The Impact of Microclimate on the Reproductive Phenology of Female Populus tomentosa in a Micro-Scale Urban Green Space in Beijing

: The spatial variation of poplars’ reproductive phenology in Beijing’s urban area has ag-gravated the threat of poplar ﬂuff (cotton-like ﬂying seeds) to public health. This research explored the impact of microclimate conditions on the reproductive phenology of female Populus tomentosa in Taoranting Park, a micro-scale green space in Beijing (range <1 km). The observed phenophases covered ﬂowering, fruiting, and seed dispersal, and ENVI-MET was applied to simulate the effect of the microclimate on SGS (start day of the growing season). The results showed that a signiﬁcant spatial variation in poplar reproductive phenology existed at the research site. The variation was signiﬁcantly affected by the microclimate factors DMT (daily mean temperature) and DMH (daily mean heat transfer coefﬁcient), with air temperature playing a primary role. Speciﬁcally, the phenology of ﬂowering and fruiting phenophases (BBB, BF, FF, FS) was negatively correlated with DMT ( − 0.983 ≤ r ≤ − 0.908, p <0.01) and positively correlated with DMH (0.769 ≤ r ≤ 0.864, p < 0.05). In contrast, DSD (duration of seed dispersal) showed a positive correlation with DMT (r = 0.946, p < 0.01) and a negative correlation with DMH (r = − 0.922, p < 0.01). Based on the ﬁndings, the increase in air convection with lower air temperature and decrease in microclimate variation in green space can be an effective way to shorten the seed-ﬂying duration to tackle poplar ﬂuff pollution in Beijing’s early spring.


Introduction
The rapid changes in urban environments driven by the massive landscape reshaping during urbanization have dramatically shifted the growth rhythm of city plants [1][2][3][4]. The variation of the phenological temporal pattern can affect public health by influencing the occurrence period of phenology-triggered diseases [5,6].
A long-term phenology-correlated public health problem in Beijing is the seasonal respiratory diseases caused by the cotton-like flying seeds from Populus tomentosa (Chinese white poplar) and other Salicaceae species in early spring [7,8]. P. tomentosa is one of the most common and widespread deciduous woody species native to Beijing. It is dioecious and wind-pollinated, blooming in early spring before leafing [9]. In the 1960s, 0.2 million cloned female poplars were planted in Beijing city in a large-scale city greening action, in consideration of their rapid propagation and fast growth at a young age. Since then, the catkin fibers have become a major trigger of respiratory ailments, skin anaphylaxis, inflammation, and other diseases in Beijing's early spring, posing a great risk to public health [10,11]. Beyond China, some other countries also face similar health problems caused by poplar fluff. For example, in Kashmir, India, the flying seeds of introduced Eastern Cottonwood (Populus deltoides) have given significant rise to the respiratory disease cases of local people in recent years [12,13]. In Abbottabad, Pakistan, as one of the main triggers of allergic asthma in spring, poplar fluff has caused severe psychological stress [14]. In Beijing, several strategies have been put forward by the Beijing Gardening and Greening Bureau to control poplar seed pollution, such as breeding sterile triploid female poplars, replacing female plants with male ones, and inhibiting catkin formation [15,16]. However, the heavy workload, high economic cost, and limited budget for treating such a huge number of targeted poplars made the implementation very difficult. Furthermore, the interindividual difference in the seed dispersal time between different regions or within green spaces extended the overall flying-seed period in Beijing's urban area [17], aggravating the threat of catkin fibers to public health. A profound understanding of the temporal pattern of poplars' reproductive phenology and the impact mechanism behind the spatial variation is the basic step to tackle this problem.
In recent years, a prominent intra-urban spatial variation of plant spring phenology has been found in many metropolises at both an urban scale (<50 km) [18][19][20][21] and microscale (<1 km) [19]. With regard to the impact mechanism associated with the intra-urban phenological variation, most of the research was conducted at an urban scale, revealing that the urban-scale variation in spring phenology was mainly driven by environmental factors, especially the near-surface air temperature that indicates urban heat [19,21,22]. However, the key influential factors of the phenological variation at the micro-scale are still largely unknown, even though a spatially uneven microclimate has been found in urban green spaces [23,24]. For urban-scale research, a significant phenological variation within a green space can affect the overall spatial pattern of studied phenology [22]. Therefore, more micro-scale studies need to be done.
In this research, we focused on the spatial variation of the reproductive phenology of female P. tomentosa in Taoranting Park, a micro-scale green space in Beijing, aiming to explore the key influential factors of this phenological variation. We attempted to answer two questions: (1) Is there a significant spatial variation in the reproductive phenology of female P. tomentosa in Taoranting Park? (2) What are the key influential factors for this variation? Assuming the impact mechanism at the micro-scale is similar to what has been widely revealed for the urban scale, we proposed two hypotheses to be tested.

