**3. Results**

The distribution of the VPD values originating from 89 locations (4 × 22 positions in the four transects and the centre position) representing three seasons is illustrated by boxplots (Figure 2). The visually identified edges of the grove are also marked in the figures to identify the below-canopy area, the edges, and the open areas. In the figures, the left sides represent the colder sides and the right sides, the warmer ones, as determined by the irradiation and shadow patterns (cardinal directions).

During a 24-h period, di fferences between the grassland, edge, and below-canopy areas can be determined based on the vapour pressure deficit. The below-canopy VPD values were consistently lower than the edge or grassland VPD values. Except for the summer measurements, the below-canopy local average VPD values were lower than the threshold. The di fferences between the local averages

and the threshold were considerable in the grassland, mostly in July. The medians of the values were very low in May and October, whereas in July, the median was mostly above the threshold. In July, the median and mean values of the below-canopy positions were very similar to each other and to the threshold. These three derivatives did not differ considerably from one another during the summer measurements (Figure 2).

**Figure 2.** Boxplots of vapour pressure deficit (VPD) values from 24-h periods of three measurement campaigns according to the cardinal directions. Blue dashed vertical lines: visual tree edges; green horizontal lines: VPD threshold (1.2 kPa); red lines: local averages.

The pattern of micrometeorological parameters was determined by the exposure and the position (distance) to the grove in each season. As with temperature fluctuations, the range of VPD was significantly smaller within the grove than in the open area. In the edges, we observed a special behaviour of the VPD: in the warmer edge, which was more exposed to irradiation, the values were higher with a more abrupt rise than in the open area. On the colder side of the transects, this elevated VPD did not occur at the edges, and the deficit increased more smoothly with distance from the tree group.

The durations of the values being over the thresholds provide important information about the spatio-temporal pattern of the VPD. Figure 3 compares the DCs of the two ends, the edges, and the centre of the SE–NW (Southeast–Northwest) transect for the three measurement periods. Among the four transects, the difference in the effect of the exposure was the most pronounced in this one, but the behaviour of the other three transects was similar. In terms of the stress threshold (1.2 kPa) exceedance rates, the summer patterns differed significantly from the spring and autumn. The exceedance rates ranged from 29% to 41% in May, from 52% to 60% in July, and from 22% to 36% in October. In terms of VPD distribution, however, further groupings emerged. The distributions of the values measured in the middle of the grove showed relatively small seasonal differences, with longer but not stronger exceedance rates measured during the summer. The exceedance rates were 29% in May, 52% in July, and 29% in October. Although the exceedance rate for 1.2 kPa showed a small variation in each measurement period, the intensity of the exceedance already di ffered significantly between the individual measurement positions. The duration curves of both the end-of-transect and edge measurement series ran close to each other in each period, but their distributions di ffered significantly. The maxima of the end-of-transect and edge series for the spring and summer measurements were in the range of 8–11 kPa, and the others were in the range of 3–5 kPa. The spring and summer measurement series ran close together, except for the summer end-of-transect (open grassland) measurement series (Figure 3). The VPDs of the latter two series of measurements, on the other hand, were well above the others, with a more-than-45% exceedance rate with respect to 3.0 kPa, while the rates for the others were below 25%. Above 2.5 kPa, the variability of the exceedance rates began to increase in all transects; therefore, we examined the exceedance at 3.0 kPa as well (Figures 3 and 4b). The exceedance rates with respect to 3.0 kPa ranged from 1% to 25% in May, from 6% to 53% in July, and from 1% to 16% in October.

The exceedance rate of the autumn values with respect to the stress threshold did not di ffer significantly from that of the spring values of cloudy weather, but despite the sunny and warm daytime weather, the distributions above the critical limit were completely di fferent. The maximum values did not exceed 5 kPa, and they were only slightly higher than the values below canopy.

**Figure 3.** VPD duration curves for the SE–NW transect from three 24-h measurement periods. Blue dashed line indicates the 1.2 kPa physiological threshold; black dashed line indicates 3.0 kPa threshold, above which the exceedance rates significantly diversified.

