Study Site

Allee Memorial Woods is an 80-ha preserved site containing old-growth areas (sensu Parker 1989) located (39.863°N, 87.280°W) near the boundary of the central hardwoods and forest–prairie ecoregions in west-central Indiana (Petty et al. 1961). Details about AMW physiography and history are reported in Sipe and Yamulla (2021). Climate data measured across 1991–2020 at the nearest National Weather Service station 46 km to the south of AMW show a mean annual temperature of 19.7 °C (monthly range 4.4–32.4 °C) and a mean total annual precipitation of 113.5 cm (monthly range 6.3–12.9 cm). The frost-free growing season averages approximately 188 days, April 19–October 13 (Indiana State Climate Office, 2022).

The A. triloba patches used in this study are located across three older AMW stands (A, B, C). Forest structure and composition measured during 2011–12 in permanent grids of 1.2 to 3.2 ha within the stands are described in detail in Sipe and Yamulla (2021). The overstories are dominated by varying proportions of Quercus alba L. (white oak), Quercus rubra L. (northern red oak), Quercus velutina Lam. (black oak), Liriodendron tulipifera L. (tulip poplar), Ac. saccharum L., and F. grandifolia Ehrh. Both the mid-sized (7.6–50.7 cm DBH) and sapling (1 m tall to 7.6 cm DBH, hereafter 1 m–7.6 cm) strata are dominated by Ac. saccharum and F. grandifolia. Very little canopy disturbance occurred in the three stands between 2012, when stand structure, composition, and understory PPFD outside A. triloba patches were measured, and 2016, when the patch PPFD patterns were quantified. Two other shrub species occur with varying frequency and abundance across AMW, Lindera benzoin L. (spicebush) and Dirca palustris L. (leatherwood). The herbaceous community is diverse and dense on the gorge slopes and bottoms, but considerably reduced in the uplands. Invasive herb and shrub species are relatively uncommon across the preserve, especially in the upland stands.

Patch Characteristics

Stands A, B, and C were surveyed to identify all A. triloba patches in or near the permanent study grids, and eight patches were selected for further study in undisturbed upland locations. The goal was to include intact patches showing wide variation in in foliar height, patch area, and ramet and foliar densities (T. Sipe in press). The eight patches have distinctive boundaries and are isolated from other patches. The patches show important differences relevant to their potential impacts on PPFD attenuation. Ramet density ranges from 1.7 to 6.7 stems m^–2, maximum foliar height from 0.83 to 1.88 m, LAI from 0.42 to 1.28 m² m^–2, and LAD from 0.53 to 1.12 m² m^–3. These variables do not change in parallel across patches, however, which produces a greater diversity of patch foliar structures and potential impacts on PPFD attenuation.

A 2-m-wide belt transect was positioned through the center of each patch parallel to its longest axis and divided into contiguous 2 ｘ 2 m quadrats placed 0–2, 2–4, 4–6 m, etc. on either side of the patch midline. Within each quadrat, the minimum and maximum foliar heights were measured and total leaf area (m²) per quadrat was estimated using leaf area scaling relationships developed previously on the basis of counting leaf clusters projected over each 2 ｘ 2 quadrat (T. Sipe in press). Leaf area index (m² m^–2) per quadrat was then calculated by dividing total leaf area by four and LAD (m² m^–3) was calculated by dividing total leaf area by the volume of space defined by the quadrat's 2 ｘ 2 m footprint and its maximum and minimum foliar heights.

Open-Site PPFD/Wind Speed and Patch PPFD Measurements

Photosynthetic photon flux density and wind speed were measured 2.5 m above ground in an open area outside AMW approximately 1 km from the A. triloba patch locations during May 27–July 16, 2016, spanning the duration of PPFD measurements in A. triloba patches. A LI-190 quantum sensor (Li-Cor, Lincoln, NE) and three-cup anemometer (model 12012, R.M. Young, Traverse City, MI) were connected to a data logger (model 23X, Campbell Scientific, Logan, UT) programmed to read every 10 sec and store 1-min averages continuously during all daylight hours.

