTESSEL) where the underlying land cover is based on the Global Land Cover Characteristics (GLCC) [87] dataset and the classification is performed using the Biosphere-Atmosphere Transfer Scheme (BATS) [88] which has a total of 20 classes, none of which represent urban structures. Further investigations on the dataset [89] revealed that most areas that according to the global map of LCZs are urban, are classified as BATS class “1” which corresponds to “Crops, Mixed Farming,” hence no urban information enters the dataset this way. [90] assessed if ERA5 [91] can reproduce the surface UHI effect; however, finding no evidence for this. All ERA5-Land pixels were averaged on a per-hour basis for each ROI, forming a rural reference air temperature .

Results

Data availability

The different rural references vary in data availability. relies on a single station located at an airport. In Paris and Berlin, both stations exhibit short periods of data outages that cannot be compensated for by nearby stations, resulting in gaps in the dataset. Nonetheless, the data availability is high with 99.7% (Berlin) and 99.9% (Paris) across the six years. For this specific case, limitations are negligible. For other PWS, especially in Berlin, the issue of data availability is more substantial (see: S1 Appendix Fig D).

In this regard, using is more robust. However, the subset of stations, from which the reference air temperature is calculated, varies. New stations were added to the network in Paris over the course of the study period (S1 Appendix Fig D). Data outages are rare and usually short. In Berlin, most stations show frequent outages from a few days to several years (S1 Appendix Fig D). Multiple stations also stopped recording during the study period. For both cities the combination of all stations still provides full temporal data coverage, allowing the calculation of at all times.

shows a dynamic data availability (S1 Appendix Fig C). Transmission outages, in combination with data removed by QC-procedures, especially during daylight hours when radiation errors are the most common [50], cause data gaps for individual CWS. This produces a varying subset of CWS with a distinct diurnal cycle of data availability (S1 Appendix Fig C). This dynamic set of stations still serves as a robustly calculated combined rural reference, as individual stations have less weight in the large sample of CWS (typically ∼50 and ∼90 rural CWS for Berlin and Paris, respectively, cf. S1 Appendix Fig C). Due to the large sample, is also without any temporal gaps.

Lastly, is similarly continuously available without any gaps.

Comparison of rural references

All four rural reference air-temperature datasets () were compared against each other (Fig 2, S1 Appendix Fig K). The comparison of to shows an overall positive deviation of +0.6 K towards for Paris. With +0.1 K, this is less pronounced for Berlin. The root mean square deviation (RMSD) (S1 Appendix Equation A) of 1.35 K (Paris) shows the highest deviation among all compared references. A similar positive deviation (+0.5 K) is also visible when comparing to . For Berlin, this is again less pronounced with (+0.3 K). Comparing to shows a lower RMSD of 1.15 K and with -0.07 K no distinct mean deviation. This is similar for Berlin. versus shows a higher RMSD of 1.28 K (1.23 K for Berlin) and a positive mean deviation of +0.4 K (different for Berlin with -0.12 K) towards . compared to shows the lowest RMSD (1.04 K), highest correlation coefficient ( = 0.99) and the highest R2 (0.98) amongst all comparisons, indicating an excellent fit. For Berlin, this differs, showing a slightly higher RMSD of 1.26 K but a similar R2 of 0.98 (S1 Appendix Fig K). and were not compared due to being a subset of .

thumbnail
Download:
Fig 2. Inter-comparison of rural references for the greater Paris region.

is based on intersecting ERA5-Land pixels, uses crowd weather stations (CWS) located in homogeneous natural local climate zones (LCZs), relies on multiple professional weather stations (PWS) in rural LCZs evenly distributed around the city, and uses a single airport station. The red line (1:1) shows the ideal fit. N indicates the hours covered, R2 is the coefficient of determination of the linear regression, MAD the mean absolute deviation (S1 Appendix Equation B), RMSD the root-mean-square deviation (S1 Appendix Equation A), and spearman’s rank correlation coefficient. The color bar indicates the number of observations per grid cell of the 2-dimensional histogram. For Paris, and agree best among all comparisons with the lowest bias, lowest RMSD, and highest R2.

https://doi.org/10.1371/journal.pclm.0000955.g002

Comparison of

First, the resulting differences in (UHI intensities) calculated from CWS measurements were compared for all times in 2019–2024 (Fig 3). and show almost identical median values for Paris (Fig 3a). For Berlin (Fig 3b), shows slightly higher values (median, 75th percentile) compared to . For both cities shows the lowest values overall (median, interquartile range (IQR)). This is similar for , however less pronounced. Performing the same for just PWS (using as a rural reference and all available urban PWS (S1 Appendix Fig E, S1 Appendix Fig D), it shows a much lower range compared to CWS.

thumbnail
Download:
Fig 3. over the entire study period (2019–2024) in Paris (a) and Berlin (b) using different rural references.

The rural reference is based on intersecting ERA5-Land pixels, uses crowd weather stations (CWS) located in homogeneous natural local climate zones (LCZs), relies on multiple professional weather stations (PWS) evenly distributed around the city, and, uses a single airport station. The box with hatching (PWS) contains only PWS for urban and rural measurements and hence relies on as a rural reference. Boxes range from 25th to 75th percentile (IQR), whiskers extend to maximum and minimum values. Above each box, the number of observations is given. represents median (line).

https://doi.org/10.1371/journal.pclm.0000955.g003

We further analyzed the differences in for the different rural references along the diurnal cycle (Paris: Fig 4, Berlin: S1 Appendix Fig L). For Paris (Fig 4), compared to shows lower values during nighttime and higher values during daytime hours. Compared to , the diurnal cycle is slightly dampened. shows a strongly dampened diurnal cycle with only marginally increased values during nighttime and an overall low IQR. Using also shows overall lower values compared to and a dampened diurnal cycle, ranging in between and . Using only professional stations for urban and rural sites (PWS), the diurnal cycle of the UHI is similarly dampened but with an overall lower IQR. and show the highest IQR for both cities. For Berlin, shows lower values during the night and higher values in the afternoon compared to (S1 Appendix Fig L).

thumbnail
Download:
Fig 4. Same as Fig 3 but as diurnal cycle (data grouped per hour) of in Paris.