Hypothesis 1 (H1):
There was no significant difference between the reproductive phenology of P. tomentosa in different sampling points of Taoranting Park (p > 0.05). H 1 : There was a significant difference between the reproductive phenology of P. tomentosa in different sampling points of Taoranting Park (p < 0.05).

Hypothesis 2 (H2):
The spatial variation in reproductive phenology had no significant correlation with any microclimate factor (p > 0.05). H 1 : The spatial variation in reproductive phenology had a significant correlation with at least one of the microclimate factors (p < 0.05).
Based on the revealed impact, this study attempted to gain some insight into more effective control and management of poplars' catkin fiber pollution in Beijing's early spring.

Research Site
The research was conducted in Taoranting Park, Xicheng District, Beijing, China. Beijing (39 • 54 N 116 • 24 E), the capital city of China, is located at the northern tip of the North China Plain. Beijing has a monsoon-influenced humid continental climate, characterized by hot, humid summers and cold, dry winters. Taoranting Park (39 • 52 21" N 116 • 22 32" E) is situated in the central downtown area of the dense capital city. As one of the largest urban public gardens in Beijing, with a 65-year development history, Taoranting Park is famous for the flourishing vegetation with rich diversity and healthy growth conditions. The dense evergreen forest dominated by Platycladus orientalis (L.) and Juniperus chinensis L. together with deciduous groves has formed various green spaces in the garden, and the uneven landscape structure of the underlying surface has resulted in a highly diverse microclimate environment.

Research Object
Female Populus tomentosa (Chinese white poplar) trees growing in Taoranting Park were the research objects for phenological observation. According to the garden construction record, the adult female white poplars in Taoranting Park were planted in 1952. The seedlings were a batch of clones that were propagated with the cutting of root suckers from a stock plant. Therefore, the influence of genetic variation on the interindividual phenological variation [25,26] could be excluded in this research.

Selection of Sample Poplars and Setting of Sampling Points
There were eight female poplar populations distributed in the green space, with each population composed of 3-5 individuals. From each population, we selected two adjacent adult individuals with good growth conditions (that performed well in yearly growth and reproduction, with no sign of diseases or being subject to serious environmental stress) as the sample poplars of this population; their canopy height ranged from 5 to 11 m. The central growing point of the two sampling trees was set as the sampling point; the distribution of eight sampling points is shown in Figure 1. From each sampled individual, three branchlets at the height of 7-9 m above ground were selected for phenological observation, with 20-30 catkins growing on each branchlet. A branchlet refers to a 2-year-old branch formed in the previous year [27]. The phenology data of each branchlet were obtained from the earliest catkin, and the data for each sampling point were the values of six branchlets (three branchlets/individual × two individuals).

Setting of the Buffer Area
A buffer area was set around each sampling point for microclimate simulation. Each buffer area was a 100 m × 100 m square centered on the sampling point, as shown in Figure 1. The size of the buffer area was determined with reference to the findings of previous research, i.e., that the microclimate of a point in urban green land can interact with the thermal environment of its surrounding area within a range of 100-200 m [24,28,29]. Considering that the distance between any two sampling points was less than 100 m in this research, the side length of the square buffer area was set to 100 m to prevent considerable overlap.