The exceedance rate with respect to the threshold of 1.2 kPa calculated from the VPD DCs also showed seasonal variability (Figure 4a). In each of the three measurement periods, at least 20% of the values were above the threshold. In the spring and autumn measurements, the exceedance rate was 20–45%, while in the summer period, it was 48–62%. Spring and autumn did not di ffer significantly from each other. The below-canopy exceedance rate was balanced and tended to be low, but the exceedance rate varied more across open areas. We also found di fferences in the VPD exceedance rate between the opposite ends of the transects. Choosing 3.0 kPa as the threshold value for the exceedance rate, the curves show a stronger di fference between the areas in the opposite edges of the

grove (Figure 4b). On sunny days, on the colder side, the lower values were even more pronounced when the Sun was at a lower altitude (October) at noon, with, sometimes, a 0–10% duration of values being above 3.0 kPa, as opposed to a 25–30% duration on the warmer side (Figure 5).

**Figure 4.** Exceedance (duration) (%) for VPD values above 1.2 kPa (**a**) and VPD values above 3.0 kPa (**b**) from 24-h period of three measurement periods; the left sides of the positions are the "cold" ends (W, NW, N, and NE), the right sides are the "warm" ends (E, SE, S, and SW), and the blue arrows are the centres of the transects.

**Figure 5.** Exceedance rates for VPD values with respect to 1.2 kPa (top panels: (**<sup>a</sup>**,**b**)) and 3.0 kPa (lower panels: (**<sup>c</sup>**,**d**)) from 24-h period of three measurement periods; (**<sup>a</sup>**,**<sup>c</sup>**): edge positions; (**b**,**d**): grassland end positions.

There were no remarkable di fferences among (all of) the edges during the spring measurements. In the case of the exceedance rate for the VPD values with respect to 1.2 kPa (Figure 5a,b), in the edges, the highest values always occurred on the eastern and south-eastern side, as well as in the grassland end positions, except for in October. In the edges, an increasing tendency could be observed from west to south-east, followed by a remarkable decreasing trend from south to west, although these trends did not appear in the case of grassland end positions. The values in the grassland varied during the three measurement periods. However, in the case of the exceedance rate of the VPD values with respect to 3.0 kPa (Figure 5c,d), the tendency was quite di fferent in the edges. In the NW (Northwest), N (North), and NE (Northeast) edges, the exceedance rate was 0% with respect to 3.0 kPa. This tendency was also observed in May and July, but the fall was not as strong as in the October data.

Depending on the weather conditions, the spatial pattern of the exceedance rate (Figure 6) also varied. At the 3.0 kPa threshold, the microclimatic di fferences among the three seasons were striking; the high VPD displayed the areas with consistently highly stressful conditions. In each of the three measurement periods, the below-canopy positions had lower values, but the values of the edges and the surrounding grassland were varied. There was only a slight di fference between the open areas on the di fferent sides of the tree group (Figure 6a). In October (Figure 6c), the transition between the warmer (E–SE (East–Southeast)) and colder (W–NW (west– Northwest)) sides was more gradual than in July (Figure 6b). In both of the later cases (summer and autumn), the measurements were made under clear skies. In July, the opposite grassland sides of the grove were in sync, NE and SW had lower values, and SE and NW had higher values. On the other hand, in October, the warmer and colder zones expanded considerably, and the transition between the two zones was sharp. In October, the warmer grassland side was E–SW, and W–NE was notably colder. In both periods, the eastern and south-eastern open areas had the highest values.

The ordinations also showed the di fferent behaviour of the open areas of the SE–NW and E–W transects. During July, the open areas of these two transects were clearly distinct from each other, in contrast to the other two transects, where the confidence ellipses largely overlapped. The below-canopy positions were separated only in the SE–NW and E–W transects, while no outstanding di fferences were observed in the microclimatic conditions of the two other transects (Figure 7).

**Figure 6.** Spatial plots of exceedance rates (%) for VPD values with respect to 3.0 kPa from 24-h period of three measurement periods (coordinates refer to the Hungarian Unified National Projection System (m)); (**a**): May, (**b**): July, and (**c**): October. White dashed line: visual tree edge. Black dashed lines: the position and extent of individual shrubs and poplar sprouts.

*Festuca vaginata* (the only species in Group A) occurred both in the grasslands and in the grove along the transect in Figure 8. Except for a few grassland species that were found on both sides of the group of trees (Group B), there were three characteristic spatial groups of species: species occurring on the south-eastern side (C) or the north-western side (D) of the grassland and species occurring mainly or exclusively in the grove (E). According to the Zólyomi-indicator values, the plant species below the canopy had a notably higher moisture and lower temperature requirement than the species in the grassland. On the other hand, there was also a di fference between the species of the two grassland sides. Group C (species occurring on the south-eastern side of the grassland) had slightly lower moisture and higher temperature requirements than Group D (species occurring on the north-western side of the grassland).

