How Cool Are

allotment Gardens?

A study on air temperature differences between allotment gardens and other urban areas in Berlin.


By Anahita Bidjanbeg, Victoria Liste, Lea Matscheroth, Annemarie Rost  & Corinna Seidel


April 2019




The climatic properties of urban green spaces, especially parks, have been studied regarding their ability to mitigate the Urban Heat Island effect (UHI). However, allotment gardens, prominent type of green space within the city of Berlin, have not yet been studied substantially in terms of climatic conditions, e.g. air temperature. In the present study, the nocturnal air temperature of 13 allotment garden clusters (AGCs) were analysed and compared to densely built-up urban areas and two different urban parks in Berlin in the summer of 2018. All AGCs were on average around 3 K cooler at night than the built-up areas. Ten of the 13 AGCs were on average 1.5 K cooler than both urban parks. Furthermore, the influence of the parameters size, shape complexity and the degree of built-up area in the surroundings of each AGC was analysed. The temperature in the allotments rose as shape became more complex and building density increased. Size had no significant influence in this analysis. It is assumed that further parameters such as the vegetation structure and the frequent irrigation at each garden patch could have had an influence on the lower nocturnal air temperatures of AGCs compared to the sampled parks and densely built-up area in Berlin.





1. Introduction

2. Methods       

      2. 1 Study site   

      2.2 Data Collection         

      2.3 Data processing       

3. Results          

     3.1 Temperature differences    

     3.2 Influence of parameters      

4. Discussion    

     Differences to urban references             


     Shape complexity           


     Other influencing factors and limitations             

5. Conclusion   

6. Publication bibliography       

Appendix I         

Appendix II       






Allotment garden


Allotment garden cluster


Individual Allotment garden cluster


Urban built-up area reference


Urban park reference with dense trees (“Tiergarten”)


Urban park reference with low plants (“Tempelhofer Field”)


Floor Space Index

Nocturnal AGC air temperature

Nocturnal URB air temperature

Nocturnal TIER  air temperature

Nocturnal TEMP air temperature

Air temperature difference of TURB and  TC


General linear model

Mean nocturnal air temperature of an AGC



  1. Introduction

Urban areas show specific climatic properties due to the lower albedo, higher thermal conductivity and higher heat capacities of building materials, causing different absorption and reflection of solar radiation (Bowler et al. 2010). Additionally, reduced convective cooling and lower evaporation rates (Gunawardena et al. 2017) lead to a warmer microclimate of urban space in comparison to its surroundings, an effect widely described as the urban heat island (UHI) (Bowler et al. 2010; Qiu et al. 2013). The effect is especially prominent at night (Grimm et al. 2008; Tan et al. 2007), when stored heat is emitted by urban surfaces, called the nocturnal urban heat island (Spronken-Smith and Oke 1999).

Rising temperatures due to climate change and the trend of urbanisation will lead to more and more people affected by the UHI and its negative effects on human wellbeing and health (Lee et al. 2017). Possible mitigation and adaptation strategies for the negative effects of UHI have been studied broadly. In particular, the impact of green spaces has been investigated by many in recent decades (Emmanuel and Loconsole 2015; Chang et al. 2007). Green spaces can function as “cool islands”. The term “Park Cool Island” (PCI) was coined to describe the condition in which the air temperature of the park is lower than the air temperature of the surrounding built-up area (Spronken-Smith and Oke 1998). The moisture, aerodynamic and thermal properties of vegetation differ from urban materials (Oke et al. 1989; Givoni 1991). Further, impermeable urban materials do not retain water for evaporation and absorb or store heat (Bowler et al. 2010). Vegetation can mitigate the UHI in three ways: evapotranspiration, shading of incoming solar radiation and altering air movement, leading to heat exchange (Greene and Kedron 2018; Zardo et al. 2017). However, the cooling capacities of green spaces are dependent on those ecosystem functions (Oke 1988; Larondelle and Haase 2013; Zardo et al. 2017), which in turn are based on many factors, e.g. the park type, shape, size and the surrounding land use (Spronken-Smith and Oke 1998). The local conditions require microscale analyses of e.g. the surrounding buildings, direction of streets, shape of the open space (Bowler et al. 2010), roughness, vegetation cover and many more small-scale factors.