Reproductive Phenology Data Collection
The reproductive phenology of female P. tomentosa was observed from February 20 until April 18 in 2019, covering the entire reproduction process from the beginning of bud break to the end of seed dispersal. The observation was conducted every day at 13:00-14:00, when the air temperature reaches its daily maximum and often drives a phenological change [30].
The BBCH (Biologische Bundesanstalt, Bundessortenamt and Chemical Industry) scale was used to identify the phenological development stages, i.e., phenophases [31], and the phenology data of each phenophase were recorded in Julian Day (the day of the year). The observed phenophases covered flowering, fruiting, and seed dispersal, including the

Data Collection of Microclimate Factors with ENVI-MET Simulation
We needed to evaluate the microclimate environment at the height of the sampled branches (7-9 m). Field data measurements are generally accepted and widely applied in the data collection of the microclimate condition, including using fixed equipment [32]. However, this method was not feasible in this research, subject to the management regulation of the Park Administrative Office, that no equipment was allowed to be attached to tree branches to prevent potential damage caused to the plants. Unable to measure the microclimate with field data, we applied ENVI-MET software (ENVI-MET 4.0, sourced from ENVI_met GmbH, Essen, Germany) for microclimate simulation in the micro-scale green space.
ENVI-MET is holistic, three-dimensional and non-hydrostatic modeling software, developed by the research team of Professor Michael Bruse at Johannes Gutenberg University Mainz [33]. Based on computational fluid dynamics, ENVI-MET is often used to simulate urban microclimate environments and evaluate the effects of small-scale variations in urban design (e.g., atmosphere, vegetation, architecture, and materials) on the microclimate [34][35][36][37][38][39][40]. ENVI-MET has some functions in favor of phenological research. Firstly, the 3D modeling can simulate meteorological conditions at a specific height, a favorable point for the phenological research of tree species, whose reproductive phenology can be directly affected by the microclimate condition at canopy height via bud temperature perception [41]. Secondly, ENVI-MET can simulate multiple microclimate parameters, e.g., air temperature, convection, and wind, which can help explore the key microclimate factors that affect plant phenology. These characteristics provide evidence for its applicability in phenological research.
The microclimate simulation in ENVI-MET mainly includes the following several steps: (1) Create a Workspace in Manage Projects and Workspaces, (2) Construct a 3D model in SPACES, (3) Run the Simulation in ENVIGuide, (4) Extract data in LEONARDO.

3D Model Construction
The satellite images of the eight 100 m × 100 m buffer areas were extracted from the Worldview−3 image of Taoranting Park in February 2019 as the background image for model construction. A network of 50 grids (x) × 50 grids (y) × 30 grids (z) with a resolution ratio of 2 m/grid was constructed for each buffer area in ENVI-MET.
The basic model elements include architecture, vegetation, DEM (topography), soil, and surface (including water). The height of landscape elements (e.g., evergreen trees and architecture) was measured with Nikon Forestry Pro, i.e., laser rangefinder equipment. The height of the architecture and single trees was measured one by one, while the height of evergreen forest was measured using 3-5 sample trees at the forest edge surrounding the sampling point. The topography data were extracted from Google Earth Pro. The vegetation and architecture layer of the constructed models is shown in Figure 3, which can reflect the various landscape structures of the underlying surface of the eight buffer areas.

