Influence of Canopy Cover on Surface Temperature

Introduction Vegetation creates microclimates that are distinct from open environments. The vegetation acts as a windbreak, affecting the heat transfer and transpiration of living beings (Smith and Jarvis, 1998). Plant transpiration also increases humidity and transforms some of the solar radiation as latent heat, which contributes to tissue cooling (Gowing et al., 2008). Canopy and canopy litter have different albedos than that of bare soil, and they prevent direct solar radiation from reaching the ground surface. All these changes can affect the local niches of ecosystems and, furthermore, could be important when planning for more effective thermal comfort in environments inhabited by humans. Maximum and minimum temperatures are important factors used by plants and other living beings to control their seasonal cycles (i.e., phenology), such as blossoming and leaf senescence (Ricklefs and Relyea, 2019). Many enzymes and biological processes within living beings also have tolerance ranges and optimal points regarding environment temperature (Brown et al., 2004). Therefore, the microclimates in open areas and in areas under canopy cover may offer distinct conditions in terms of ecosystem development. Along the process of ecological succession, the development of larger trees with a thicker canopy would then change the environmental conditions for different living beings in the ecosystem. These effects of tree cover on temperature are also important when planning better thermal comfort zones in green areas, such as parks, squares and gardens. Tree shade can be a good ISSN:1984-2295 Revista Brasileira de Geografia Física Homepage: https://periodicos.ufpe.br/revistas/rbgfe Revista Brasileira de Geografia Física v.13, n.07 (2020) 3275-3286.


Introduction
Vegetation creates microclimates that are distinct from open environments. The vegetation acts as a windbreak, affecting the heat transfer and transpiration of living beings (Smith and Jarvis, 1998). Plant transpiration also increases humidity and transforms some of the solar radiation as latent heat, which contributes to tissue cooling (Gowing et al., 2008). Canopy and canopy litter have different albedos than that of bare soil, and they prevent direct solar radiation from reaching the ground surface. All these changes can affect the local niches of ecosystems and, furthermore, could be important when planning for more effective thermal comfort in environments inhabited by humans.
Maximum and minimum temperatures are important factors used by plants and other living beings to control their seasonal cycles (i.e., phenology), such as blossoming and leaf senescence (Ricklefs and Relyea, 2019). Many enzymes and biological processes within living beings also have tolerance ranges and optimal points regarding environment temperature (Brown et al., 2004). Therefore, the microclimates in open areas and in areas under canopy cover may offer distinct conditions in terms of ecosystem development. Along the process of ecological succession, the development of larger trees with a thicker canopy would then change the environmental conditions for different living beings in the ecosystem.
These effects of tree cover on temperature are also important when planning better thermal comfort zones in green areas, such as parks, squares and gardens. Tree shade can be a good ISSN:1984ISSN: -2295 Revista Brasileira de Geografia Física Homepage: https://periodicos.ufpe.br/revistas/rbgfe refuge for people and other animals during periods of hot temperatures, such as heat waves. Planning that includes access and equipment (i.e., seats and tables) under trees may add even more value to the ecosystem services of these green areas (Tyrväinen et al., 2005). Additionally, the current trend of global warming may increase the importance of the ecological services provided by trees in humaninhabited areas (Salmond et al., 2016). The objective of this study was to investigate the relationships between the canopy cover and variations in the surface temperature. These relationships are investigated through the analysis of mean difference tests and multiple regression. The main underlying process under investigation is the reduction in direct solar radiation caused by canopy cover, and the primary hypothesis assumes that the canopy decreases the maximum and minimum daily ground temperatures, as well as the daily temperature range, at the microclimate scale.