https://doi.org/10.1371/journal.pclm.0000955.g004

Lastly, the calculated values of were analyzed along the annual cycle (Fig 5, S1 Appendix Fig M). shows almost no annual cycle for Paris (slightly more pronounced in Berlin (S1 Appendix Fig M)), while and show the strongest annual cycle with being the highest in spring and summer months (March to August) and lower in winter and fall (September to February). again shows lower values but still a stronger annual cycle than . again ranges in between with a clear annual cycle visible but overall lower values of . For Berlin, shows a dampened annual cycle compared to PWS (S1 Appendix Fig M).

thumbnail
Download:
Fig 5. Same as Fig 3 but as annual cycle (data grouped per month) of in Paris.

https://doi.org/10.1371/journal.pclm.0000955.g005

In general, shows the highest values and the strongest diurnal and annual cycles, and performs best for Paris in comparison to , which is considered the best possible rural reference in terms of quality of measurement and covering spatial variability. For Berlin (S1 Appendix Fig L, S1 Appendix Fig M), shows a dampened diurnal cycle compared to . This is especially pronounced during the months of September, October, and November (S1 Appendix Fig M).

Discussion

The choice of a representative, robust rural reference to calculate spatial air-temperature differences between a city and its surroundings is critical for any UHI study and its choice can enable or limit the comparability of results to other studies [13,16,21,92].

Crowd weather stations

[26] state that “[...] it is important for consistency that the reference site is also a Netatmo station rather than a standard site.” In general, using only a single reference is problematic as it may be temporarily influenced by other conditions such as heat advection [45]. Selecting multiple rural reference stations can overcome this limitation, making it robust against site-specific influences and characteristics and more representative of the entire region’s meso climate. With CWS also being available in the rural surroundings of a city, solely relying on them would create a simple and consistent approach. However, it comes with a certain limitation. Varying data availability (see: Data availability) makes this reference already inconsistent between individual timestamps, albeit with an overall suitable number of rural CWS available (S1 Appendix Fig C). Across the current, over 2000-day observing period (as of 11/2025), stations in the crowd database were active for an average of 896 days (∼2.5 years). Less than 1% of the stations cover the entire period in a consistent way. Our results indicate that using multiple rural CWS as a combined rural reference () has major shortcomings. Calculated values of show a strongly dampened diurnal and annual variation (see Fig 4, Fig 5, S1 Appendix Fig L, S1 Appendix Fig M), which is uncharacteristic for mid-latitude cities like Paris and Berlin [93,94]. While during the day shortwave radiation errors due to imperfect radiation shielding and placement are likely to be the most dominant source of error, the nocturnal dampening is most likely caused by CWS being close to buildings and hence being influenced by emitted heat from the buildings (long-wave radiation, anthropogenic heat), even though they are located in rural areas. While filtering for LCZ-homogeneity ( of selected natural LCZ in a 250 m buffer), the observed effects may also be caused by the remaining LCZ-heterogeneity within the buffer. In addition, systematic biases due to an unequal distribution of CWS as, e.g., described by [95], may also contribute to the reported effects. All mentioned factors negatively impact the use of CWS as a rural reference. We argue that CWS in rural areas are hence not providing a “true” rural air-temperature signal, resulting in overall lower values of and underestimating the UHI intensity.

In some cases a well-chosen single CWS, as used by [26], may not suffer from strong influences by buildings. However, it will have similar limitations as a single PWS. The station may be temporarily influenced by the close-by city and only representative of a small area. In addition, due to the crowdsourced nature of the CWS data, exact site-specific information remains unknown. This is particularly true for Netatmo CWS, while some other CWS networks provide further meta data if uploaded by users (e.g., Weather Observations Website (WOW), [96]). Yet, these other networks have a much lower CWS density compared to the Netatmo network.

Professional weather stations

In contrast to , PWS are less likely to suffer from heat emitted by close-by buildings since they are typically set up farther away from buildings, following stricter guidelines [19]. However, as mentioned, single reference stations may lack robustness against local-scale effects such as heat advection [38], which can lead to long-term biases of the time series. [45], e.g., report a bias due to UHA of 0.1 K for the city of De Bilt, NL [45].

Using a multitude of rural PWS would be the best case. Yet, it is highly uncommon for cities to have a large number of rural reference stations evenly distributed around the city, forming an ideal [97]. Paris or Berlin can provide this and hence were used here as test beds. Previously, few studies were able to utilize multiple PWS surrounding the city. For example, [43] used two stations for Berlin, [39] used six for Paris, ensuring similar altitudes and even distributions. Further, [18] used six rural stations for the Nagano Basin (Japan), however, mentioning that they were unable to select stations in all directions. [40] used four stations for Regina, CA, [17] used three rural stations for Seoul, KR, and finally [42] was able to select six stations surrounding Oklahoma City, US. The majority of UHI studies, however, had to rely on a single station with all its known shortcomings [44].

Data availability

Another aspect worth considering when choosing a suitable rural reference, especially considering CWS, is the general data availability and applicability in different regions worldwide. While PWS may be the best possible rural reference, their availability and data access is typically limited. Although a city may have a sufficient number of CWS, it may lack even a single PWS, let alone multiple evenly distributed PWS around the city. In addition, focusing on megacities or megacity-like developments such as Tokyo (Japan) or the Ruhr-Area (Germany), that have a large number of CWS covering the intra-urban variability in great detail, simply have no rural areas that are not influenced by urban effects. Suitable sites only exist far away and hence are not representative of the local to regional climate conditions [19]. Other options need to be explored to be able to more consistently perform UHI studies based on CWS measurements.

ERA5-Land

shows similar results in comparison with (Fig 2, S1 Appendix Fig K) and is, in contrast to , available worldwide. Since ERA5-Land does not include any urban information, values can be extracted at the exact location of the city, making this also useful for mega-cities where setting up uninfluenced rural reference stations in the same meso-scale climate is impossible. follows a similar approach to , aiming at providing an artificial meso-scale-climatic baseline forming the “sea level,” minimizing an overrepresentation of local-scale influences. To further assess the global applicability of , additional cities in different climate zones should be considered for further analyses. However, this is always constrained by data availability of all references.