Microclimate Simulation
Simulation time. The number of days for microclimate simulation varied between different studies, ranging from one day [42,43] to three days [44] or more [33,45]. Generally, the simulation of the microclimate for each hour would take 0.5-1.5 h in ENVI-MET (when running on equipment with a Core i7 processor and 8 Gb RAM), which varied depending on the complexity of the models and the running speed of the equipment. In this research, the ENVI-MET simulation for one-day microclimate of the eight models (eight buffer areas) would take more than 10 days with tests and final running time taken into account.
Spring phenology in temperate zones is affected by the microclimate of the development period prior to the phenophase occurrence [46]; therefore, the microclimate during this period was expected to be simulated in this research. Considering the high time consumption of ENVI-MET simulation for the eight buffer areas, we chose to simulate the microclimate condition on 25 February 2019, the start of the growing season (SGS) for P. tomentosa in Beijing. SGS was chosen for simulation considering that the above-threshold air temperature at the onset of the growing season could be an effective stimulus to initiating spring phenology, such as sprouting and flowering [46]. SGS is defined as the beginning of the first six consecutive days with local daily mean air temperature > 5 • C, the general threshold temperature for tree growth in temperate zones [47,48]. The date of SGS was calculated based on the dynamics of daily mean temperature in Beijing's early spring (shown in Appendix A, Figure A1). The simulation extended a period of 24 h from 0:00 to 24:00. Each simulation had been repeated twice before to ensure a steady running condition in the final round.
Initial meteorological condition preset. We preset the initial meteorological condition including wind, temperature, and humidity. We used min/max temperature bounds for forcing in ENVI-MET. The data for the meteorology preset came from the Hourly Observation  Table 1. Simulation. The microclimate simulation was run in ENVIGuide. The output data were stored in SIMX files.

Data Extraction and Calculation of Microclimate Parameters
Data extraction. The output of the ENVI-MET simulation provided hourly data of multiple microclimate parameters. In this research, the air temperature ( • C), heat transfer coefficient (m 2 /s), and wind speed (m/s) were selected as the microclimate factors to be analyzed. The heat transfer coefficient/heat exchange coefficient is a proportionality constant between the heat flux and the thermodynamic driving force for the flow of heat [49]. In thermodynamics, the heat transfer coefficient usually indicates the overall heat transfer rate, reflecting the efficiency and activeness of air convection, and is considerably affected by the pattern of air-flow and the geometry of solid space [49,50]. In this study, besides the two basic factors, air temperature and wind speed, the heat transfer coefficient was also used to explore the impact of convection efficiency on reproductive phenology.
Calculation of microclimate parameters. Considering that the reproductive phenology of sampling branchlets could be directly affected by the surrounding microclimate condition, the microclimate data of each sampling point was extracted from the central eight grids in the 4 m (x-axis) × 4 m (y-axis) × 2 m (h-axis) model space. The 4 m × 4 m horizontal area was constituted by four grids (25,25), (25,26), (26,25), (26,26) at each height, the vertical range of 2 m covered the height of 7 m and 9 m (the height of sampled branchlets). The values of the daily mean air temperature (DMT), daily mean heat transfer coefficient (DMH), and daily mean wind speed (DMW) for each grid were calculated with the hourly simulated data. Taking air temperature as an example, the DMT for each sampling point was calculated with Formula (1).
where DMT n refers to the DMT of sampling point n; x,y,h refer to the respective x-ordinate, y-ordinate, and h-ordinate of a grid in the model; (DMT n ) xyh refers to the DMT on the grid (x,y,h) of sampling point n.

The Accuracy Evaluation of ENVI-MET Microclimate Simulation
The measured and simulated air temperature 1.5 m above ground was used for the accuracy evaluation. The air temperature of 1.5 m above ground (T a ) is one of the standard indexes for the performance evaluation of ENVI-MET microclimate simulation [39,51,52]. The field data of air temperature at eight sampling points were simultaneously collected at 9 a.m., 11 a.m., and 13 p.m. on February 25, with TES−1314, a high-accuracy handheld digital hygrothermograph.
A suite of quantitative indices for model evaluation recommended by Willmott (1982) [53], was used to evaluate the model performance, including the root mean square error (RMSE), systematic errors (RMSE s ), unsystematic errors (RMSE u ), and the index of agreement (d). With a purpose to obtain an accurate simulation outcome, the magnitude of RMSE s should approach 0, the value of RMSE u should approach RMSE, and the value of d should approach 1. RMSE, RMSE s , RMSE u , and d were calculated with Formulas (2)-(8) [54,55].
where O i refers to the observed value, P i refers to the simulated value,P i refers to the fitted value from the linear regression equation between the observed value and simulated value, and O i refers to the average of observed values.