**Figure 7.** Principal coordinate analysis (PCoA) ordination of VPD quantiles in each transect in July 2018; 95% confidence ellipses are drawn according to vegetation patches.

**Table 1.** Species groups of the SE–NW transect ordered according to Figure 8. (A) ubiquitous species; (B) species occurring in the grassland; (C) species occurring on the south-eastern side of the grassland; (D) species occurring on the north-western side of the grassland; (E) species occurring in the grove; (F) low-frequency species not marked in Figure 8.


**Figure 8.** Distribution of the species along the SE–NW (Southeast–Northwest) transect in July 2018, together with the Zólyomi ecological indicator values, which quantify the temperature (TZ) and moisture (WZ) optimums for the plant species. Horizontal bars indicate coexisting species groups: (A) ubiquitous species; (B) species occurring in the grassland; (C) species occurring on the south-eastern side of the grassland; (D) species occurring on the north-western side of the grassland; (E) species occurring in the grove. List of species is shown in Table 1.

The ordination of the microcoenological data shows a clear difference between the three vegetation patches (Figure 9). The two grassland sides were notably separated from each other, and the grove just slightly overlapped with the NW grassland area. This slight overlap also applied to the VPD distribution (Figure 7, lower-left panel), indicating a more gradual transition toward the grove. The SE grassland side separated from the grove, similarly to the vegetation data, and the small overlap between the two grassland sides referred to the microclimatic similarities at the transect ends, which could also be observed in Figure 8. However, the two grassland sides showed remarkable microcoenological differences, and this difference was also manifested in the microclimatic conditions.

**Figure 9.** PCoA ordination of microcoenological data from the SE–NW transect. Jaccard index was used on summer measurements. Ellipses representing 95% confidence are drawn according to vegetation patch.

## **4. Discussion**

Habitats sensitive to environmental changes, such as transition zones, have complex vegetation dynamics formed by the interaction of the biotic and abiotic factors of the two vegetation components [4]. This interaction is called an edge effect, which is manifested in the microclimatic and vegetation structural differences between the forest edge and the interior of the forest [5]. In the sandy forest-steppe habitat, the edge effect is caused by the different structural complexity of the surrounding grassland, with the outstandingly open, treeless, plain vegetation structure and low biomass [5,10,14].

To test some of the abiotic–biotic relationships for transition zones, we carried out microclimate measurements in a sandy forest-steppe habitat in Hungary. From the air temperature and air humidity measured, we calculated the vapour pressure deficit (VPD), which is an important limiting factor in plant growth, determining plant photosynthesis, because CO2 uptake is strongly limited due to stomatal closure to prevent water loss [5,22,24].

We found that during a 24-h period, microclimatic differences between grassland, edge, and below-canopy areas can be determined based on the VPD (Figure 2). The range of the VPD was significantly lower within the grove than in the open area. These results are in good agreemen<sup>t</sup> with previous temperature data of several other edge studies [4,5,10,13]. In the edges, we also observed a special behaviour of the VPD in that the values were higher with a more abrupt rise in the warmer edge than in the open area, which can be explained by the heat-reflecting properties of the sunny side of the tree group. However, on the colder side of the transects, in line with the shadow effect, this elevated VPD does not occur at the edges and the deficit increases with distance from the tree group. This effect was more pronounced due to the lower altitude of the Sun (October) in the eastern, south-eastern, and southern borders but also in relatively cloudy weather conditions (May).

However, the extreme values, such as the maximum and the minimum, often occur only for short periods. We introduced a new method of applying the duration curve, showing the distribution of a variable over a measurement period [34,35]. Thus, the duration of values being over a threshold provides important information about the spatio-temporal VPD pattern. In this case, the exceedance rate with respect to the critical value of 1.2 kPa calculated from the 24-h VPD DCs showed a stress effect [22,24]. The duration curve method is presented in Figure 3 for the SE–NW transect, where the difference in the effect of the exposure was the most pronounced, in contrast to that for the N–S orientation, which is referred to as the coolest–warmest gradient in other studies [4,13,20]. For example, a study [4] of the edges of an oak-chestnut forest in the U.S. showed a difference of more than 5 ◦C between the south and north edges at the same measurement time. Another study in a Douglas-fir forest in the U.S. [13] also observed a remarkable difference between the south and north edges in short-wave radiation and air humidity during a diurnal period. In our study, in each season, at least 20% of the values were above the threshold, indicating semi-desert conditions in this habitat [9,10]. This new method also showed an increase in the variability of the exceedance rates (Figure 3), thus highlighting the importance of examining significant differences between the parts of the study site with a higher threshold.