Incident PPFD above the A. triloba canopy (3 m) and transmitted PPFD near the ground (0.25 m) were measured simultaneously using two LI-190 quantum sensors mounted on rods extending 0.25 m from each of eight 5-cm-diameter polyvinyl chloride (PVC) towers affixed vertically to steel stakes and distributed along the 2-m-wide belt transects in patches. The 0.25-m sampling height was used because pawpaw foliage often occurs down to this level. Herb, other shrub, and seedling foliage were sparse to absent within the eight patches and thus had little impact on PPFD at 0.25 m. The horizontal distances between sensor towers differed among patches since eight towers were used in all cases but belt transect lengths varied by threefold (8–24 m).

All sensors were calibrated against a factory-certified LI-190 sensor before the field sampling. The 16 sensors in each patch were connected to two data loggers (model 21X, Campbell Scientific) that were programmed to read and store PPFD every 6 sec during 14-hr days (0600–2000 Central Standard Time) centered approximately around solar noon for at least 10 days to span a range of weather conditions. Only two patches could be measured simultaneously because of limited numbers of sensors (32) and data loggers (4). Therefore four pairs of patches were measured during successive periods of 10–14 days between May 30 (Julian 151) and July 15 (Julian 197) in 2016.

The open-site and A. triloba patch data were screened to identify data gaps due to sensor or data-logger problems. Eight full 14-hr days of measurements were identified per patch pair for which open-site data were complete. One-way parametric ANOVA was used to determine whether the four patch-pair sampling periods differed significantly for open-site weather variables to guide sample day selection for A. triloba patch PPFD analyses. Daily mean PPFD (P = 0.012), mean wind speed (P = 0.018), and median wind speed (P = 0.032) differed significantly overall across the four sampling periods, whereas median PPFD did not (Kruskal–Wallis one-way ANOVA, P = 0.062). Tukey pairwise comparisons for the first three variables showed that the second period (June 11–20) was both brighter and windier compared with the fourth period (July 3–12), whereas the first (May 31–June 9) and third (June 22–July 1) periods were intermediate.

PPFD Analysis

Photosynthetic photon flux density means and medians are both reported here because of the highly skewed nature of understory irradiance regimes caused both by long periods of low PPFD in morning and evening and by sunflecks. Interpretations of PPFD regimes on the basis of means alone may overestimate photosyn- thetically usable photons while simultaneously underrepresenting the extent and photosynthetic impact of shaded conditions.

Photosynthetic photon flux density readings above and below the A. triloba canopy were not combined for any analysis. Unless otherwise noted, “patch” refers to A. triloba patches, and “canopy” refers to A. triloba patch foliage. Several additional terms are used throughout the paper for different hierarchical levels of data averaging. “Sensor” calculations denote a statistic for an individual sensor across an entire individual sample day (e.g., daily sensor mean or median). Patch values denote averages across all sensors at one of the measurement heights for either individual 6-sec readings or all-day sensor means. “Grand patch” values refer to statistics calculated across all sample days for a given patch, using the patch means or patch medians as inputs. Finally, the term “all-patch” is used for some analyses that involve descriptive statistics or inferential tests calculated across all patches for selected sample days. A “full sun” value of 2,000 lmol m^–2 sec^–1 has often been used as a referential constant for temperate latitudes, corresponding to incident PPFD in the open on a horizontal surface at solar noon on a clear day near summer solstice. Photosynthetic photon flux density values are sometimes expressed below as “percent full sun” using this 2,000 lmol m^–2 sec^–1 baseline. All statistical inference tests and percentile distributions were done with Sigmastat (v. 12.5, Systat, San Jose, CA) or SPSS (v. 27, IBM, Armonk, NY).