Data Collection
The Open University of the United Kingdom organized a collaborative data measurement project that collected data from different places around the world between 29 October and 14 December 2016. In total, the collaborators obtained 363 measurements, each consisting of the maximum and minimum surface temperatures in an open area and an area under the canopy (i.e. 1 metre from the trunk) during a continuous 24-hour period. Each collaborator used two Tmin/Tmax thermometers to record the maximum and minimum air temperatures at the surface level, unshielded and unaspirated (i.e. measurement under natural ventilation).
Most measurements (91.5%) were collected in the United Kingdom; however, 6.3% of measurements were collected in other European countries, and 2.2% were collected elsewhere in the world (including four sites in the Southern Hemisphere) (Figure 1). Therefore, while the deciduous and semi-deciduous trees in the Northern Hemisphere were in the process of losing their leaves during autumn, the deciduous and semi-deciduous trees in the Southern Hemisphere were sprouting their new leaves during spring. The possible influencing environmental factors were noted on a standardized data sheet, which is available as a supplement to this article. Canopy cover was estimated using a reference sheet (Vasconcelos (2021a). Canopy cover was estimated in intervals of 10%, and the values were transformed into a canopy index that represented the mid-values of each interval. Each collaborator was instructed to select trees that were located as far as possible from buildings and light sources (preferably 20 metres or more). Additionally, collaborators were instructed to select trees that were mature and had well-formed deciduous canopies. Davies and Gowing (2016) described the prescribed methods for collection and analysis in more detail. After collection, the data were synthesized by The Open University team and then returned to the collaborators, who could then perform their own statistical analysis.
There was also a prescribed list of tree species, which included the following, in order of preference: Quercus robur (pedunculated oak, comprising 56.7% of the trees), Quercus petraea (sessile or durmast oak, comprising 19.3%), other oaks (8%), other deciduous trees (13.2%), other trees (1.7%) and other shrubs (1.1%, including woody plants with no clear trunk and foliage within 0.5 m of the ground). The preference was given to oak species because they retain their leaves for a longer period of time in autumn, providing more uniformity within the measurement period. The ground coverage also followed a prescribed preference: grass (53.1%), litter (20.8%), undergrowth (17.5%), bare soil (7.1%) and artificial surface (1.5%). Among other environmental variables, the collaborators noted whether the environment was rural (44.6%) or urban (55.4%, which included villages and suburban areas), the geographical coordinates, the distance from the thermometer under the canopy to the edge of the canopy, and the distance to the nearest light source.

Modelling Framework
This article follows a theoretical framework that is based on King and Roberts (2014); specifically, rather than selecting what would be the best model among conventional parametric, robust or non-parametric alternatives, these modelling approaches could be interpreted as complementary perspectives of the same phenomena under analysis. Therefore, if similar patterns emerged from these distinct modelling approaches that used the same database, there was stronger evidence that these patterns were valid. On the other hand, if there was significant disagreement among these distinct modelling approaches, it was usually an indicator that there was a need for further analysis and modelling improvement. The significance values (p-levels) were interpreted in conjunction with the effect sizes, as proposed by Henson and Smith (2000). The power was evaluated using the recommendations of Hair et al. (2018) when measuring the risk of type II errors.
The individual daily maximum and minimum temperature measurements would undoubtedly diverge because the collaborators collected data from various locations under distinct weather and other environmental conditions. To address this variation, a paired analysis was used to compute temperature differences between each pair of open area and canopy cover measurements. The difference between the maximum and minimum temperatures in the open area and the area under the canopy was calculated using Equations 1 and 2, respectively. The difference between the temperature range in the open area and the area under the canopy, termed D, was calculated using Equation 3. Table 1 presents these three equations. A paired t-test was conducted to evaluate the overall effect of the tree canopy on the three variables listed in Table 3. The parametric assumption of normality was verified using the Shapiro and Wilk (1965) W test. The assumption of homogeneity of variance was checked using Levene's (1960) test. The significance of the paired t-test was re-checked using the Yuen (1974) robust paired t-test on 20% trimmed means and the Wilcoxon (1945) non-parametric signed rank test with continuity correction.
In addition to significant differences obtained in the t-tests, two multiple linear regression analyses were performed to understand the effects of possible causes on these processes. One multiple regression evaluated the effect of the percentage of canopy cover and urbanization (i.e., the independent variables) on the difference in the maximum temperature values (i.e., the dependent variable). Urbanization was coded as a dummy variable (rural = 1, urban = 0).
The other multiple regression analysis evaluated the effects of the canopy index and the distance to the edge of the canopy (i.e. independent variables) on D (i.e. a dependent variable). Nonlinear relationships were verified between the independent and dependent variables. The residual autocorrelation was verified using the Durbin and Watson (1971) test. The homoscedasticity was evaluated using the score test for non-constant error variance of Breusch and Pagan (1979). Multicollinearity was evaluated by the variance inflation factor (VIF), according to the guidelines of Hair et al. (2018) The outliers were evaluated by counting the cases in which Cook's distance was higher than 1, as advised by Cook and Weisberg (1982), and the number of hat (i.e., leverage) values that were higher than 3(k+1)/n, where k is the number of parameters and n is the number of cases, as advised by Pituch and Stevens (2015). In addition to the conventional ordinary least squares (OLS) method, the multiple linear regressions were evaluated using the robust standard deviations (White,1980), the robust iterated reweighted least squares (IRWLS) with a method-of-moments (MM) estimator (Yohai and Zamar, 1988;Koller and Stalhel, 2011), and the non-parametric quantile regression (Koenker and Bassett, 1978) with pseudo-R 2 (Cragg and Uhler, 1970;Nagelkerke, 1991). The MM-estimator for IRWLS was preferred over the M-estimator (Holland and Welsch, 1977) because it can generate robust R 2 and significance values (Renaud and Victoria-Feser, 2010), which makes it easier to compare the results with those of the other modelling approaches.
Three independent sample t-tests, with the Welch (1938) adjustment for heterogeneity of variance, were conducted to evaluate the effects of urbanization and the distance from light sources on the independent variables. The analogous diagnostics as well as the robust (with bootstrapping) and non-parametric tests used in the paired samples were also used in the independent sample t-tests. The goal of performing the independent t-tests was to separately evaluate the effects of potential explanatory variables whose coefficients were not significant when incorporated into the multiple regression analyses together with other variables.
The null and alternate hypotheses for each model are described in Table 2. The parametric assumptions of each model were tested, and the statistical significance of the results was doublechecked using robust and non-parametric alternatives. The code in R used for the statistical analysis and the respective datasets are provided respectively in Vasconcelos (2018) and Vasconcelos (2021b).