Often and in contrast to PWS, data access for ERA5-Land is easy via the Copernicus Data Store (CDS) (https://cds.climate.copernicus.eu/datasets/reanalysis-era5-land?tab=download) [82] or the Google Earth Engine data catalog (https://developers.google.com/earth-engine/datasets/catalog/ECMWF_ERA5_LAND_HOURLY) [83]. One downside is, however, that it is not available in near real-time but only with a delay of typically 3–7 days and as an intermediate product (ERA5-Land-T) [98]. , on the other hand, has instant, near-real time data availability via the public application programming interface (API).

Covering the spatial variability of the UHI

A comparison with only using PWS (as a rural reference and as intra-urban measurements) was compiled to showcase UHI estimates solely based on PWS (Fig 3). For all stations exact metadata are available, the measurement devices have generally high accuracies, are set up to minimize influences from buildings, and are well maintained. Our results indicate a low IQR of PWS, especially for Paris (Fig 4, Fig 5). For Berlin it is higher, but compared to the CWS data set again lower (Fig 4, S1 Appendix Fig L, Fig 5, S1 Appendix Fig M). The reason might be the low number of available PWS for Paris (only 8 urban stations (S1 Appendix Fig E)), and that the PWS do not fully cover the intra-urban local-scale settings and resulting air-temperature variability, whereas CWS potentially do. The PWS dataset for Berlin is more robust with 23 urban stations and the spatial variability is better captured, but still likely not sufficient, as certain areas of the city are not covered by PWS. In any case, comparing city-wide UHI estimates derived from CWS and PWS is only fair if a similar subset of local-scale areas, located in similar meso-scale settings of a city, are compared, as CWS also suffer from spatial biases [95]. Networks such as the Birmingham, UK Urban Climate Laboratory Network [99] and the Leuven.cool network (2019–2025) in Leuven, Belgium [100], both now partially discontinued, as well as recently completed networks in Freiburg, Germany (2022-today) [101], Bristol, UK (2024-today) [102] with 39 PWS, and the large Dortmund network (2025-today) with 76 PWS [103], may represent upcoming options suited for fair and detailed comparisons.

With 360 rural reference CWS for Paris and 151 rural reference CWS for Berlin, a large spatial coverage can be achieved (Fig 1). The robustness against mesoscale effects and its representativeness for the area is large, but influences by buildings and highly variable data availability limit its overall applicability.

Recommendations and limitations

Varying results for Paris and Berlin indicate some city-specific characteristics of the UHI that warrant further investigation in future studies. Potential reasons for deviations include differences in urban boundary-layer characteristics, regional-scale differences in synoptic forcing (e.g., [48] S7) and background climate (Berlin is located in the transition zone between oceanic and continental climate). Furthermore, our approach was applied only to two mid-latitude cities in temperate climates. The performance of ERA5-Land as a rural reference should also be tested in other climate zones (e.g., arid, tropical, cold), in regions with strong terrain-forced flows, and in coastal cities where sea breezes may dominate the regional signal. Our approach with using a fixed ELR might introduce large deviations in regions with stronger topographical features. An option for improvement would be to compute a daily ELR, similarly done in the downscaling of ERA5 to ERA5-Land, following [57,104]. As ERA5-Land is a single-level product at the surface, this cannot be directly derived from the data used here and other data sources would be needed (e.g., coarser-resolution ERA5). Such a daily calculation of a daily ELR would be most beneficial in regions with large differences in elevation. For mostly flat areas, however, this can be neglected. Both approaches will likely neglect temporary near-surface inversions [104].

The reported metrics in Fig 2 (top row, 2nd panel) for are relative to . For Paris, the RMSD of 1.04 K indicates an uncertainty when using instead of and the bias of -0.102 K indicates a systematic deviation. However, while may be considered the best available rural reference, it cannot be considered perfect, hence those metrics may only be used for assessing similarity between references instead of absolute correctness. Considering the reported bias and also Fig 3, -based values of tend to be slightly (0.1 K) higher than -based values of . Comparing the deviations (RMSD, bias, IQR, amplitude) to the expected range of (Fig 3), the annual and diurnal variation (Fig 4, Fig 5), deviations in between using and can be, in the long-term, considered low and spatial and temporal patterns to be very similar. Based on the results for Paris and Berlin, ERA5-Land is able to capture the spatial heterogeneity of the rural surroundings, providing most similar results compared to using the multitude of PWS (). However, it should be noted that the resulting computed rural air temperature, which is based on pixels, represents an artificial regional rural air-temperature average.

We suggest that ERA5-Land may be used as a rural reference for UHI studies, making them consistent and inter-comparable, and creating additional potential for the use of CWS globally. This combination of crowdsourced CWS data with reanalysis data can enable detailed and comparable urban climate studies in cities without any professional reference stations.

Conclusion

This study examines how the selection of rural reference air-temperature data influences UHI studies based on CWS. Data from multiple rural PWS would provide the most reliable and representative rural reference. However, as they are often unavailable globally, we explored alternatives. Specifically, we investigated using a single airport PWS, multiple CWS in rural LCZs, and ERA5-Land reanalysis data to represent rural reference air temperature, in comparison to using multiple PWS. Analyses were carried out for the cities of Paris and Berlin during 2019–2024.

The analysis reveals a strong dampening effect of when using rural CWS data, compared to the ideal case of using multiple rural PWS, as rural CWS suffer from urban-like effects due to their typical placement close to buildings. Using a single PWS indicates insufficient coverage of the spatial rural air-temperature variability. The comparison of ERA5-Land with multiple rural PWS shows most similar results, both in terms of absolute rural air temperature as well as derived values of (UHI intensity).

Using ERA5-Land as a rural reference allows for better comparisons between UHI studies from different cities and also enables urban climate analyses in regions that, besides CWS, have no measurement stations. With a large number of Netatmo CWS and other CWS networks available globally, a new spatial and temporal scale of (urban) climate studies is possible. For example, in March 2024, 285,000 active Netatmo stations for air temperature, humidity, and atmospheric pressure, 110,000 of them also equipped with a precipitation sensor and 57,000 also with a wind sensor were available. Yet, to utilize the global characteristic of the dataset, especially for UHI studies, a suitable consistent rural reference is critical.