Pearson Correlation Analysis between Reproductive Phenology and Microclimate Factors
Pearson correlation between the microclimate parameters and reproductive phenology was analyzed to explore the key microclimate factors that had a significant influence on the spatial variation of studied phenology and to reveal the quantitative impact. The independent variables for Pearson correlation analysis included DMT (daily mean temperature), DMH (daily mean heat transfer coefficient) and DMW (Daily mean wind speed); the analyzed dependent variables included the phenology of the reproductive phenophases-BBB, BF, FF, FS, BSD, ESD, DSD.

Multiple Regression Analysis of Reproductive Phenology in Relation to Multiple Microclimate Factors
Multiple regression analysis was performed to explore the overall correlation pattern between the phenology of each reproductive phenophase and multiple microclimate factors (DMT, DMH, DMW). According to the primary multiple regression analysis, there was a high-degree multicollinearity between DMT and DMH with VIF (variance inflation) >10. In response to the problem of multicollinearity, we applied Ridge Regression in the multiple regression analysis [56,57].

Simulated Microclimate Conditions in Taoranting Park
With respect to the accuracy evaluation of the air temperature simulation in ENVI-MET, RMSE = 1.057 • C, RMSE s = 0.684 • C, RMSE u = 0.806 • C, d = 0.962. With RMSE u approaching RMSE and d approaching 1, the model performance can be evaluated as 'good', which means the ENVI-MET simulation could reflect the actual thermal environment in this research. The detailed data of the accuracy evaluation are shown in Appendix A, Table A1 and Figure A2.
Among the eight sampling points, DMT (daily mean temperature) ranged from 4.578 to 5.071 • C, SD = 0.147 • C; DMH (daily mean heat transfer coefficient) ranged from 0.440 to 1.403 m 2 /s, SD = 0.300 m 2 /s; DMW (daily mean wind speed) ranged from 0.017 to 0.160 m/s, SD = 0.079 m/s. A significant correlation was found between DMT and DMH (r = −0.920, p <0.01), which means a higher air temperature was associated with a lower convection rate. The simulated data of the microclimate factors are shown in Appendix A, Table A2.

Spatial Variation of the Reproductive Phenology of Female P. tomentosa in Taoranting Park
The multivariate tests (MANOVA) showed that there was a statistically significant difference in the phenology of analyzed reproductive phenophases among the eight sampling points (p < 0.01). As seen in Figure 4, the phenology of BBB showed the most significant spatial difference, and the between-point phenological variation in flowering phenophases (BF, FF) was more significant than that in the fruiting and seed dispersal phenophases (FS, BSD).
In the micro-scale green space, the entire reproduction process of female P. tomentosa occurred from Julian day 58 (27 February)-Julian day 106 (16 April), from the earliest flower bud breaking to the latest time of seed dispersal. The beginning of bud break (BBB) ranged from Julian day 58 to day 63, the beginning of flowering (BF) ranged from day 65 to day 71, full flowering (FF) ranged from day 67 to day 72, fruit set visible (FS) ranged from day 70 to day 73, beginning of seed dispersal (BSD) ranged from day 93 to day 96, end of seed dispersal (ESD) ranged from day 102 to day 106, and duration of seed dispersal (DSD) ranged from 6 days to 12 days. A significant positive correlation was found between BBB, BF, FF, FS, BSD, the phenophases at flowering and fruiting stage (0.781 ≤ r ≤ 0.986, p < 0.05), while they showed a significantly negative correlation with DSD (−0.930 ≤ r ≤ −0.840, p < 0.01). This contrast indicated that the seed dispersal of female P. tomentosa could be a different process inconsistent with the flowering and fruiting development. The raw phenology data are presented in Appendix A, Table A3.