A way to describe the shape of a green area is the edge-to-area ratio. Research shows that areas with a low edge-to-area ratio are cooler than these with a high ratio (Ren et al. 2013). If there is a high edge-to-area ratio the temperature rises, due to the strong surface heterogeneity in urban environments (Hagishima et al. 2007; Meier and Scherer 2012). The size of green areas can be influential as well. Larger green areas are cooler than smaller ones. However, this correlation is not linear (Bowler et al. 2010; Cao et al. 2010). Furthermore, impervious surroundings lead to lower temperature variability, whereas natural surroundings lead to higher fluctuations. Green areas with a less densely-built up vicinity are cooler (Egerer et al. 2019). Though difficult and complex, research on small-scale processes offer valuable insights that are especially of interest for applications in landscape planning and urban planning, as well as in adaptation to climate change (Zölch et al. 2016; Bartesaghi Koc et al. 2017).

Allotment gardens (AGs) are a type of green space that has not been investigated as much as parks with regards to their climatic properties and possible cooling potential (Cabral et al. 2017a; Cabral et al. 2017b). Most studies of AGs have focused on other ecosystem services inherent to them, e.g. social benefits (Cabral and Weiland 2016; Breuste and Artmann 2015). Little information is available on the microclimatic conditions of AGs and even less regarding which properties of AGs are most beneficial for cooling. AGs have been an important green area type in central and northern Europe historically and are still expanding in southern Europe (Cabral et al. 2017a). The Federal Allotment Gardens Law (Bundeskleingartengesetz - BKleingG) defines AGs as gardens which are used for non-commercial horticulture, in particular to produce horticultural products for their own use, and for recreation. They are located in complexes in which several individual gardens with communal facilities are grouped together (§1 BKleingG, revised 9/19/2006). AGs have very distinct properties that distinguish them different from parks, e.g. greater plant diversity (Speak et al. 2015). It is mainly regulations that determine the characteristics of AGs, e.g. tall trees are prohibited (BKleingG, revised 9/19/2006). The lack of tall trees in AGs results in quicker heat emission at night in comparison to parks with tall trees. The emission of stored heat can be inhibited by the shielding of the tree canopies (Bowler et al. 2010). Typically, AGs are also frequently irrigated, which could lead to different cooling properties, since irrigated vegetation has lower surface temperatures than water-stressed vegetation (Norton et al. 2015) and insufficient water availability causes plant transpiration reduction (Leuzinger et al. 2010; Shashua-Bar et al. 2011).

However, the number of AGs is declining steadily because of real estate pressure (Cabral et al. 2017a). For example, in Berlin, from 2012 to 2018, the number of AG plots decreased by 3% (Senatsverwaltung für Umwelt, Verkehr und Klimaschutz 2018), another 1% will be built on and a further 7% on private land are very likely to be removed (Fahrun 2018; Fröhlich 2018).

Therefore, this study will address the following research questions: (1) whether AGs are cooler than urban parks and densely built-up areas of Berlin, and (2) and whether the size, shape complexity and degree of built-up area surrounding the AG is influencing the air temperature of the AGs. Following previous research, the expected outcome is a lower air temperature for AGs compared to urban parks and densely built-up areas in Berlin (Hypothesis 1). Larger AGs with less complex shapes and lower densities of built-up area in the surroundings are expected to show the lowest air temperatures (Hypothesis 2).

  1. Methods

2. 1 Study site

Berlin, the capital and largest city of Germany by both area (89,170 ha) and population (3,723,914 inhabitants) (Amt für Statistik Berlin Brandenburg 10/2/2018), is characterised by high latitude westerlies and is influenced by a subcontinental climate with cold winters and warm summers (Endlicher and Lanfner 2003). The UHI in Berlin is not very pronounced due to high ventilation caused by its flat landscape (elevation 34 m) and relatively dry surrounding areas (Kottmeier et al. 2007). However, the measuring period during the summer of 2018 was most characterised by high air temperatures, long sunshine hours, and low precipitation rates in Germany and especially in Berlin (Deutscher Wetterdienst 8/30/2018), which will intensify negative UHI impacts (Luber and McGeehin 2008; Bowler et al. 2010). Urban heat maximum in air temperature during the day in summer 2018 was 37°C in August and the mean summer temperature was 20.6°C, which marked a new record (Deutscher Wetterdienst 8/30/2018).