This study only examined two mid-latitude, temperate oceanic climate cities that provide sufficient data for the analyses carried out. Extending the analyses to cities in different climates could further strengthen the applicability. Nonetheless, dense networks of CWS in cities across the globe will now potentially be able to be exploited with regard to urban heat analyses and comparison with surrounding rural areas. However, sensitivity to the choice of rural reference should, if possible, always be tested, reported, and discussed with regard to local climate and topographic characteristics.

We conclude and propose, with a focus on fully utilizing the potential of CWS globally, the usage of ERA5-Land reanalysis data as an independent, ubiquitously available, easy-to-access, and near-real-time dataset.

Supporting information

S1 Appendix. Table A UHI studies and their respective (rural) reference used.

A curated list of studies following [2].

Fig A Spatial density of CWS. The number of stations available for Paris (a) and Berlin (b) in the study period after applying all data availability criteria (see: Sec Crowd weather station data in the main article). Note the different color range.

Fig B Distribution of CWS altitudes. The altitudes were derived from the Multi-Error-Removed Improved-Terrain Digital Elevation Model (MERITDEM) [84] based on a station’s geographic positions. The red line indicates the median (), the green line the mean (). Note the different axes range since there are more stations in Paris (a) than in Berlin (b).

Fig C Availability of CWS serving as rural reference (RefCWS). Data across the entire study period of 2019–2024. The left panel shows the daily number of available stations across the years, the right panel the diurnal cycle of mean data availability per year for Paris (a) and Berlin (b). Note different y axis range.

Fig D Daily data availability over the 6-year study period (2019–2024) for professional weather stations. The color bar indicates the percentage of available data for that day. Stations used in this study are shown in their respective LCZ color. Stations that were excluded. are marked as gray with an asterisk indicating from which subset the station was excluded.

Fig E Professional urban weather stations. The station “Montsouris” in Paris consists of two sensors close-by (“double”), however, appears as one marker in this figure. Projection: EPSG:3035 ETRS89-extended / LAEA Europe, background map: [62].

Fig F Comparison of PWS measurements vs. ERA5-Land data for Berlin, DE. Note the per-city colorbar ranges.

Fig G Comparison of PWS measurements vs. ERA5-Land data for Berlin, DE. Note the per-city colorbar ranges. Continuation of S1 Appendix Fig F.

Fig H Comparison of PWS measurements vs. ERA5-Land data for Berlin, DE. Note the per-city colorbar ranges. Continuation of S1 Appendix Fig G.

Fig I Comparison of PWS measurements vs. ERA5-Land data for Paris, FR. Note the per-city colorbar ranges.

Fig J Comparison of PWS measurements vs. ERA5-Land data for Paris, FR. Note the per-city colorbar ranges. Continuation of S1 Appendix Fig I.

Fig K Inter-comparison of rural references for the greater Berlin region. RefERA5 is based on intersecting ERA5-Land pixels, RefCWS uses crowd weather stations (CWS) located in homogeneous natural local climate zones (LCZs), RefmultiPWS relies on multiple professional weather stations (PWS) in rural LCZs evenly distributed around the city, and, RefsinglePWS uses a single airport station. The red line (1:1) shows the ideal fit. N indicates the hours covered, R2 is the coefficient of determination of the linear regression, MAD the mean absolute deviation (S1 Appendix Equation B), RMSD the root-mean-square deviation (S1 Appendix Equation A), and ρ Spearman’s rank correlation coefficient. The color bar indicates the number of observations per grid cell of the 2-dimensional histogram.

Fig L ∆Ta over the entire study period (2019–2024) in Berlin using different rural references. The rural reference RefERA5 is based on intersecting ERA5-Land pixels, RefCWS uses crowd weather stations (CWS) located in homogeneous natural local climate zones (LCZs), RefmultiPWS relies on multiple professional weather stations (PWS) evenly distributed around the city, and, RefsinglePWS uses a single airport station. The box with hatching (PWS) contains only PWS for both urban and rural measurements and hence relies on RefmultiPWS as a rural reference. Boxes range from 25th to 75th percentile (IQR), whiskers extend to maximum and minimum values. Same as Fig 4 but for Berlin.

Fig M Same as S1 Appendix Fig L but as annual cycle (data grouped per month) of ∆Ta in Berlin. Same as Fig 5 but for Berlin.

Equation A Root mean square deviation (RMSD).

Equation B The mean absolute deviation (MAD).

https://doi.org/10.1371/journal.pclm.0000955.s001

(PDF)

Acknowledgments

The authors thank all owners of Netatmo CWS for sharing their data publicly and Netatmo for making the data available via an API. The authors thank METEO-France for providing PWS data from over 30 stations in Paris and the Freie Universität Berlin and Technische Universität Berlin for also providing the data from their stations, as well as the DWD for providing additional PWS data. Thanks to all World Urban Database and Access Portal Tools (WUDAPT) [105] contributors who provided LCZ training labels to develop a global map of Local Climate Zones. Lastly, the authors would like to thank all open-source software creators, maintainers, and contributors, without claiming to be exhaustive and without any order, namely:

contextily

CPython

CrowdQC+ [25]

geopandas [106]

matplotlib [107]

numpy [108]

pandas [109]

PostgreSQL

PostGIS

pre-commit

psycopg2

pytest [110]

scikit-learn [111]

seaborn [112]