Key Microclimate Factors Affecting the Reproductive Phenology of Female P. tomentosa in Taoranting Park
According to the Pearson correlation analysis between reproductive phenology and the daily mean value of three microclimate parameters associated with SGS, we found that the phenology of the reproductive phenophases was heavily dependent on air temperature and the heat transfer coefficient, while wind speed showed no significant correlation.

Air Temperature vs. Reproductive Phenology
As shown in Figure 5, the DMT (daily mean temperature) associated with SGS was significantly negatively correlated with the phenology of BBB, BF, FF, FS, BSD, which means a higher air temperature at the onset of the growing season could drive an earlier occurrence of flowering and fruiting in female P. tomentosa. For BBB, BF, FF, and FS that happened successively within 15 days from the beginning of reproduction, the correlation was extremely significant (−0.983 ≤ r ≤ −0.908, p < 0.01), while the phenology of BSD that happened over 20 days after the appearance of fruit set showed a weaker correlation with DMT (r = −0.754, p < 0.05). Besides, according to the linear regression analysis, the absolute slope of the regression decreased with the occurrence time of the phenophases, with a change rate of −12.64 day/ • C for BF to −5.23 days/ • C for BSD. This difference demonstrated that the 'driving effect' of the air temperature at the onset of the growing season could become weaker for those phenophases occurring at later reproduction stages.
With respect to ESD and DSD, the correlation pattern reversed. Their positive correlation with DMT (r = 0.749, p < 0.05; r = 0.946, p < 0.01) indicated a different impact mechanism behind the seed dispersal process.

Heat Transfer Coefficient vs. Reproductive Phenology
As shown in Figure 5, the DMH (daily mean heat transfer coefficient) associated with SGS was significantly positively correlated with BBB (r = 0.864, p < 0.01), BF (r = 0.769, p < 0.05), FF (r = 0.788, p < 0.05), and FS (r = 0.799, p < 0.05). The positive correlation means a lower air convection rate could advance the occurrence of flowering and fruiting phenophases, especially in the early development stage. According to the linear regression between phenology and DMH, the slope of the regression decreased with the phenophase occurrence time, from 5.11 day/(m 2 /s) for BF to 1.88 day/(m 2 /s) for BSD, which indicated a declining influence of the heat transfer coefficient at SGS on the late-occurring phenophases. In addition, similar to DMT, the correlation between DMH and phenology ceased to be negative with respect to DSD (r =−0.922, p < 0.01).
Compared with air temperature, the heat transfer coefficient had a less significant impact on the reproductive phenology, reflecting on a lower absolute r (correlation coefficient) for most analyzed phenophases.

Multiple Regression of Reproductive Phenology in Relation to Microclimate Factors
Based on Ridge Regression analysis, the regression models of the phenology for various reproductive phenophases (BBB, BF, FF, FS, BSD and DSD) are shown as follows.
where D refers to the phenology (Julian day) or duration (days) of phenophases; X DMT refers to the value of the daily mean temperature at SGS, X DMH refers to the value of the daily mean heat transfer coefficient at SGS, X DMW refers to the value of the daily mean wind speed at SGS; k refers to the ridge parameter; * refers to the factor that had a significant influence on the analyzed phenology in the ridge regression model (* p < 0.05, ** p < 0.01).
In general, the regression models for various phenophases showed a satisfactory fit (R 2 = 0.883, p < 0.05) especially for BBB and DSD, which means the three microclimate factors could explain more than 88% of the spatial variation in reproductive phenology. In the regression of flowering and fruiting phenophases, BBB, BF, FF, FS, and DMT had a significantly negative influence on the phenology, while DMH showed a weaker positive correlation. As regards to the regression of DSD, the correlation pattern shifted, where DMT exerted a significantly positive impact on seed dispersal duration while DMH showed a negative effect. H 1 was true based on the MANOVA results, i.e., that the phenology of the analyzed reproductive phenophases all showed a statistically significant difference among the eight sampling points (p < 0.05). This is consistent with the finding that a significant spatial unevenness of plant phenology could appear in a small-scale urban space [19] or natural space [58][59][60]. Affected by this spatial variation, the duration of seed dispersal for female P. tomentosa in the green space extended from the average of 8.8 days for an individual to 13 days in total, aggravating the threat of poplar fluff to public health, let alone the impact at the urban scale, which highlights the need for more micro-scale studies in this field.