Berlin's area is made up of green spaces and water bodies (45%), infrastructural areas and transport (20%) and built-up areas around (35%) of 3% of the area of Berlin consists of AGs (Dugord et al. 2014). Berlin has 890 AG colonies on 2,932 ha and 71,473 plots as of May 2018 (Senatsverwaltung für Umwelt, Verkehr und Klimaschutz 2018). The size of AG colonies in Berlin ranges from 0.035 ha up to 44.16 ha (Senatsverwaltung für Stadtentwicklung und Wohnen 2018).


2.2 Data Collection

Selection of sites and metadata

AG colonies are defined as administrative entities only, but colonies right next to each other still form one continuous green area, which is relevant for the urban climate. However, streets can work as barriers and inhabit different climatic characteristics such as high amount of air pollutants (Chapman and Thornes 2005). To take this into account, all AG colonies in Berlin that were less than 20m apart from each other, based on the geodata of the Umweltatlas Berlin (Senatsverwaltung für Stadtentwicklung und Wohnen 2018), were aggregated into one entity via GIS analysis. A limit of 20m was chosen to avoid the influence of traffic passing through the entity, based on the average size of a street (Chapman and Thornes 2005). This leads to 530 allotment garden clusters (AGC) of various sizes and shapes distributed all over the city of Berlin. Through further GIS analysis, the parameters size (area in m²), perimeter (in m, capture scale 1:1.000) and edge-to-area ratio of each AGC were calculated. The edge-to-area ratio is calculated by dividing the perimeter by the area.

As the surrounding area has proven to influence the local microclimate, particularly up to a distance of 500m (Chudnovsky et al. 2004; Huang et al. 2008), the floor space index (FSI) was used as a factor to describe the environment surrounding the AGC. The FSI is defined as the ratio of a building's total floor area to the size of the piece of land upon which it is built. Therefore, the mean FSI was calculated for the 500m buffer surrounding the outer AGC border via GIS analysis, with data from Umweltatlas Berlin (Senatsverwaltung für Stadtentwicklung und Wohnen 2018).

Based on this analysis, 15 AGCs were selected that resembled a wide variety of the calculated parameters (Fig. 1, see also Appendix I). From each AGC, individual garden patches were chosen as measuring sites based on their location from the edge to the centre of each cluster to get a wide variety. This led to a total of 39 individual measuring sites distributed over the 15 AGCs, with a minimum of two stations per AGC to reduce the risk of measuring errors.

Air temperature sampling

At each of the 39 measuring sites, the air temperature was measured with Easy Log EL-USB-2 (resolution of 0.5°C, specific device error of 0.5 K between 0-30°C) at 10-minute intervals over the period from mid-July to mid-October 2018. Each logger was covered with a radiation protection case (white plastic, no ventilation), tied to a metal pole 1.70m above the ground and placed facing south. Before the installation of the measurement device the sky view factor (SVF) was calculated for each site from a fish-eye photo (taken from a 1.20m height) using the software SOLWEIG 2015a 1D. Only sites within the garden patches with a SVF higher than 0.59 were selected to make them comparable by reducing the influence of shade.

Throughout the measurement period, each logger was checked once to ensure they were all working properly and to save the data collected to date.

To compare the air temperatures from the AGCs with other urban areas in Berlin, five reference stations (two in parks and three urban stations in densely built up areas) were selected, based on Local Climate Zone (LCZ) categorisation and availability. As park reference stations, Tiergarten “TIER” (large size, irrigated, dense trees; LCZ: A) and Tempelhofer Feld “TEMP” (large size, not irrigated, low plants; LCZ: D) were selected based on park type and availability of data (Fenner et al. 2017). For the three densely built-up urban stations, Alexanderplatz ALEX (LCZ: 2), Bamberger Straße BAMB (LCZ: 2B) and Dessauer Straße DESS (LCZ: 2) were selected (Fenner et al. 2017). These three stations were aggregated to one densely built-up reference “URB”.