Timescale ...and many more

References

  1. 1. Arnfield AJ. Two decades of urban climate research: a review of turbulence, exchanges of energy and water, and the urban heat island. Int J Climatol. 2003;23(1):1–26.
  2. 2. Stewart ID. A systematic review and scientific critique of methodology in modern urban heat island literature. Int J Climatol. 2011;31(2):200–17.
  3. 3. Stewart ID. Why should urban heat island researchers study history? Urban Clim. 2019;30:100484.
  4. 4. Peppler A. Das Auto als Hilfsmittel der Meteorologischen Forschung. Das Wetter. 1929;46:305–8.
  5. 5. Oke T. The energetic basis of the urban heat island. Q J R Meteorol Soc. 1982;108(455):1–24.
  6. 6. Oke TR, Mills G, Christen A, Voogt JA. Urban climates. Cambridge University Press; 2017. https://doi.org/10.1017/9781139016476
    • 7. Stewart ID, Oke TR. Local Climate Zones for Urban Temperature Studies. Bull Am Meteorol Soc. 2012;93(12):1879–900.
    • 8. Stewart ID, Oke TR, Krayenhoff ES. Evaluation of the ‘local climate zone’ scheme using temperature observations and model simulations. Int J Climatol. 2014;34(4):1062–1080.
    • 9. Bechtel B, Demuzere M, Mills G, Zhan W, Sismanidis P, Small C, et al. SUHI analysis using local climate zones—A comparison of 50 cities. Urban Clim. 2019;28:100451.
    • 10. Brousse O, Georganos S, Demuzere M, Dujardin S, Lennert M, Linard C. Can we use local climate zones for predicting malaria prevalence across sub-Saharan African cities?. Environ Res Lett. 2020;15(12):124051.
    • 11. Demuzere M, Bechtel B, Mills G. Global transferability of local climate zone models. Urban Clim. 2019;27:46–63.
    • 12. Fenner D, Meier F, Bechtel B, Otto M, Scherer D. Intra and inter ‘local climate zone’ variability of air temperature as observed by crowdsourced citizen weather stations in Berlin, Germany. Meteorol Z. 2017;525–47.
    • 13. Roth M, Sanchez B, Li R, Velasco E. Spatial and temporal characteristics of near‐surface air temperature across local climate zones in a tropical city. Intl J Climatol. 2022;42(16):9730–52.
    • 14. Demuzere M, Kittner J, Bechtel B. LCZ Generator: A Web Application to Create Local Climate Zone Maps. Front Environ Sci. 2021;9.
    • 15. Runnalls KE, Oke TR. Dynamics and controls of the near-surface heat island of Vancouver, British Columbia. Phys Geogr. 2000;21(4):283–304.
    • 16. Hawkins TW, Brazel AJ, Stefanov WL, Bigler W, Saffell EM. The Role of Rural Variability in Urban Heat Island Determination for Phoenix, Arizona. J Appl Meteor. 2004;43(3):476–86.
    • 17. Kim Y-H, Baik J-J. Spatial and Temporal Structure of the Urban Heat Island in Seoul. J Appl Meteorol. 2005;44(5):591–605.
    • 18. Sakakibara Y, Matsui E. Relation between Heat Island Intensity and City Size Indices/Urban Canopy Characteristics in Settlements of Nagano Basin, Japan. Geogr Rev JPN Chirigaku Hyoron. 2005;78(12):812–24.
    • 19. World Meteorological Organization (WMO). Guide on the Global Observing System. No. 488 in Publication. Geneva, Switzerland: World Meteorological Organization; 2010.
      • 20. World Meteorological Organization (WMO). Guidance on Measuring, Modelling and Monitoring the Canopy Layer Urban Heat Island (CL-UHI). Geneva; 2023. 1292 p.
        • 21. Martin-Vide J, Sarricolea P, Moreno-García MC. On the Definition of Urban Heat Island Intensity: The “Rural” Reference. Front Earth Sci. 2015;3.
        • 22. de Vos LW, Leijnse H, Overeem A, Uijlenhoet R. Quality Control for Crowdsourced Personal Weather Stations to Enable Operational Rainfall Monitoring. Geophys Res Lett. 2019;46(15):8820–9.
        • 23. de Vos LW, Droste AM, Zander MJ, Overeem A, Leijnse H, Heusinkveld BG, et al. Hydrometeorological Monitoring Using Opportunistic Sensing Networks in the Amsterdam Metropolitan Area. Bull Am Meteorol Soc. 2020;101(2):E167–85.
        • 24. Jacobs A, Top S, Vergauwen T, Suomi J, Käyhkö J, Caluwaerts S. Filling gaps in urban temperature observations by debiasing ERA5 reanalysis data. Urban Clim. 2024;58:102226.
        • 25. Fenner D, Bechtel B, Demuzere M, Kittner J, Meier F. CrowdQC+ — A Quality-Control for Crowdsourced Air-Temperature Observations Enabling World-Wide Urban Climate Applications. Front Environ Sci. 2021;9.
        • 26. Chapman L, Bell C, Bell S. Can the crowdsourcing data paradigm take atmospheric science to a new level? A case study of the urban heat island of London quantified using Netatmo weather stations. Intl J Climatol. 2017;37(9):3597–3605.
        • 27. Bechtel B, Alexander PJ, Beck C, Böhner J, Brousse O, Ching J. Generating WUDAPT Level 0 Data – Current Status of Production and Evaluation. Urban Clim. 2019;27:24–45.
        • 28. Taesler R. Studies of the development and thermal structure of the urban boundary layer in Uppsala. Part I: Experimental program. Uppsala: Meteorological Institution, Uppsala University; 1980. 63 p.
          • 29. Taesler R. Studies of the development and thermal structure of the urban boundary layer in Uppsala. Part II: Data analysis and results. Uppsala: Meteorological Institution, Uppsala University; 1980. 61 p.
            • 30. Bowling S, Benson C. Study of the Subarctic Heat Island at Fairbanks, Alaska. Washington, D.C.: U.S. Environmental Protection Agency; 1978.
              • 31. Kłysik K, Fortuniak K. Temporal and spatial characteristics of the urban heat island of Łódź, Poland. Atmosph Environ. 1999;33(24–25):3885–95.
              • 32. Fast JD, Torcolini JC, Redman R. Pseudovertical Temperature Profiles and the Urban Heat Island Measured by a Temperature Datalogger Network in Phoenix, Arizona. J Appl Meteorol. 2005;44(1):3–13.
              • 33. Yow DM, Carbone GJ. The Urban Heat Island and Local Temperature Variations in Orlando, Florida. Southeastern Geographer. 2006;46(2):297–321.
              • 34. Emmanuel R, Johansson E. Influence of urban morphology and sea breeze on hot humid microclimate: The case of Colombo, Sri Lanka. Clim Res. 2006;30:189–200.