All Patches and Sample Days. For each patch, the grand patch mean, grand patch median, and their standard deviations were calculated separately above and below the canopy and presented in a single bar graph for comparison across the entire PPFD data set. Patch percent transmission values based on the patch means or medians were also calculated for each patch. All-patch descriptive statistics were then calculated across the eight patches for above-canopy PPFD, below-canopy PPFD, and percent transmission. Inference tests were not done for these variables to determine whether patches differed significantly since they were measured during periods with different sky and wind conditions. The data are presented because they collectively represent the strongest expression of the overall impact of pawpaw patches on near-ground PPFD levels across the study site, even though all eight patches were not sampled simultaneously.

All Patches on Similar Individual Days. Weather differences among the four sampling periods constrained the ability to compare all eight patches directly and to relate PPFD patterns to A. triloba LAI and LAD. Therefore scatterplots of open-site daily PPFD versus daily wind speed for all 32 sample days (4 periods ｘ 8 days/period) were examined to identify a set of days that would allow direct comparison of all patches. Four days (one per patch pair) with similar average light and wind conditions of intermediate magnitudes were identified (Table 1).

Table 1.

Daily statistics for open-site weather station photosynthetic photon flux density (PPFD; lmol m^–2 sec^–1) and wind speed (m/s) during (a) the 4 days used for direct comparison of PPFD patterns across all eight patches and (b) the days used to contrast PPFD patterns under sunny versus cloudy conditions for four patches.

These PPFD data were analyzed several ways. First, within each patch, the daily sensor means were used to determine through paired t tests or Wilcoxon signed-rank tests whether above-canopy means or medians differed significantly from their respective below-canopy values (N = 8 sensors per height). Second, individual patch means at the two measurement heights were used to test for the overall difference in above- versus below-canopy PPFD across all patches (N = 8 patches). A paired t test was used for patch means and a Wilcoxon signed-rank test for patch medians since the latter were nonnormal and heteroscedastic. Third, Kruskal–Wallis one-way ANOVA was used to determine whether the eight patches differed among themselves in patch means or medians above or below the canopy (four separate tests, N = 8). Dunn's test was used for pairwise patch differences if the overall Kruskal–Wallis result was significant. Fourth, within-patch spatial variation was explored by plotting means and medians both above and below the canopy for all sensor towers in each patch.

Attenuation versus LAI and LAD across All Patches. The relationship between foliage density and PPFD transmission was explored at the whole-patch scale, rather than the finer 2 ｘ 2 m quadrat scale, because of the complexity with which both direct and diffuse radiation interact with three-dimensional patch structure inside and outside the 2-m belt transect across the day. Patch percent transmission values calculated previously using patch means or medians were plotted against mean patch LAI and LAD values separately for all eight patches on the days with similar open-site PPFD and wind conditions. A linear regression was fit to each of the four scatterplots to determine whether there were predictable impacts of LAI or LAD on light attenuation.

Effects of Sunny versus Cloudy Skies on Patch Attenuation. Four days were identified that allowed direct comparison of the impacts of sunny versus cloudy conditions on PPFD patterns for four of the eight patches (A1, B8, C1, C7; Table 1). Sunny versus cloudy contrasts could not be made for the remaining four patches, since the second sampling period (patches B9, C10) did not include any sufficiently cloudy days, whereas the fourth period (patches C2, C3) did not include any sufficiently clear days. These sunny and cloudy designations are defined on the basis of mean daily PPFD and do not represent completely clear or overcast days, respectively.

The impacts of sky conditions were examined in numerous ways. First, the difference in average grand patch mean PPFD above versus below the A. triloba foliage was tested separately for sunny and cloudy days using paired t tests (N = 4), and the results were compared between sky conditions. Second, the difference in mean percent transmission between sunny and cloudy days was tested directly using paired t tests (N = 4). Third, within each patch, PPFD differences above versus below the A. triloba foliage were examined with a paired t test or Wilcoxon test (N = 8) and the results were compared between sky conditions.

Finally, diurnal PPFD patterns were generated by (a) converting 6-sec means across sensors to 30-min means above and below the canopy within each patch, (b) calculating mean percent transmission for each of these half-hour intervals, and (c) plotting the resulting values versus time of day.