All reference stations were maintained by either the Technische Universität Berlin climate department or the Deutscher Wetterdienst (DWD). All measurement sites are shown in Fig. 1, for details see also Appendix II.

Fig. 1: Map of Berlin with the location of the AGCs (black: included samples, grey: excluded samples) and reference stations.


2.3 Data processing

The data obtained by the measurement campaign was first examined thoroughly for errors, which led to the exclusion of c15 and c3 (Fig. 1) due to missing values. This resulted in a sample size of 13 AGCs and 35 garden patches. The 13 AGCs were analysed with the statistical computing programme R, in order to answer the aforementioned research questions. Night-time air temperatures (), from 9:00 pm to 5:00 am UTC+2 based on the diurnal cycle of the sun (Thiele 2019) and calculated rate of change (Huang et al. 2008), were chosen for the analysis due to the distinctiveness of the UHI effect particularly during the night (Lin et al. 2018; Spronken-Smith and Oke 1999). Day-time values were not further investigated due to uncertainties about the robustness of the experimental set-up during intense solar radiation. The final dataset comprises a two-month period from 31st July to 3rd October 2018.

After a descriptive analysis, the air temperature differences between the AGCs ( ) as well as urban stations () and the two park references, dense trees () and low plants park () were tested. The following equation based on the PCI-equation of Spronken-Smith and Oke(1998) is used for calculating the differences of Tc and the reference stations:


 refers to the TC of each AGC, and  refers to the air temperature of each reference station. AGCs with a positive or negative  are referred to as colder or warmer than the reference station, respectively.

In order to test the significance of , a Welch-Anova (Bartesaghi Koc et al. 2017; Dag et al. 2018) and a Games-Howell post-hoc test (Delacre et al. 2017; Shingala and Rajyaguru 2015) were conducted, since a normal distribution could be verified, but no homogeneity of variance of . The statistical test was applied to all Tc in the respective time frame and aggregated into daily night-time means (nightly-Tc), which were used to calculate nightly differences (nightly ,) of each AGC from the nightly mean of all AGCs combined. In the same way, nightly differences (nightly , ,  ) of each AGC from the reference means (, , ) per night were calculated.

Following the statistical analysis of , the influence of different parameters on  was analysed. The selection of parameters was based on a correlation between generated parameters (FSI, size, edge-to-area ratio, perimeter) and an evaluation regarding the relevance and the quality of the parameters. The selected parameters, and the significance of their influence on the response variable, were tested in a univariate analysis with linear regressions as well as a multivariate analysis with a general linear model (GLM) (Egerer et al. 2019). The mean difference of the TURB for all nights during the time frame and the  for all nights during the time frame of each individual AGC (Chang et al. 2007), also abbreviated with an overbar: , was used as response variable for the linear regressions and for the GLM.  

3. Results

3.1 Temperature differences

Fig. 2: Boxplot of the temperature of AGCs and urban references, sorted by

Concerning the first research question, the results show that AGCs are cooler during the declared night period compared to the reference stations. All AGCs investigated in this study are significantly cooler than URB with reference to the overall mean (see Fig. 2). Both park references are significantly warmer than 11 of the 13 AGCs. The two warmest AGs, c12 and c10, are not significantly different to either park references and from each other, with a  of 15.9°C for both AGCs. Furthermore, the results show that the AGCs have a high internal variability, with a mean standard deviation of 5.0°C.

The two coldest AGCs, c14 with a Tc of 13.6°C and c6 with a of 13.7°C, are not significantly different from each other, but from the remaining AGCs. This shows a maximum  between  (17.7 °C) and the coolest station of 4.1 K (see Tab. 1). On average the AGCs were 3.0 K colder than URB and 1.5 K colder than the mean of both park references. AGCs with air temperatures not significant different to the external stations (c10, c12 for both parks and c4 additionally for TEMP) were excluded from the urban parks mean difference in air temperature.