              • 35. Chow WTL, Roth M. Temporal dynamics of the urban heat island of Singapore. Int J Climatol. 2006;26(15):2243–60.
              • 36. Fenner D, Holtmann A, Krug A, Scherer D. Heat Waves in Berlin and Potsdam, Germany – Long-term Trends and Comparison of Heat Wave Definitions from 1893 to 2017. Int J Climatol. 2019;39(4):2422–37.
              • 37. Varentsov M, Fenner D, Meier F, Samsonov T, Demuzere M. Quantifying Local and Mesoscale Drivers of the Urban Heat Island of Moscow with Reference and Crowdsourced Observations. Front Environ Sci. 2021;9.
              • 38. Bassett R, Cai X, Chapman L, Heaviside C, Thornes JE. The effects of heat advection on UK weather and climate observations in the vicinity of small urbanized areas. Boundary-Lay Meteorol. 2017;165(1):181–96.
              • 39. Haeffelin M, Ribaud JF, Céspedes J, Dupont JC, Lemonsu A, Masson V. Impact of Boundary Layer Stability on Urban Park Cooling Effect Intensity. Atmosph Chemist Phys. 2024;24(24):14101–22.
              • 40. Stewart ID. Influence of meteorological conditions on the intensity and form of the urban heat island effect in Regina. Canadian Geograph. 2000;44(3):271–85.
              • 41. Schatz J, Kucharik CJ. Seasonality of the Urban Heat Island Effect in Madison, Wisconsin. J Appl Meteorol Climatol. 2014;53(10):2371–86.
              • 42. Basara JB, Hall PK Jr, Schroeder AJ, Illston BG, Nemunaitis KL. Diurnal cycle of the Oklahoma City urban heat island. J Geophys Res. 2008;113(D20).
              • 43. Fenner D, Holtmann A, Meier F, Langer I, Scherer D. Contrasting changes of urban heat island intensity during hot weather episodes. Environ Res Lett. 2019;14(12):124013.
              • 44. Lowry WP. Empirical Estimation of Urban Effects on Climate: A Problem Analysis. J Appl Meteor. 1977;16(2):129–35.
              • 45. Brandsma T, Können GP, Wessels HRA. Empirical estimation of the effect of urban heat advection on the temperature series of De Bilt (The Netherlands). Intl J Climatol. 2003;23(7):829–45.
              • 46. Heaviside C, Cai X.‐M., Vardoulakis S. The effects of horizontal advection on the urban heat island in Birmingham and the West Midlands, United Kingdom during a heatwave. Quart J Royal Meteoro Soc. 2015;141(689):1429–41.
              • 47. Brousse O, Simpson C, Walker N, Fenner D, Meier F, Taylor J, et al. Evidence of horizontal urban heat advection in London using six years of data from a citizen weather station network. Environ Res Lett. 2022;17(4):044041. pmid:37600746
              • 48. Kittner J, Fenner D, Demuzere M, Bechtel B. Analysis of nocturnal urban heat advection using crowd weather stations. Quart J R Meteoro Soc. 2025;151(773):e5065.
              • 49. Muller CL, Chapman L, Johnston S, Kidd C, Illingworth S, Foody G, et al. Crowdsourcing for climate and atmospheric sciences: current status and future potential. Intl J Climatol. 2015;35(11):3185–203.
              • 50. Meier F, Fenner D, Grassmann T, Otto M, Scherer D. Crowdsourcing air temperature from citizen weather stations for urban climate research. Urban Clim. 2017;19:170–91.
              • 51. Middel A, Nazarian N, Demuzere M, Bechtel B. Urban Climate Informatics: An Emerging Research Field. Front Environ Sci. 2022.
              • 52. Fenner D, Christen A, Grimmond S, Meier F, Morrison W, Zeeman M. Urbisphere-Berlin Campaign: Investigating Multiscale Urban Impacts on the Atmospheric Boundary Layer. Bull Am Meteorol Soc. 2024;105(10):E1929-61.
              • 53. Naserikia M, Nazarian N, Hart MA, Sismanidis P, Kittner J, Bechtel B. Multi-city analysis of satellite surface temperature compared to crowdsourced air temperature. Environ Res Lett. 2024;19(12):124063.
              • 54. Krelaus L, Apfel J, Bechtel B, Sismanidis P. Surface and Canopy-Layer Urban Heat Island Intensities in Europe – Quantifying Differences in the Diurnal Cycle for Three Summer Periods. In: 2023 Joint Urban Remote Sensing Event (JURSE). 2023. p. 1–4.
                • 55. Wolters D, Brandsma T. Estimating the Urban Heat Island in Residential Areas in the Netherlands Using Observations by Weather Amateurs. J Appl Meteorol Climatol. 2012;51(4):711–21.
                • 56. Bassett R, Cai X, Chapman L, Heaviside C, Thornes JE, Muller CL, et al. Observations of urban heat island advection from a high‐density monitoring network. Quart J Royal Meteoro Soc. 2016;142(699):2434–41.
                • 57. Muñoz-Sabater J, Dutra E, Agustí-Panareda A, Albergel C, Arduini G, Balsamo G, et al. ERA5-Land: a state-of-the-art global reanalysis dataset for land applications. Earth Syst Sci Data. 2021;13(9):4349–83.
                • 58. Kottek M, Grieser J, Beck C, Rudolf B, Rubel F. World Map of the Köppen-Geiger climate classification updated. metz. 2006;15(3):259–63.
                • 59. Gerstengarbe FW, Werner PC. Katalog der Grosswetterlagen Europas (1881-2004). Nach Paul Hess und Helmu Brezowsky. Potsdam: Potsdam-Institut für Klimafolgenforschung e.V.; 2005.
                  • 60. Bissolli P, Dittmann E. The objective weather type classification of the German Weather Service and its possibilities of application to environmental and meteorological investigations. metz. 2001;10(4):253–60.
                  • 61. Deutscher Wetterdienst (DWD). Objective Weather Type Classification Data. 2025. Available from: https://www.dwd.de/EN/ourservices/wetterlagenklassifikation/online_wlkdaten.txt?view=nasPublication&nn=495490
                    • 62. Demuzere M, Kittner J, Martilli A, Mills G, Moede C, Stewart ID, et al. A global map of local climate zones to support earth system modelling and urban-scale environmental science. Earth Syst Sci Data. 2022;14(8):3835–73.
                    • 63. Varentsov M, Konstantinov PI, Shartova NV, Samsonov TE, Kargashin PE, Varentsov AI, et al. Urban Heat Island of the Moscow Megacity: The Long-Term Trends and New Approaches for Monitoring and Research Based on Crowdsourcing Data. IOP Conference Series: Earth and Environmental Science. 2020;606(1):012063.