Sunfleck Regimes. Sunflecks were quantified above and below the canopy for the same four patches used for sunny- versus cloudy-day comparisons above. Within each patch, three of the eight sensors were selected at each height for which their mean daily PPFD values were closest to (a) the overall mean at that height, (b) the mean + 1 SD, and (c) the mean – 1 SD. The goal was to distribute the sampling across most of the PPFD range at each height in a consistent way across patches. For one patch (B8), the sensor daily means above the canopy were unusually bimodal, with no single sensor sufficiently close to the patch mean (32.1 lmol m^–2 sec^–1). Therefore the closest sensors higher (55.0 lmol m^–2 sec^–1) and lower (16.3 lmol m^–2 sec^–1) than the mean were both included so their joint contribution to the overall patch mean would substitute for a single sensor. This yielded a total of 25 sensors for the sunfleck analyses (13 above, 12 below).

Sunflecks are usually identified as transient PPFD occurrences above a fixed threshold (e.g., 50 lmol m^–2 sec^–1). Defining the appropriate threshold is challenging given that the diffuse background at each understory location changes diurnally and can vary considerably within and across forest stands. The goal here was not to determine a “correct” sunfleck regime, but to contrast sunfleck patterns above versus below the canopy using appropriate fixed threshold(s) determined through careful examination of diurnal PPFD traces. Six-second PPFD readings were plotted versus time of day for each of the 25 sensors above and below the canopy, and the peak diffuse background at midday was estimated visually from the graph for each sensor in a manner similar to the procedure used by Brantley and Young (2009) to define thresholds for different plant communities. High and low estimates were averaged in a few cases where it was difficult to select a single value. The midday peak diffuse PPFD per sensor spanned 8.0–18.0 lmol m^–2 sec^–1 (overall mean 11.7, N = 13) above the canopy and 3.0–11.0 lmol m^–2 sec^–1 (mean 6.0, N = 12) below the canopy. A sunfleck threshold of 20 lmol m^–2 sec^–1 was then selected since it was sufficiently high for the brightest diffuse background above the canopy (18 lmol m^–2 sec^–1) and represents ∼ 1% full sun. Careful examination of the diurnal trace plots showed that a 20-lmol m^–2 sec^–1 threshold was a suitable compromise between a higher threshold, which would exclude too many distinctive fleck events, and a lower threshold, which would include too many minor PPFD transients above the diffuse background.

A sunfleck event was identified if 6-sec readings exceeded the 20-lmol m^–2 sec^–1 threshold for at least three measurements and ended when there were at least three successive readings below the threshold. Readings < 20 lmol m^–2 sec^–1 during the event were included so long as there were no more than two in a row.

Four variables were calculated for each fleck event, including the duration, mean PPFD, maximum PPFD, and fluence (mean PPFD ｘ duration). Descriptive statistics were then calculated for these same four variables across all fleck events for that sensor on that day. The total number of fleck events, total daily fluence (mean PPFD ｘ 14 hr ｘ3,600 sec/hr), total daily fleck duration (sum of all fleck-event durations), and total daily fleck fluence (sum of all fleck-event fluences) were also calculated for each sensor. Finally, the percentage of the day and the percentage of total fluence represented by all fleck events were calculated.

The analysis above was repeated for each of the 25 sensors. Means were calculated across the three (or four, for B8 above canopy) sensors per height per patch, yielding four data points above and below the patch canopies for statistical tests. Paired t tests (two-tailed, d.f. = 3) were used for each of the 10 variables above to determine whether and to what degree the A. triloba canopy significantly altered sunfleck regime components.

The 20-lmol m^–2 sec^–1 threshold is appropriate given the unusually low midday diffuse background peaks, and it is consistent with the PPFD patterns measured outside A. triloba patches across the site (Sipe and Yamulla 2021). However, it is nontrivially lower than the 50- to 100-lmol m^–2 sec^–1 range of values that have been used in some studies of forest understory sunfleck regimes (Nilsen et al. 2001; Brantley and Young 2009). To determine whether and how a higher threshold would affect the contrast in sunfleck regimes above versus below the A. triloba canopy, all the analyses described above were repeated for a fixed threshold of 40 lmol m^–2 sec^–1 (∼ 2% full sun).