Tmaxopen -Tmaxcanopy
Tree canopy has no effect on daily maximum temperature Tree canopy changes daily maximum temperature Tminopen -Tmincanopy Tree canopy has no effect on daily minimum temperature Tree canopy changes daily minimum temperature D Tree canopy has no effect on the daily temperature range (D) Tree canopy changes daily temperature range (D)

Tmaxopen -Tmaxcanopy
The canopy cover and urbanization have no effect on the difference between the daily maximum temperatures in the open area and the area under the canopy The canopy cover and urbanization affect the difference between the daily maximum temperatures in the open area and the area under the canopy D The canopy cover and distance from the edge of the canopy have no effect on the difference between the daily temperature ranges (D) in the open area and the area under the canopy The canopy cover and distance from the edge of canopy affect the difference between the daily temperature ranges (D) in the open area and the area under the canopy

Tminopen -Tmincanopy
The distance from the light source has no effect on the difference between the daily minimum The distance from the light source has an effect on the difference between the daily minimum temperatures in The distance from the light source has no effect on the difference between the daily temperature ranges (D) in the open area and the area under the canopy The distance from the light source has an effect on the difference between the daily temperature ranges (D) in the open area and the area under the canopy The urbanization has no effect on the difference between the daily temperature ranges (D) in the open area and the area under the canopy The urbanization has an effect on the difference between the daily temperature ranges (D) in the open area and the area under the canopy Table 3 shows the results of the paired ttests. The Shapiro-Wilk normality test on the three variables shown in Table 1 were significant with 99.9% confidence, indicating that the data were significantly different from that of a normal distribution. However, even if the sample size was large enough to generate a normal distribution based on the central limit theorem, Table 3 also presents the p-values for the robust and nonparametric alternatives to the paired sample tests. Levene's test also showed that, at least for D, the variance in the open area and the area under the canopy was significantly distinct, violating another parametric assumption. Nevertheless, the p-values of the parametric, non-parametric and robust tests for the three variables were significant at the 99% confidence level. The areas under the canopy exhibited lower maximum and minimum daily temperatures and presented narrower daily temperature ranges. Nevertheless, the average difference and effect size (i.e., the Cohen "d") of the daily maximum temperatures were lower than those of the other two variables.

