Nature-Based Preschool Gardens for Thermal Condition Regulation: Implications for Climate-Sensitive Policy and Planning

Poznań has a population of approximately 545,000 inhabitants and an area of 262 km², and is one of the largest cities in Poland (Statistics Poland 2022). The city is located in the temperate climate zone and is characterised by average temperature over multiple years (1991–2020) of 9.4°C and average precipitation over multiple years (1991–2020) of 538 mm. However, recent data show an increase in the average annual temperature coupled with a decline in annual precipitation levels (IMWM 2023).

As with many other cities, Poznań faces the problem of climate change, which is expressed through rising temperature, more frequently occurring heat waves and tropical nights (Półrolniczak et al. 2018), among other factors, and those phenomena are predicted to increase by 2050 (UCAP 2019).

In Poznań, there are 263 preschools attended by 20,297 children (Department of Education of Poznań City Hall 2019), of which 87% (230) have their own outdoor space that plays the role of a PG (Fig. 1). However, they vary in terms of the share and composition of green spaces. According to the national regulations, preschools operate yearround; however, during summer holidays, they may operate on a limited basis. Still each child is guaranteed care at their home preschool during the summer duty period.

Fig. 1.

Location of PGs in Poznań based on data from the Department of Education of Poznań City Hall (2019) and the orthophoto of Poznań (SISP 2021). PGs - preschool gardens

We analysed thermal conditions of all PGs (N = 230) against the background of the city’s land surface temperature (LST). To provide a local-scale perspective, we further examined three detailed case studies: preschools No. 42, No. 87 and No. 115, located in one of the most densely built-up areas of Poznań.

The first step was to identify and map preschools with their own outdoor area – a PG (Fig. 2). Next, maps of average LST (ALST), tree canopy cover (TCC) and a distribution of the normalised difference vegetation index (NDVI) were prepared to recognise thermal conditions of PGs and their relationship to the existing vegetation.

Fig. 2.

Study framework.

ALST – average land surface temperature; NDVI – normalised difference vegetation index; TIRS – thermal infrared sensor; T_air – air temperature; TCC – tree canopy cover; PG – preschool garden; LULC –c land use and land cover; SIP – spatial information system; PGCE – preschool garden cooling effect; GIS – geographic information system

Then, based on the ALST map, the air temperatures (T_air) were modelled, showing the thermal condition of PGs for children. Finally, PGCE was assessed, taking into account differences in ALST between each PG and its closest neighbourhood within a 45 m ring buffer, and the thermal profiles were analysed. We calculated average building heights in ring buffers of PGs using the BDOT10k (2022) topographic database as a secondary factor for analysing PGCE. The methods applied in the following steps are presented in the next sections.

The thermal conditions of PGs were recognised by analysing the distribution of LST obtained from satellite imaging, which is a common approach in urban climate studies (Voogt, Oke 2003, Tomlinson et al. 2011, Zhan et al. 2013, Zhou et al. 2019), as well as in studies assessing the thermal impact of green spaces (Cao et al. 2010, Estoque et al. 2017, Jang et al. 2024).

We utilised the Landsat Surface Temperature product and selected bands of Landsat 8 Operational Land Imager (OLI)/Thermal Infrared Sensor (TIRS) scenes from Collection 2 Level 2 Science Products (EROS 2024). The data were processed according to Cook et al. (2014) and the thematic guide (LSDS 2020). Using Earth Explorer, we searched for images that captured the city’s warmest weather conditions, ensuring that cloud cover was < 3%. Our selection focused solely on images free of cloud interference within the study area, allowing for accurate and clear observations. As a result, we included the thermal infrared bands ST_B10 from five multispectral satellite images that met the mentioned criteria. They were captured on the following dates: 7 June 2018, 3 June 2019, 26 June 2019, 23 September 2019 and 8 August 2020. The images were acquired during clear weather conditions, characterised by a cloud-free sky and full sun exposure over the entire city, at the time of 9:449:51 GMT (11:44–11:51 CET). These images depict sunny weather typical of late spring, summer and early autumn – referred to as the warm season – when high temperatures may negatively impact the thermal comfort of Poznań residents.