Furthermore, the overall mean of all AGCs together, compared to URB and both park references, shows a significant   and provides evidence that the AGCs are, on average, colder. To sum up, was 1.6 K, was 1.4 K and   was 3.0 K.  Further details are presented in Appendix I.  


 in K

Min.  in K (AGC)

Max.  in K (AGC)



1.8 (c10)

4.1 (c14)



1.0 (c4)

2.5 (c14)



0.9 (c2)

2.3 (c14)

Tab. 1 Differences in air temperature from AGCs to the reference stations


Fig. 3: Boxplot of nightly mean differences between the overall mean of all AGCs and each individual AGC

When looking at the nightly differences (nightly  ) from the AGCs’ overall mean temperature there is a more distinguished pattern, which is shown in Fig. 3 (see also Appendix II). Some of the AGCs i.e. c5 and c13 show a small variance (0.16 and 0.22), whereas c14 or c7 display a greater variance (1.53 and 1.14). C12 showed the highest deviation from the mean, being 3.48 K warmer than the nightly , whereas c14 showed the coldest maximum difference of 4.38 K. On average each AGC deviated by 0.02 K from the overall mean.


Fig. 4: Boxplot showing nightly mean differences of all urban references and each AGC

Respectively, the nightly differences from the reference means show that there was only one single night in c8 where the Tc was 0.02 K warmer than TURB, as is visible in Fig. 4. The coolest night had a TURB of 9.09 K and occurred in c14. This AGC also had the coolest   and  in K with a maximum of 6.97 K and 6.09 K respectively, whereas  was 4.1 K. When looking at the two park references, the results show that several nightly Tc in all AGCs, were warmer than the reference in the same night. The highest negative nightly was detected in c12 with 2.55 K, whereas the highest negative  with 1.33 K was found in c9. The nightly-was 1.14 K and the TIER reference had a 1.09 K warmer nightly-. C14, the AGC with the coolest nights compared to all references, also had the biggest variance compared to all references. C12 and c10 had the smallest variance compared to all three references.


3.2 Influence of parameters

The parameters chosen to test on their influence on  were size, FSI and edge-to-area ratio as shape complexity of each AGC. Tab. 2 presents an overview of the parameter’s variation within the 13 AGCs. All three parameters show a wide range of values, which emphasize the diversity of the sample group.

Size in ha

Shape complexity




















Tab. 2: Selected parameters and their variation within sample groups (Details see Appendix I)

In order to set the  in relation to the parameters, boxplots including the  and AGCs were created and then sorted and colour-scaled by each parameter individually (Fig. 5 to Fig. 7). Fig. 5 depicts the AGCs sorted by shape, from the highest to the lowest complexity. The data suggests that an AGC with a more complex shape tends to have a higher  value. Accordingly, c10 and c12, which are the warmest AGCs, have the highest complexity values of 0.07 and 0.1. The coolest AGCs, c14 and c6, have both the sample group’s minimum value of 0.02. A similar trend is recognizable in Fig. 6, in which the AGCs and their TC are sorted by FSI, also from lowest to highest value. The higher the value of FSI, meaning the more built-up the surroundings of the AGC are, the higher the Tc. Again, c14 and c6, with a FSI lower than 0.4 are in contrast to c10 and c12, with the second (1.5) and third highest (1.83) FSI value. However, c2 has a higher FSI of 2.11, despite rather moderate temperatures. Fig. 7 displays a similar tendency, in which the warmest and coolest AGCs are not necessarily presenting the highest or lowest parameter values within the sample group. C6 with a size of 13.67 ha is the fourth largest AGC, c14 with a size three times as large (42.02 ha) is the second largest AGC, after c7 with 68.83 ha. However, c7 has a relatively moderate  despite being the largest AGC of the sample group, with a low FSI of 0.39 and a minimum shape complexity of 0.02. Nevertheless, c10 with a size of 0.61 ha and c12 with 0.29 ha are the smallest AGCs of the sample group.