                    • 64. Mandement M, Caumont O. Contribution of personal weather stations to the observation of deep-convection features near the ground. Nat Hazards Earth Syst Sci. 2020;20(1):299–322.
                    • 65. Coney J, Pickering B, Dufton D, Lukach M, Brooks B, Neely RR III. How useful are crowdsourced air temperature observations? An assessment of Netatmo stations and quality control schemes over the United Kingdom. Meteorol Appl. 2022;29(3):e2075.
                    • 66. Büchau YG. Modelling shielded temperature sensors. An assessment of the Netatmo citizen weather station. [Master’s thesis]. Hamburg: University of Hamburg; 2018.
                      • 67. METEO-France. API Données d’observation. 2025.
                        • 68. Deutscher Wetterdienst (DWD) Climate Data Center (CDC). Stündliche Stationsmessungen der Lufttemperatur in 2 m Höhe in °C für Deutschland, Version V24.03. 2025.
                          • 69. Holtmann A, Meier F, Scherer D. [UC]2 LTO Measurements. Technische Universität Berlin, Fachgebiet Klimatologie; 2025.
                            • 70. Langer I. Mikroklimatische Messungen im Berliner Stadtgebiet vom Institut für Meteorologie der Freien Universität Berlin des Jahres 2019. Freie Universität Berlin; 2023. https://doi.org/10.17169/REFUBIUM-41443
                              • 71. Langer I. Mikroklimatische Messungen im Berliner Stadtgebiet vom Institut für Meteorologie der Freien Universität Berlin des Jahres 2020. Freie Universität Berlin; 2024. https://doi.org/10.17169/REFUBIUM-41456
                                • 72. Langer I. Mikroklimatische Messungen im Berliner Stadtgebiet vom Institut für Meteorologie der Freien Universität Berlin des Jahres 2021. Freie Universität Berlin; 2024. https://doi.org/10.17169/REFUBIUM-41993
                                  • 73. Langer I. Mikroklimatische Messungen im Berliner Stadtgebiet vom Institut für Meteorologie der Freien Universität Berlin des Jahres 2022. Freie Universität Berlin; 2024. https://doi.org/10.17169/REFUBIUM-42179
                                    • 74. Langer I. Mikroklimatische Messungen im Berliner Stadtgebiet vom Institut für Meteorologie der Freien Universität Berlin des Jahres 2023. Freie Universität Berlin; 2024. https://doi.org/10.17169/REFUBIUM-42854
                                      • 75. Langer I. Mikroklimatische Messungen im Berliner Stadtgebiet vom Institut für Meteorologie der Freien Universität Berlin des Jahres 2024. Freie Universität Berlin; 2025. https://doi.org/10.17169/REFUBIUM-46513
                                        • 76. Kirchner I, Gebauer P, Langer I, Sorge I, Bergmann T, Undeutsch R. Meteorologische Messungen im Berliner Stadtgebiet mit MEVIS Stationen des Instituts für Meteorologie der Freien Universität Berlin für das Jahr 2019. Freie Universität Berlin; 2023. https://doi.org/10.17169/REFUBIUM-37944
                                          • 77. Kirchner I, Gebauer P, Langer I, Sorge I, Bergmann T, Undeutsch R. Meteorologische Messungen im Berliner Stadtgebiet mit MEVIS Stationen des Instituts für Meteorologie der Freien Universität Berlin für das Jahr 2020. Freie Universität Berlin; 2023. https://doi.org/10.17169/REFUBIUM-37935
                                            • 78. Kirchner I, Gebauer P, Langer I, Sorge I, Bergmann T, Reinmar U. Meteorologische Messungen im Berliner Stadtgebiet mit MEVIS Stationen des Instituts für Meteorologie der Freien Universität Berlin für das Jahr 2021. Freie Universität Berlin; 2023. https://doi.org/10.17169/REFUBIUM-37858
                                              • 79. Kirchner I, Gebauer P, Langer I, Sorge I, Bergmann T, Undeutsch R. Meteorologische Messungen im Berliner Stadtgebiet mit MEVIS Stationen des Instituts für Meteorologie der Freien Universität Berlin für das Jahr 2022. Freie Universität Berlin; 2023. https://doi.org/10.17169/REFUBIUM-38662
                                                • 80. Kirchner I, Gebauer P, Langer I, Sorge I, Bergmann T, Undeutsch R. Meteorologische Messungen im Berliner Stadtgebiet mit MEVIS Stationen des Instituts für Meteorologie der Freien Universität Berlin für das Jahr 2023. Freie Universität Berlin; 2024. https://doi.org/10.17169/REFUBIUM-42016
                                                  • 81. Kirchner I, Gebauer P, Langer I, Sorge I, Bergmann T, Hoffmann R. Meteorologische Messungen im Berliner Stadtgebiet mit MEVIS Stationen des Instituts für Meteorologie der Freien Universität Berlin für das Jahr 2024. Freie Universität Berlin; 2025. https://doi.org/10.17169/REFUBIUM-48243
                                                    • 82. Copernicus Climate Change Service. ERA5-Land Hourly Data from 2001 to Present. ECMWF; 2019. https://doi.org/10.24381/CDS.E2161BAC
                                                      • 83. Gorelick N, Hancher M, Dixon M, Ilyushchenko S, Thau D, Moore R. Google Earth Engine: planetary-scale geospatial analysis for everyone. Remote Sens Environ. 2017;202:18–27.
                                                      • 84. Yamazaki D, Ikeshima D, Tawatari R, Yamaguchi T, O’Loughlin F, Neal JC. A high-accuracy map of global terrain elevations. Geophys Res Lett. 2017;44(11):5844–53.
                                                      • 85. Bechtel B, Demuzere M, Sismanidis P, Fenner D, Brousse O, Beck C, et al. Quality of Crowdsourced Data on Urban Morphology—The Human Influence Experiment (HUMINEX). Urban Sci. 2017;1(2):15.
                                                      • 86. Bechtel B, Demuzere M, Stewart ID. A weighted accuracy measure for land cover mapping: Comment on Johnson et al. local climate zone (LCZ) map accuracy assessments should account for land cover physical characteristics that affect the local thermal environment. Remote Sens. 2020;12(11):1769.
                                                      • 87. Loveland TR, Reed BC, Brown JF, Ohlen DO, Zhu Z, Yang L, et al. Development of a global land cover characteristics database and IGBP DISCover from 1 km AVHRR data. Int J Remote Sens. 2000;21(6–7):1303–30.