To calculate LST, raw values of each pixel of the surface temperature bands were first multiplied using a scale factor and then enlarged by including additive scale factor (ASF) [Eq. (1), see LSDS (2020)]. In the next step we converted the LST values from Kelvins to Celsius degrees [Eq. (2)]. 1LSTK=(DN×MSF)+ASF{\rm{LS}}{{\rm{T}}_{\rm{K}}} = ({\rm{DN}} \times {\rm{MSF}}) + {\rm{ASF}} 2LSTC=LSTK−273.15{\rm{LS}}{{\rm{T}}_{\rm{C}}} = {\rm{LS}}{{\rm{T}}_{\rm{K}}} - 273.15 where:

• –

  LST_k - LST in Kelvins (K),

• –

  DN - digital number (value of each pixel of surface temperature band),

• –

  MSF - multiplicative scale factor equal to 0.00341802,

• –

  ASF - additive scale factor equal to 149,

• –

  LST_C - LST in Celsius degrees (°C).

As proposed by Majkowska et al. (2017), we created a raster map of the ALST_C from five images using the raster calculator in ArcMap [Eq. (3)]. 3ALST=∑i=1nLSTCn{\rm{ALST}} = {{\sum\nolimits_{i = 1}^n {{\rm{LS}}{{\rm{T}}_{\rm{C}}}} } \over n} where:

• –

  ALST - average LST (°C),

• –

  LST_C - LST in Celsius degrees (°C) for a given pixel,

• –

  n - number of scenes (pixels).

In this study, we included the greenness level by employing the NDVI, which is commonly used in studies related to surface temperature (Walawender 2009). With NDVI, we were able to group PGs according to the greenness levels of vegetation (Lo, Quattrochi 2003, Feyisa et al. 2014). We calculated NDVI as proposed by Lillesand et al. (2004) based on Landsat bands 4 [red band (RED)] and 5 [near infrared band (NIR)] [see Eq. (4)]. For each of the PGs, the mean NDVI was calculated using the zonal statistics tool [Eq. (5)]. 4NDVI=(NIR−RED)(NIR+RED){\rm{NDVI}} = {{({\rm{NIR}} - {\rm{RED}})} \over {({\rm{NIR}} + {\rm{RED}})}} 5ANDVI=∑i=1nNDVIn{\rm{ANDVI}} = {{\sum\nolimits_{i = 1}^n {{\rm{NDVI}}} } \over n} where:

• –

  NDVI - normalised difference vegetation index,

• –

  NIR - near infrared band, Landsat 8 band 5,

• –

  RED - red band, Landsat 8 band 4,

• –

  ANDVI - average NDVI of a given polygon,

• –

  n - number of NDVI pixel values intersected by a given polygon.

For PGs and their buffers, we calculated TCC as the percentage share of canopy cover (%). It was based on the tree canopy layer (SISP 2018), which provides information on the canopy coverage at a 0.5 m spatial resolution in the year 2018. We also analysed how building height in PG buffers (measured by the number of floors) affects thermal conditions in PGs. We identified a number of floors using the vector Topographic Objects Database (BDOT10k 2022).

The relationships between selected variables and ALST and PGCE were assessed using the Pearson correlation coefficient.

While analysing the thermal conditions in PGs, we also considered the air temperature distribution, estimated using the LST data and the non-linear regression model developed for Poznań by Majkowska et al. (2017) [see Eq. (6)]. The authors of the model used thermal data from TM Landsat 5 satellite images, along with meteorological data that represent air temperature at 2 m above the ground in various areas of the city of Poznań, obtained from in situ measurements with HOBO sensors. The model is notable for its high coefficient of determination (r² = 0.84). 6Tair =b0×(ALSTc)b1{T_{{\rm{air }}}} = {b_0} \times {\left( {{\rm{ALS}}{{\rm{T}}_{\rm{c}}}} \right)^{{b_1}}} where:

• –

  T_air - estimated air temperature at a height of 2 m above ground level (°C),

• –

  b₀, b₁ - model coefficients: b₀ = 0.633 and b₁ = 1.035,

• –

  ALST_C - average LST (°C).

We compared the estimated air temperature values in PGs to the number of preschoolers from the city register, taking into account thresholds from the Atlas of Poland Climate (Tomczyk, Bednorz 2022), in which warm days (25.1–30.0°C) and hot days (> 30.0°C) are distinguished.

The PGCE was defined as the mean LST difference between the PG and its buffer as proposed by Cao et al. (2010) and Chibuike et al. (2018), who measured a park’s cooling island intensity. A similar approach is commonly used in studies on the intensity of surface urban heat islands measured as LST differences between urban and surrounding reference areas (Zhou et al. 2019). We calculated PGCE using Eq. (7) based on Chibuike et al. (2018). Negative values of PGCE indicate a cooling effect (ALST_pg < ALST_b) and positive ones a warming effect of given PGs (ALST_pg > ALST_b). 7PGCE=ALSTpg−ALSTb{\rm{PGCE}} = {\rm{ALS}}{{\rm{T}}_{{\rm{pg}}}} - {\rm{ALS}}{{\rm{T}}_{\rm{b}}} where:

• –

  ALST_b - average LST (°C) calculated for a given buffer of PG,

• –

  ALST_pg - average LST (°C) calculated for a given PG.