Testing of Hypothesis 2: The Spatial Variation of Reproductive Phenology Had a Significant Correlation with at Least One of the Microclimate Factors (H 1 )
H 1 was true based on the results of the correlation and regression analyses, i.e., that the DMT and DMH at SGS were significantly correlated with the reproductive phenology of female P. tomentosa (p < 0.05). Besides, air temperature played a key role in the multiple regression of reproductive phenology for various phenophases, showing a more significant impact on phenology than other factors. This also illustrated that in the studied garden area, the primary influence on poplar's reproductive phenology was air temperature, and other microclimate factors might exert their effects by affecting or interacting with temperature. This significant correlation showed that a slight microclimate variation could lead to an obvious spatial unevenness of reproductive phenology in such a micro-scale green space.

Air Temperature
The significant linear correlation between the air temperature and phenology of the reproductive phenophases demonstrated the prominent influence of the micro-scale thermal environment on poplar's reproductive development in Beijing's early spring. This finding accords with the widely accepted conclusion that air temperature is the critical climate driver of spring flowering and sprouting phenology in temperate zones [60][61][62][63][64], including poplar's reproduction [65].
The simulated air temperature at SGS performed well in the correlation and regression analyses, but the limitation of the one-day simulated data should be mentioned as well. In this research, the R 2 (coefficient of determination) of the linear regression between air temperature and phenology decreased as the phenophases occurred later, which means the ability of the temperature data at SGS to explain the spatial variation in late-occurring phenophases (e.g., FS, BSD) declined. Some studies revealed that multi-day thermal accumulation, e.g., Growing Degree Days, is the decisive initiator of spring phenology and has been widely applied in phenology modeling [66,67]. From this perspective, the oneday simulated microclimate data may not be robust enough for phenology prediction, and multi-day data simulation or observation is needed in future research.

Heat Transfer Coefficient
In addition to air temperature, the heat transfer coefficient was also found to have a significant impact on the reproductive phenology of female poplars. In Beijing in late February, a much lower heat transfer coefficient in an urban microclimate usually indicates a less frequent air disturbance from the outer cold airflow (e.g., Skimming flow regime), which is conductive to the establishment of a stable flow pattern with independent circulatory vortexes and a steady thermal field [49] that can provide a favorable condition for reproductive development. As a micro-scale parameter, the role of the heat transfer coefficient in affecting phenological variation was rarely explored before and deserves more attention.

Wind Speed
As an important impact factor of air temperature [68,69] and the heat transfer coefficient [70] in urban microclimates, wind speed is expected to show a clear correlation with the studied phenology. However, in this research, the correlation between wind speed and the phenology of reproductive phenophases was not significant, even though wind speed had a weak positive impact on the flowering and fruiting phenophases and a negative influence on the duration of seed dispersal in the multiple regression models.
One possible reason for this discrepancy could be the poor simulation accuracy of wind speed in ENVI-MET. Due to the lack of measurement data for comparison and validation, the performance of wind simulation could not be evaluated in this research. A challenge in the wind simulation is due to the unique wind environment in the microscale green space. Different from open urban spaces characterized with strong wind fields (e.g., urban canyons) [71], the large area of evergreen woods in the research site could function as a wind obstacle, decreasing the turbulence of high-momentum fluid [72], forming small vortexes with changeable airflow [73]. Further, the weak wind field in Beijing city on the simulated day (1.89 m/s at 10 m height), the changeable wind pattern, and low wind speed could bring a large challenge to the wind simulation in ENVI-MET [74,75].
The unexpected correlation pattern between wind speed and reproductive phenology showed the limitation of ENVI-MET in wind environment simulation, highlighting the need for field data as a support.