Fig. 5: Boxplot of  of each AGC sorted by shape complexity ascendingly

Fig. 6: Boxplot of  of each AGC sorted by FSI ascendingly

Fig. 7: Boxplot of  of each AGC sorted by size (in ha) ascendingly

In order to statistically determine the relationship between Tc and the selected parameters, a regression analysis was conducted as described in chapter 2.3. The results of each linear regression as well as the correlation coefficient of the response variable and the predictors shape complexity, FSI and size are shown in Tab. 3. The shape complexity (shape) and FSI show a high significant linear relation to . The R-squared values (Tab. 3) indicate that FSI and shape complexity each explain around 46% of the variances. Furthermore, both parameters’ correlation coefficients show a strong negative correlation with the response variable. The parameter size, however, has neither a significant p-value nor a significant correlation coefficient. On that account, the general linear model (GLM) was run using the parameters FSI and shape complexity. Testing both variables in a multivariate analysis, neither of the two parameters show a significant p-value. Nevertheless, the coefficient of -0.72 shows a strongly negative correlation between the model and , and the R-squared value indicates that this calculated model explains 61.2% of the variances.







0.0058 **




0.0064 **




0.3261 .


Shape + FSI


Shape: 0.108

FSI:      0.119


Tab. 3: Regression results with target variable

In addition, Fig. 8 and Fig. 9 visualize the linear correlation between the response variable and the two significantly influential parameters, FSI and shape complexity respectively, the regression line and its equation. The regression line of the shape complexity as predictor shows a negative slope of -20.48, whereas the line for the FSI predictor has a shallower but also negative slope of -0.88, conforming to the results in Tab. 3.

These figures provide strong evidence in support of our hypothesis that a less complex shape and a lower degree of built-up surrounding area of the sampled AGCs can result in a higher , meaning a lower  and cooler AGC.

Fig. 8: Linear regression results and regression line of shape complexity and

Fig. 9: Linear regression results and regression line of FSI and  


4. Discussion

Differences to urban references

Densely built-up areas in Berlin are warmer than allotment gardens. This result is in line with previous studies about other urban green spaces, e.g. for Germany from Dugord et al.(2014) or Cabral et al.(2017b) and with the climatic aspects of AGs in Sydney detected by Egerer et al.(2019). Evapotranspiration of plants in AGCs leads to a lower temperature than in areas consisting of bare concrete, with a high sensible heat due to the absence of moisture. Therefore, like other urban green spaces, AGs could be valued as a contributing factor for UHI-mitigation (Bowler et al. 2010).

High trees, which are seen as the main driver of the cooling aspect of vegetation (Bowler et al. 2010; Vieira et al. 2018), are less common in AGs. Nevertheless, the analysed AGCs are cooler than the urban park TIER (LCZ A), which consists of dense trees. Heat storage underneath the trees during the night in the park TIER could lead to higher temperatures than in allotment gardens, where tall trees are prohibited by law (Bowler et al. 2010; BKleingG, revised 9/19/2006). However, it is possible that tall trees can be found in community areas (Cabral et al. 2017b) or at patches, if the trees had reached a certain size before the Federal Allotment Gardens Law (BKleinG) came into force. Then they were not felled or pruned. Furthermore, it is very likely that AGs are watered more frequently and more extensively than TIER and TEMP, especially during the dry summer of 2018. This favours evaporative cooling which could additionally lead to the temperature difference (Egerer et al. 2019; Deutscher Wetterdienst 8/30/2018).