We analysed PGCE within a 45 m distance, which results from the spatial resolution and geometric parameters of Landsat 8 TIRS that collects data with a 100 m spatial resolution resampled to 30 m (USGS 2019). Storey et al. (2014) have shown that the geometric accuracy varies from 7.4 m in the case of TIRS Band-to-Band Registration Accuracy (requirement value < 18 m) to 32.7 m in the case of TIRS Absolute Geodetic Accuracy (requirement value < 76 m). Taking the above into account, we applied buffers, the size of which exceeded the spatial resolution (pixel size) and geometric performance values of the input data (Zhan et al. 2013). A similar justification was presented by Chibuike et al. (2018).

However, it is necessary to stress that there is no standard approach in delimitation buffer zones thus far. Other studies that quantified greenspace cooling effects applied various buffer parametrisations, e.g., from 30 m intervals from the edges of parks to a maximum distance of 420 m (Feyisa et al. 2014), 100 m intervals to a maximum distance of 200 m (Łowicki, Kuklińska 2025) and 100 m around each case study greenspace (Shih 2017).

For the three selected case study preschools, we investigated the PGCE using the thermal profiles (stack profiles) showing the temperature change within a distance interval (e.g. Walawender 2009, Majkowska et al. 2017). We used an ALST raster and a sampling interval of 30 m according to the map resolution.

Climate change results in the increased number of hot days and heatwaves, making it harder to maintain comfort in cities. This affects also preschools, where young children (ages 3–6 years) play outside in warm season. Yet, climate policies or space design often overlook children’s needs (Pegram, Colon 2020).

This study has shown that PGs located in a diverse urban context vary in terms of size, NDVI and TCC, which translates into different thermal conditions illustrated by ALSTs ranging from ≈33°C to > 39°C.

The air temperature estimated based on ALST shows that 94.5% (18,073) of preschoolers can experience temperatures > 25°C to 30°C, while 0.2% (44) might be exposed to air temperatures > 30°C that characterise hot days. According to Półrol·niczak et al. (2018), hot days occurred in Poznań on average 18 times per year (within the period of 1966–2015), with a maximum of 37 days in 2006. The frequency of this phenomenon is increasing (Wibig 2021).

The need to deal with high temperatures affects preschools and school yards in various geographical contexts. The review conducted by Antoniadis et al. (2020) revealed three factors contributing to heat issues, namely land cover materials with high heat capacities and/or heat conductivities, like asphalt or concrete, a lack of shade and proximity to other hot areas.

Bäcklin et al. (2021) have revealed that an important factor that shapes local climate conditions is shade provided mainly by trees. Their results indicate that PGs with fewer trees are more exposed to heat stress than those with many trees. Zhang et al. (2017) have also shown that planting trees is the most effective solution for improving school outdoor thermal comfort. Shading recreational areas for children can also lower the risk of heat stress during hot weather (Vanos et al. 2017) and reduce the risk of sunburn and skin cancer through reducing ultraviolet radiation exposure (Gage et al. 2019). Lanza et al. (2021) have confirmed that more children interact with trees during periods of high heat index than during periods of moderate heat index, which highlights the crucial role of trees in cooling the play space during the hottest periods.

The green spaces in PGs can provide not only a better thermal environment for pupils using those spaces but also thermal impacts beyond the preschool’s borders that can contribute to living conditions of local communities. However, ameliorating the impact of climate heat on children’s health is not common practice in designing play spaces in an urban context (Vanos 2015). Not surprisingly, schoolyards are often ‘heat islands’ with higher temperatures than the surrounding streets (Moogk-Soulis 2002, 2010).

The results of this study have shown that PGs could be cooler or warmer than the surroundings. Similarly, there was strong variation in thermal conditions of PGs (Bäcklin et al. 2021). Although the average cooling potential of PGs in the neighbourhood was rather low, the thermal profiles of the selected case studies illustrated that such an impact can be much higher, up to 2.5°C. When placed within the broader range of 1–7°C reported for urban green spaces, this demonstrates a meaningful cooling potential (Soltanifard, Amani-Beni 2025).

The potential of GI for cooling the neighbourhood is related to vegetation type. As our results have shown, the PGs with higher TCC were characterised by cooler thermal conditions. A more specific study conducted by Lehnert et al. (2021) showed that the cooling impact of trees on the heat stress index was detected up to 10.5°C in Universal Thermal Climate Index (UTCI), while low vegetation was able to reduce the UTCI by not more than 2.3°C.