Results
Tables 4 and 5 present the data for the multiple regressions of the two formulas (Tmaxopen -Tmaxcanopy and D as dependent variables). The two formulas were significant at the 99.9% confidence level using OLS (including the robust standard deviation correction) and IRWLS, and the formulas were significant at the 99% confidence level using quantile regression. The coefficients of the dependent variables were also significant at the 95% confidence for OLS (even including the robust standard deviation correction), IRWLS and quantile regression. However, all models had very low R 2 values, from 3.3% to 5.9%, indicating that there may have been many other environmental factors influencing the behaviour of the temperature both under the canopy and in the open. The Durbin-Watson tests with W values near 2 indicated that autocorrelation between residuals would not adversely affect the outcome of either model.
The VIF also indicated low multicollinearity for both models. However, the Breusch-Pagan test showed that the hypothesis of homoscedasticity of the residuals was not significant for both models; thus, paying special attention to the robust and non-parametric models is recommended.
For each additional 1% of canopy cover, the difference between the maximum daily temperatures in the open area and the area under the canopy increased by 0.026°C (OLS), 0.016°C (IRWLS), or 0.012°C (quantile regression). Therefore, a tree with 95% canopy cover would have a maximum daily temperature that was 2.94°C (OLS), 1.52°C (IRWLS), or 1.14°C (quantile regression) lower than that in the adjacent open areas. In rural areas, the difference between the maximum daily temperatures in the open area and the area under the canopy was 0.782°C (OLS), 0.937°C (IRWLS), or 0.675°C (quantile regression) higher than those in the urban areas. Comparing the standardized coefficients, the canopy cover had more influence than did urbanization on the difference between the maximum daily temperatures in the open area and the area under the canopy using OLS (even including the correction for robust standard deviation); however, the relative influence of urbanization was higher using quantile regression and even higher using IRWLS. In OLS, there were no outliers identified by the Cook's distance method; furthermore, only one outlier was identified using the leverage values, and it had a small influence (0.8%) relative to all the leverage values of the model.