Some Insights into the Alleviation of Catkin Fiber Pollution from Female Poplars
Based on the revealed impact of microclimate factors on the reproductive phenology of female P. tomentosa, some potential strategies to alleviate the catkin fiber pollution can be proposed.
(1) Shorten the seed-flying period by increasing air convection and decreasing spatial variation of the microclimate in the green space.
The phenological response of seed dispersal to microclimate factors showed that the duration of seed dispersal (DSD) was significantly correlated with air temperature (r = 0.946) and the heat transfer coefficient (r = −0.922). Considering the heat transfer coefficient, the indicator of air convection, has been proved to be positively correlated with the spatial openness [49], to improve the air convection by increasing spatial openness can be an effective way to shorten the duration of seed dispersal for each individual poplar. In addition, the decrease in microclimate variation in the green space can help standardize the thermal environment around target trees and reduce the spatial variation of reproductive phenology, so as to shorten the overall duration of seed dispersal. Of course, the implementation of this potential strategy can be subject to reality and a comprehensive trade-off consideration, which deserves further discussion.
(2) Time arrangement of flying seed control based on the phenological response to air temperature.
According to the linear regression, with an increase in the daily mean temperature at SGS of 1 • C, BSD could advance 5.23 days and DSD could extend over 10 days. Combining local meteorological data, the phenological response of seed dispersal to air temperature variation can provide a reference to the timing of flying-seed control in Beijing's different urban heat environments .

Research Prospect-Other Possible Influential Factors Besides the Microclimate?
As shown in Figure 4, among the eight sampling points, the female P. tomentosa at point 8 showed a significantly later phenology than at other points for most analyzed phenophases (BBB, BF, FF, FS, BSD), and the poplars at points 2 and 6 showed a significantly earlier phenology than that at other points. According to the landscape pattern of the eight buffer areas shown in Figure 3, point 8 was situated in a highly open space with a wide wind corridor, while points 2 and point 6 were highly enclosed by evergreen groves. This indicates a possible correlation between the spatial structure (e.g., spatial openness) and reproductive phenology. Besides, the spatial structure of the underlying surface has been proved to have a significant influence on the air temperature in urban microclimates [24,49,76]. Therefore, a possible impact mechanism of the spatial structuremicroclimate-reproductive phenology should be explored in follow-up research.

Conclusions
This research investigated the impact of microclimate factors on SGS (start of the growing season) on the spatial variation in the reproductive phenology of P. tomentosa in a micro-scale green space in Beijing. We found that the phenology of the flowering and fruiting phenophases of female poplars was significantly negatively correlated with DMT (daily mean temperature) and positively correlated with DMH (daily mean heat transfer coefficient), while the duration of seed dispersal was positively affected by DMT and negatively affected by DMH. Based on the findings, an increase in air convection with lower air temperature and a decrease in the spatial variation of microclimates in green spaces can be an effective way to shorten the seed-flying duration, so as to help tackle poplar's catkin fiber pollution in Beijing. The discovery of this research can help fill the knowledge gap in the impact mechanism of the microclimate with respect to plant phenology in a micro-scale urban green space. It can also provide some empirical guidance for the alleviation of catkin fiber pollution in Beijing and other countries facing a similar problem.
Meanwhile, some limitations of this research need to be noted. ENVI-MET microclimate simulation (especially thermal environment simulation) was suitable for this phenological study, but field data measurements are still strongly advocated for the data collection for microclimate conditions to ensure data accuracy. Subject to the management regulation of the Park Administrative Office, we were unable to collect microclimate data with field measurements in this research. Regarding this, we will try to improve the access to field data in our future work. Another limitation was that the one-day simulated microclimate was not robust enough for phenology prediction, which addressed the importance of multi-day data collection in follow-up work.