The thermal conditions of preschools are also impacted by the surrounding urban context. As simulations conducted by Zhang et al. (2017) indicated, the temperature of the playground area might increase by approximately 2.0°C at 2 p.m. under a high-rise neighbourhood scenario.

The ability to regulate local climate conditions by small-scale solutions such as PGs in densely built-up areas can contribute to (thermal) comfort within PGs and in neighbouring public spaces. However, the strength of thermal regulation is highly related to local complex conditions and land-use management. The results of this study are consistent with a review showing that cooling intensity is shaped by vegetation type, spatial configuration and urban morphology (Soltanifard, Amani-Beni 2025).

Shaping comfortable thermal conditions in urban settings is among the various challenges that face Poznań and many other cities where climate change effects along with urban development can lead to unfavourable heat (UCAP 2019).

This study provides evidence of usefulness in identifying thermal ‘hotspots’ as the priority areas for interventions towards better thermal conditions. The hotspot areas are characterised by a high density of built-up structures and a high share of artificial surfaces. Such spaces have limited opportunities for implementing new large-scale green elements. Still, there is room for greening existing open spaces and introducing new small-scale NBSs. As Oke et al. (2017) emphasised, increasing vegetation cover and incorporating natural landscape features into already built-up areas are the best means of managing urban climate effects at all scales. This approach has recently been applied in the context of educational facilities, where transformation initiatives are being implemented (coolschools.eu).

This study has shown that NDVI can be an accessible measure for monitoring green areas and a supporting measure for indicating places for interventions towards better thermal conditions. NDVI of ≥0.3 showed lower ALST and can be used as a threshold for planning for better thermal conditions and delimiting the areas that have the potential to mitigate urban heat. The given NDVI threshold has been used in other studies to identify GI and its effect on temperature (Arellano, Roca 2022). In Barcelona, Spain and its metropolitan area, NDVI > 0.4 and a minimum size threshold of 0.5 ha (Barcelona Regional 2019) have been used for delimiting green areas that play the role of outdoor climate shelters (‘refugis climàtics’) (Baró et al. 2021). According to Barcelona Regional (2019), a climate shelter is considered an indoor or outdoor accessible space that during extreme weather episodes provides the population with thermal comfort, rest and safety. With this definition in mind, Polish cities can enhance their adaptation strategies by adopting the successful practices applied in Barcelona. This can be achieved as part of urban climate adaptation plans by identifying potential climate shelters, particularly through an assessment of institutional greenery based on our research findings.

This study has shown that PGs that are present in various city locations, including dense urban areas, are valuable reservoirs of open spaces. However, PGs, similar to schoolyards, are often designed in a traditional way with large asphalt surfaces and limited green spaces (Danks 2010). Planning, designing and retrofitting PGs and/or schoolyards can transform their space into local cooling spots. Such an approach was applied in Barcelona, where selected schoolyards are used as climate shelters (Baró et al. 2021). By introducing blue, green and grey technical solutions, schoolyards support the city’s adaptation to climate change and ameliorate the exposure of the school population to urban heat. The transformed schoolyards play the role of climate shelters for pupils and local residents as they stay publicly open (Cartalis 2020). Similarly, the project OASIS developed in Paris (Baró et al. 2021) focuses on transforming pilot school playgrounds into cool islands using NBSs and a co-design approach (Sitzoglou 2020).

Incorporating bioclimatic urban design principles into reshaping public spaces could contribute to more comfortable environments for urban communities; in particular, increasing focus on bioclimatic design can be translated across urban spaces (Cortesão et al. 2016).

The use of NDVI and LST enabled the assessment of vegetation and thermal conditions across 230 PGs in Poznań without the need for labour-intensive in situ measurements. Increasingly accurate satellite imagery, such as Landsat 8/9, provides standardised, spatially explicit and fully open data that are useful for urban climate research. Despite some criticisms regarding its coarse resolution, NDVI remains a widely applied green space metric (Huang et al. 2021), including also in studies involving children (Kabisch et al. 2017). However, the spatial resolution of the Landsat bands utilised (30 m) can be a limitation, particularly for smaller research areas. In this study, the larger size of PGs relative to pixel size helped mitigate uncertainty. While direct field-based measurements (i.e. using microclimate sensors) and remote sensing based on unmanned aerial vehicles (UAVs) offer higher precision for site-scale studies, they are costly and face challenges related to spatial range (e.g. as using a UAV to conduct measurements simultaneously across many distant locations is limited). Bridging the gap between fine-scale assessments and broadly available data remains crucial for effective climate-sensitive urban planning.