D = (T max open -T min open) -(T max canopy -T min canopy)
+ 20 metres to light source Regarding the other multiple regression model, the tests with non-linear models indicated that the squared variable for canopy cover was more strongly correlated with the difference between the daily temperature ranges in the open area and the area under the canopy (D). Therefore, canopy cover was transformed in this way before running the linear regression. Versions including the linear and quadratic transformations of the canopy cover in the same equation did not improve the regression, and thus, the quadratic transformation was used alone. According to the model, a tree with 5% canopy cover would have a daily temperature range that was 0.0075°C (OLS) or 0.0050°C (IRWLS and quantile regression) narrower under the canopy than in the open; furthermore, these values changed to 0.75°C (OLS) or 0.5°C (IRWLS and quantile regression) for a tree with 50% canopy cover and 2.71°C (OLS) or 1.80°C (IRWLS and quantile regression) for a tree with 95% canopy cover. For each metre between the thermometer and the edge of the canopy, the daily temperature range was 0.127°C (OLS and IRWLS) or 0.142°C (quantile regression) narrower than that in the open. Comparing the standardized coefficients, the canopy cover had more influence than did the distance to the edge of the canopy on the difference between the daily temperature range in the open area and the area under the canopy in OLS (even including correction for robust standard deviation); however, this effect gradually decreased in IRWLS and decreased even more in quantile regression. In OLS, no outlier was detected using Cook's distance; however, four outliers were detected using leverage values, influencing 12.5% of the model results. Theoretically, IRWLS and quantile regression would decrease the influence of these outliers in the model and compensate for this bias. Table 6 presents the results for the t-tests with independent samples. Although Levene's tests indicated that the variance should be homogeneous for the three tests, the Welch adjustment was still applied to all tests as an extra precaution. The three tests had small effect sizes (Cohen "d"); however, they were significant at the 95% confidence level for the parametric and non-parametric versions of the tests. However, the influence of urbanization on D was not significant in the robust version of the test, demanding higher caution when making inferences from this statistical effect. In measurements that were at least 20 metres from light sources, the daily minimum temperatures under the canopy were 0.78°C higher and the temperature ranges under the canopy were 1.17°C narrower than the respective values in the open area.
The difference between the daily temperature range in the open and under the canopy was 0.87°C higher in rural areas than the values in urban areas. Therefore, all the null hypotheses presented in Table 2 can be rejected with at least 95% confidence using the parametric and non-parametric models. The robust models also supported the same patterns with 95% confidence, except for the influence of urbanization on the temperature range (D). Over a broad spectrum, the modelling results corroborated the respective alternative hypotheses from Table 2. All tests presented a power higher than 0.8, which was advised by Hair et al. (2018) as a cautious threshold that could be used to avoid type II errors. The high power of the tests also indicated a good balance in the experimental design regarding the sample size, effect size, and control of type I and II errors. Table 7 synthesizes the underlying processes that could have caused the effects described in the models. These processes can be discussed based on how Carlson and Boland (1978) developed a surface heat flux/temperature model to compare the canopy effects in urban and rural environments. They proposed that the main factor controlling the contrasting results between urban and rural environments is the ground moisture availability and ground conductance (thermal inertia). According to these authors, the ground moisture availability in cities is usually lower because of reduced areas in evapotranspiration, intensifying the daytime urban heat island effect. Ground conductance tends to be higher in urban areas because structures and artificial surfaces retain heat during the day and release it during the night, increasing night-time minimum temperatures. These processes show the relevance of evapotranspiration and lateral heat emissions (including horizontal radiation and wind advection), which were used for the explanation of many of the patterns described in Table 7. An atmospheric heat flow model proposed by Shashua- Bar and Hoffman (2003) demonstrated how the tree canopy changes the microclimate of urban areas through heat-pump and windbreak effects. The model resulted in general cooling and a decrease in temperature variation (consistent with Table 7), especially during daytime. Heat pumping through tree evapotranspiration (evaporative cooling) was also emphasized by Lewis (1998) as a relevant factor associated with decreasing ground temperatures, measured in forested and deforested areas. The light source may provide a lateral heat flux to places under the canopy Proximity to a light source decreases the buffering effect of canopy on the temperature range The light source may provide a lateral heat flux to places under the canopy Lower temperatures under tree canopies than on unshaded ground in urban environments were previously reported by Mascaró and Mascaró (2009). Robinette (1972) reported that the tree canopy controls solar radiation and increases air humidity, thus decreasing temperature variation and reducing the thermal amplitude under the vegetation, consistent with the patterns in Table 7. Both Robinette (1972) and Mascaró and Mascaró (2009) also reported that the canopy effect was more effective during the summer because the leaf density and tree evapotranspiration were more intense. Both also found that the canopy effect of groups of trees was cumulatively more intense than that of isolated trees, especially when the various layers of the canopy amplified the absorption of solar radiation and the stratification of air temperature under the vegetation.

Underlying processes
Various studies using remote sensing data supported the phenomena that tree removal increases average and maximum temperatures 1 in 1 It is necessary, nevertheless, to keep in mind that remote sensing studies would mainly reflect surface temperatures, not air temperatures, and would mainly reflect the top of canopy temperatures, not those under canopy. For example, the influence of canopy on the land cover temperature could be very significant in a both urban (Nichol, 1996;Elmes et al., 2017;Geene and Milward, 2017) and non-urban areas (Godinho et al., 2016) (the latter study also focused on oaks, as this study did), and the findings are consistent with the patterns observed in this study. An exception would be the trees in snow-covered areas, where the trees have a lower albedo and thus absorb more light energy than the bare snow (Bonan et al., 1992). Li et al. (2015), based on remote sensing analysis, proposed that the balance between evaporative cooling and albedo warming of trees would result in higher net cooling at lower latitudes that would decrease towards higher latitudes until reaching negative net values in boreal forests of snow-covered areas. In temperate zones, where most of the measurements of the current study were taken, Li et al. (2015) and Alkama and Cescatti (2016) proposed that the net cooling would be higher in summer and negative during winter (because of snow). However, as the measurements of this experiment were taken in autumn (Northern remote sensing image, yet evapotranspirational effects would be weakly observable because they are mixed through the air volume. Hemisphere) and spring (Southern Hemisphere), these snow-covered areas would have been restricted to high latitudes; therefore, the positive net cooling is consistent with these remote sensing analyses. Alkama and Cescatti (2016) also found that removing tree cover in areas/periods without snow increased maximum air temperatures and amplified diurnal temperature variation, consistent with the patterns in Table 7.