The urban park TEMP (LCZ D) has a negligible tree density and mainly ground-level vegetation and shrubs, as well as sealed cover. Nevertheless, TEMP is also significantly warmer than the AGCs, but not to such an extent as TIER. TEMP is almost four times larger than the largest AGC (c7) (Appendix I). The lower amount of irrigation and therefore less evaporative cooling at TEMP and the lower density of vegetation (mainly grassland), which leads to a lower cooling effect (Vieira et al. 2018; Weng et al. 2007), could be an explanation for the findings. Different percentages and varieties of vegetation cover, imperviousness within the green space and the intensity of irrigation can all lead to varied air temperatures. This is also confirmed by Edmondson et al.(2016), Schwarz et al.(2012) and Egerer et al.(2019) for urban parks or urban gardens, respectively. It is stated that there can be two different kinds of park cool island effects (PCI), - daytime and night-time PCI (Erell et al. 2011). With a small number of trees and little irrigation, TEMP would be a typical park for a night time-PCI, whereas TIER fulfils the criteria for a daytime-PCI. AGCs show characteristics of both park types, however the nocturnal air temperature of TEMP depicts higher similarities to the AGCs. This might be caused by the lack of tall trees and less vegetation cover than TIER, although TIER and AGCs are frequently irrigated, unlike TEMP.


The size of the analysed AGCs was not a significant predictor for ,  and only showed a weak positive correlation. On the one hand, size is an important predictor for the air temperature for parks (Upmanis et al. 1998; Spronken-Smith and Oke 1998; Bowler et al. 2010), but results from Du et al.(2017), Yu et al.(2017) and Feyisa et al.(2014) suggest that after a specific threshold, the cooling effect does not extent any further. According to Dugord et al.(2014), this threshold is 10 ha for green spaces in Berlin. Additionally, Egerer et al.(2019) found that a larger garden has an increased percentage of vegetation cover, leading to lower air temperature variability. On the other hand, it has also been stated that there is no linear correlation between the size of a park and its air temperature (Bowler et al. 2010; Chen et al. 2014). In the present study, the small sample size of 13 AGCs and a lower scattering in size (only a few large AGCs) could be reasons for the insignificant regression.

Shape complexity

The results show that a complex shape of AGs can lead to higher air temperature. This is confirmed by many studies about urban parks (Feyisa et al. 2014; Ren et al. 2013), where the same relationship was shown. A complex shape could lead to a higher influence of the surrounding area, which is often urban landscape, and vice versa (Ren et al. 2013), resulting in an increased air temperature (Lin et al. 2018). However, with a complex shape the cold air from AGs has a higher chance to move into their surroundings. For urban green in Berlin, Dugord et al.(2014) stated that large and complex-shaped forested green spaces, as well as interconnected and spatially aggregated green spaces,, reduced the temperatures significantly. The type of vegetation and the interaction between size and shape are stated as essential influencing factors. Tree-covered spaces get colder with increasing shape and size, whereas grass-covered areas behave differently depending on their climate zone (Yu et al. 2017). The study of Jaganmohan et al.(2016) on green spaces in Leipzig showed that green spaces larger than 6.27 ha have a positive correlation between shape and PCI, whereas green spaces smaller than 6.27 ha have a negative correlation between these two factors. Due to a non-significant result regarding the size predictor, the interaction between shape and size was not tested in this study.



As presented in the results, with an increasing built-up area surrounding the AGCs, the air temperature of AGCs increases as well. The stored heat is emitted overnight by sealed surfaces and built-up areas such that it impacts the surroundings, especially at night (Oke 1973). The warm air masses move along the gradient into the cooler, adjacent AGCs and increase its air temperature. The influence of the surroundings on urban gardens was also studied by Egerer et al.(2019) and by Lin et al.(2018). Both studies concluded that the surrounding environment had an influence on the gardens, meaning UHI effects were more visible in gardens with densely built-up surroundings.

Furthermore, Quanz et al.(2018) showed that areas close to park-like spaces are cooler and can experience a cooling effect e.g. due to cold air flows and heat exchange. The results of this study show that the PCI effect can be found in allotment gardens.  A certain cooling effect of AGCs could be expected towards the surrounding area, albeit dependent on a variety of parameters. For example, for c14, all described parameters seem to have the expected impact: it is the coldest AGC with the highest difference in air temperature to all reference stations in mean and in total. Furthermore, the highest variance occurred at this AGC. It has the lowest shape-complexity as well as FSI and is the second-largest AGC. However, this AGC was also located in the outskirts of Berlin surrounded by a highly vegetated surface cover. C2, however, shows a contrary picture. It has the highest FSI, but shows rather moderate temperatures. This could b