                                                      • 88. European Centre for Medium-Range Weather Forecasts (ECMWF). IFS DOCUMENTATION – Cy45r1. PART IV: PHYSICAL PROCESSES; 2018. Available from: https://www.ecmwf.int/sites/default/files/elibrary/2018/80895-ifs-documentation-cy45r1-part-iv-physical-processes_1.pdf
                                                        • 89. Earth Resources Observation And Science (EROS) Center. Global Land Cover Characterization. U.S. Geological Survey; 2017. https://doi.org/10.5066/F7GB230D
                                                          • 90. Nogueira M, Hurduc A, Ermida S, Lima DCA, Soares PMM, Johannsen F, et al. Assessment of the Paris urban heat island in ERA5 and offline SURFEX-TEB (v8.1) simulations using the METEOSAT land surface temperature product. Geosci Model Dev. 2022;15(14):5949–65.
                                                          • 91. Hersbach H, Bell B, Berrisford P, Hirahara S, Horányi A, Muñoz-Sabater J, et al. The ERA5 Global Reanalysis. Q J R Meteorol Soc. 2020;146(730):1999–2049.
                                                          • 92. Stewart ID, Mills G. The Urban Heat Island: A Guidebook. Elsevier; 2021.
                                                            • 93. Wouters H, De Ridder K, Demuzere M, Lauwaet D, Van Lipzig NPM. The diurnal evolution of the urban heat island of Paris: A model-based case study during summer 2006. Atmos Chem Phys. 2013;13(17):8525–41.
                                                            • 94. Fenner D, Meier F, Scherer D, Polze A. Spatial and Temporal Air Temperature Variability in Berlin, Germany, during the Years 2001–2010. Urban Clim. 2014;10:308–31.
                                                            • 95. Brousse O, Simpson CH, Poorthuis A, Heaviside C. Unequal distributions of crowdsourced weather data in England and Wales. Nat Commun. 2024;15(1):4828. pmid:38902290
                                                            • 96. Kirk PJ, Clark MR, Creed E. Weather Observations Website. Weather. 2021;76(2):47–9.
                                                            • 97. Muller CL, Chapman L, Grimmond CSB, Young DT, Cai X. Sensors and the city: a review of urban meteorological networks. Int J Climatol. 2013;33(7):1585–600.
                                                            • 98. European Centre for Medium-Range Weather Forecasts (ECMWF). ERA5-Land: Data Documentation. 2026. Available from: https://confluence.ecmwf.int/display/CKB/ERA5-Land%3A+data+documentation#ERA5Land:datadocumentation-Dataupdatefrequency
                                                              • 99. Warren EL, Young DT, Chapman L, Muller C, Grimmond CSB, Cai XM. The Birmingham Urban Climate Laboratory—A High Density, Urban Meteorological Dataset, from 2012–2014. Sci Data. 2016;3(1):160038.
                                                              • 100. Beele E, Reyniers M, Aerts R, Somers B. Quality control and correction method for air temperature data from a citizen science weather station network in Leuven, Belgium. Earth Syst Sci Data. 2022;14(10):4681–717.
                                                              • 101. Plein M, Kersten F, Zeeman M, Christen A. Street-Level Weather Station Network in Freiburg, Germany: Station Documentation. Zenodo; 2024. https://doi.org/10.5281/ZENODO.12732551
                                                                • 102. Ludwig S, Plein M, Barlow JF, Escobar-Ruiz V, Grimmond S, Matthews J, et al. Street-Level Weather Station Network in Bristol, UK: Station Documentation. University of Freiburg; 2025. https://doi.org/10.5281/ZENODO.15537605
                                                                  • 103. Hüser C, Weickhmann L, Kittner J, Reinhart V, Sismanidis P, Bechtel B. Data2Resilience – Real-time Biometeorological Observations for Thermal Comfort Assessment at City Scale. 2025. https://doi.org/10.5194/icuc12-1013
                                                                    • 104. Dutra E, Muñoz-Sabater J, Boussetta S, Komori T, Hirahara S, Balsamo G. Environmental Lapse Rate for High-Resolution Land Surface Downscaling: An Application to ERA5. Earth Space Sci. 2020;7(5):e2019EA000984.
                                                                    • 105. Ching J, Mills G, Bechtel B, See L, Feddema J, Wang X, et al. WUDAPT: An Urban Weather, Climate, and Environmental Modeling Infrastructure for the Anthropocene. Bull Am Meteorol Soc. 2018;99(9):1907–24.
                                                                    • 106. Jordahl K, Bossche JVD, Fleischmann M, Wasserman J, McBride J, Gerard J. Geopandas/Geopandas: V0.8.1. Zenodo; 2020. https://doi.org/10.5281/ZENODO.3946761
                                                                      • 107. Hunter JD. Matplotlib: A 2D Graphics Environment. Comput Sci Eng. 2007;9(3):90–5.
                                                                      • 108. Harris CR, Millman KJ, van der Walt SJ, Gommers R, Virtanen P, Cournapeau D, et al. Array programming with NumPy. Nature. 2020;585(7825):357–62. pmid:32939066
                                                                      • 109. The Pandas Development Team. Pandas-Dev/Pandas: Pandas. Zenodo; 2023. https://doi.org/10.5281/ZENODO.3509134
                                                                        • 110. Krekel H, Oliveira B, Pfannschmidt R, Bruynooghe F, Lauger B, Bruhin F. Pytest 7.2. 2004.
                                                                          • 111. Pedregosa F, Varoquaux G, Gramfort A, Michel V, Thirion B, Grisel O. Scikit-learn: Machine Learning in Python. J Mach Learn Res. 2011;12(85):2825–30. http://jmlr.org/papers/v12/pedregosa11a.html
                                                                          • 112. Waskom M. seaborn: statistical data visualization. JOSS. 2021;6(60):3021.