Model reliability
The reliability of these statistical models may be higher in the regions that had higher concentrations of sampling sites, i.e., primarily in the United Kingdom and secondarily in Europe. Therefore, further studies should include a more balanced distribution of measurements around the globe to verify the extent at which these patterns are uniform or differ within distinct regions. The same models as presented in this article were tested using geographical subsets (i.e. only samples from the Northern Hemisphere, from Europe, or from the United Kingdom); however, the changes in the effect sizes, R 2 values, and p-values were so small that they did not justify the removal of the other measurements, especially considering the trade-off with the reduction in spatial coverage. In the same way, model runs that used subsets that excluded shrubs, as well as excluding measurements on artificial ground surfaces, resulted in very small changes that did not justify their exclusion.
Overall, the leverage values of distant measurements in this study were not higher than those in the United Kingdom. Additionally, the outliers detected by the leverage values in the multiple regressions did not include shrubs or samples measured on artificial ground surfaces. Four of these outliers were in the United Kingdom, two were in continental Europe, and none were outside Europe. This result may indicate that the patterns and processes found in this study are robust in many environmental situations; however, there is still a clear need for further research.

Conclusions
The paired t-tests showed a significant effect of tree canopy; specifically, the tree canopy decreased the lower maximum and minimum daily temperature measurements, and it shortened the daily temperature range. The multiple regressions showed that the percentage of canopy cover significantly affected the difference between the maximum temperatures in the open areas and the areas under the canopy, as well as the difference between the temperature ranges. The regressions also showed that urbanization decreased the effect of the canopy on the maximum daily temperatures; additionally, the distance to the edge of the canopy was positively correlated with the difference between the daily temperature ranges in the open areas and the areas under the canopy. However, the low R 2 values for both regression models indicated that there may be a large effect of other environmental variables that influence temperature. The independent sample t-tests showed that the distance from the light source was significant in decreasing the difference between the minimum daily temperatures in the open areas and the areas under the canopy; however, the distance from the light source increased the difference between the daily temperature ranges. The independent sample t-tests also indicated that urbanized areas have lower differences between daily temperature ranges in the open areas and the areas under the canopy according to the parametric and non-parametric tests but not according to the robust test.
The main underlying processes that could explain these results include canopy shade and wind insulation, litter insulation of the ground surface, and tree heat pumps through evapotranspiration. Light sources and other heat sources, especially in urban environments, may add lateral heat transfer, which would influence the results. The results from the models and the discussion illustrate the relevant influence of trees on the microclimate, and these influences have consequences on ecosystem processes as well as on the thermal comfort in areas inhabited by humans.
This study is also a great example showing how collaborative data collection can be an effective approach to gathering information on environmental processes in a way that integrates local contexts from many parts of the globe. Most previous studies that evaluated the effects of the canopy cover on a large scale were based on remote sensing (Nichol, 1996;Elmes et al., 2017;Greene and Milward, 2017;Godinho et al., 2016;Bonan et al., 1992;Li et al., 2015;Alkama and Cescatti, 2016), and the ones based on field measurements were usually of a limited scale. The novelty of this study, which was based on a collaborative framework, was the performance of field measurements to a larger spatial extent. Nevertheless, the design also adds some uncertainty regarding the uniformity of the data collection procedures among all collaborators. To increase the uniformity, reduce the uncertainty, and allow a larger number of participants, the experimental design needed to be simplified compared with that of previous similar studies, which focused on a limited spatial scale but with a more detailed collection of weather and environmental variables (Carlson and Boland, 1978;Lewis, 1998;Mascaró and Mascaró, 2009;Robinette, 1972). As discussed in the previous subsection, this study helped corroborate previous hypotheses on processes that have been proposed both from detailed small-scale experiments that lacked confirmation on broader spatial scales and from large-scale remote sensing studies that lacked detailed field measurement validation. Therefore, this study contributes to filling this gap as an interface between existing large-scale and small-scale experiments regarding the effect of the canopy cover on temperature.
Further modelling studies, with appropriate fieldwork designs, could investigate the joint effects of other variables on temperature together with the variables investigated in this article. These variables could include longitude, type of ground surface, tree species, tree high, distance to other tree canopies, distance to buildings (or walls) and distance to coasts.