We also found seasonal differences among the VPD distributions: the exceedance rate was much lower in spring and autumn than in the summer measurements (Figures 4 and 6). Spring and autumn did not differ significantly from each other due to the cloudy weather conditions in May. We also found differences in the VPD exceedance rate between opposite ends of the transects, which is a clear indication of the stronger stress effects in the warmer grassland areas during autumn. Choosing 3.0 kPa as the threshold value for the exceedance rate, the curves show a stronger moderating effect of the grove: a 0–10% duration of values being above 3.0 kPa in the colder grassland areas as opposed to a 25–30% duration in the warmer areas in the autumn. In the case of the above-3.0 kPa rate, this significant difference between the opposite areas showed that the distribution of the values above the threshold was notably varied within a seemingly homogeneous grassland (Figures 3 and 5). Thus, our first hypothesis that the VPD-modifying effect of the grove would gradually decrease in all directions away from the edge was only partially confirmed because this modifying effect can be detected even at a

distance of 6–10 meters from the edge, which is slightly different from the results of the other edge studies. Considering the small size of the grove we examined, this effect was stronger than suggested in other studies [4,5,10,13]. Our second hypothesis, which says that the spatial microclimate patterns do not differ from season to season and only the intensity of the modifying effect changes should also be rejected since the seasonal differences were also manifest in the changes in the position of the edge effect.

In the edges of the grove, we found an interesting pattern in the VPD exceedance rate (Figure 5a,c). We observed an increasing tendency from west to south-east and then a remarkable decreasing trend from south to west. However, these trends did not appear in the case of the grassland end positions (Figure 5b,d). Thus, the cardinal and intercardinal directions clearly define the characteristics of stressful conditions in the edges of the forest patches. During the summer measurements, a sharp temperature difference developed at the southern, southeastern, and eastern edges at higher altitudes of the Sun, while no pronounced edge effect was detected in the north-eastern and south-western borders. Due to the exposure and the distance from the grove, the open areas were also influenced by other parameters such as the altitude and shadows of other groves and vegetation patches. Other studies [4,20] showed that due to the larger exposure to irradiation, the south edge should differ the most from the forest interior. In the U.S., oak-chestnut forests showed a difference of at least 5 ◦C between the south edge and the forest interior [4]. According to our measurements, this heat gradient did not develop in summer (Figure 6b), but it was visible in the SW–NE line in autumn (Figure 6c); thus, the southern areas were not persistently the warmest in each season. Above the 3.0 kPa threshold in October (Figure 5c), in the NW, N, and NE edges, the exceedance rate was 0%, which is a good indication of the shading effect due to the lower altitude of the Sun in October. Ordinations based on the distribution of the VPD values (Figure 7) also support the microclimate-modifying effect of the grove. This effect was present in all periods, but it was most pronounced during the summer measurements. We also found clear differences between the edges based on the cardinal and intercardinal directions with this analysis, which also partially refutes our first hypothesis stating that similar trends of VPD-modifying effect can be observed in all directions from the grove.

The patterns explored highlight the importance of fine-scale sampling and analysis. Owing to our microcoenological data, we also found spatial heterogeneity in plant species distributions (Figure 8) in the open sandy grassland plant associations (*Festucetum vaginatae*). Except for a few grassland species that were found on both sides of the group of trees, there were three characteristic spatial groups of species: species occurring on the south-eastern side or the north-western side of the grassland and species occurring in the grove. This indicated the microclimatic differences in the study site, mostly between the grove and the surrounding, opposite grassland areas. Some species within the association behaved as differential species and indicated well the microclimatic differences in a seemingly homogeneous grassland area with a small group of trees. This distribution of plant species is a good indication of the different environmental conditions underneath even such a small group of trees. In accordance with the spatial heterogeneity of plant species distributions in our study, previous studies also showed that the forest edge may have a higher species number than the surrounding open area or the forest interior; thus, the edge acts as a transition zone with higher diversity [5,10,12,14,42,43]. In our case, the shading effect of the tree group has significant influencing power, which can be observed in the surrounding grassland. Therefore, our third hypothesis that the coenological and indication structure of the herb layer is homogeneous in the open grassland must also be rejected.