PPFD Comparison Inside versus Outside Patches. A key focus of this study is whether A. triloba canopies reduce incident PPFD for tree seedlings to levels below that in the surrounding understory outside A. triloba patches. This question was addressed by comparing extensive PPFD measurements at a height of 0.5 m outside A. triloba patches across all three stands in 2012 with values measured at 0.25 m below the A. triloba canopy in 2016.

Details concerning the 2012 sampling are in Sipe and Yamulla (2021) and are summarized here. Nine sensor arrays, including three arrays in each stand (A, B, C), were placed in upland areas outside A. triloba patches with no canopy gap influence. Each array included eight LI-190 sensors mounted on PVC stakes at 0.5 m inside a 14-m radius circle attached to a data logger, with two sensors randomly positioned within each 90° quadrant. One array was set up in each stand and PPFD was measured simultaneously in the three stands for 14 consecutive days (May 30–June 12). Readings were taken every 6 sec and 1-min means were recorded across 14-hr days. The arrays were then moved to a second location in each stand and measured as before (June 15–28), and this was repeated for a third set of arrays (June 30–July 13). Thus the entire 2012 data set included nine arrays (designated A1–A3, B1–B3, and C1–C3) with eight sensors each, measured across a total of 42 days.

Photosynthetic photon flux density percentile distributions were generated for the 2012 array data using the 1-min mean data values. For each array, the mean across the eight sensors was calculated for each minute during each sample day (N = 840). These means were then collated across the 14 sample days (N = 11,760 total) and percentiles were calculated for each array from the collated data. To make the temporal resolutions comparable between A. triloba patches and arrays, the below-canopy 6-sec means across sensors from the patch measurements were condensed to 1-min means for each day (N = 840). These condensed sets were collated across sample days for each patch (N=6,720), and a percentile distribution was generated for each patch.

This percentile calculation process yielded similarly structured data sets for 1-min mean PPFD measurements in A. triloba patches and in the surrounding understory arrays across diverse sky and wind conditions. There were eight A. triloba patches with eight sensors each at 0.25 m measured over 8 days, nine arrays with eight sensors each measured at 0.5 m over 14 days, and the two data sets were measured on similar spatial scales (i.e., 14-m radius circular arrays versus 8- to 24-m patch transects) with the same instruments. Therefore the arrays and patch patterns were compared in two ways. First, PPFD percentile distributions were averaged across the nine arrays to produce a mean (± SD) percentile distribution for the arrays, and the eight A. triloba below-canopy distributions were averaged to produce a mean (± SD) distribution for the patches. Second, the 25th, 50th (median), 75th, 85th, and 95th percentiles were calculated for each data set, averaged (± SD) across all the arrays or patches, and graphed. Differences between the array and patch data were not tested statistically since they were measured in different years and at slightly different heights.

Patterns across All Patches and Sample Days

All-patch mean PPFD calculated across all sensors, patches, and sample days was reduced by over half from above the A. triloba canopy (22.0 lmol m^–2 sec^–1) to below the canopy (9.0 lmol m^–2 sec^–1) and the difference was significant (paired t test: t = 5.02, P = 0008, d.f. = 7) (Fig. 1). All-patch mean PPFD was approximately 1% full sun above the canopy and less than 0.5% below the canopy. All-patch median PPFD values were approximately half that of means, the medians differed significantly on average above (9.3 lmol m^–2 sec^–1) versus below (3.7 lmol m^–2 sec^–1) the canopy overall (paired t test: t = 9.96, P < 0.0001, d.f. = 7), and they showed a slightly lower overall transmission percentage (average 40%, range 27–53%) compared with PPFD means (average 44%, range 28–96%).

Fig. 1.

Grand patch mean and median photosynthetic photon flux density (PPFD) above and below the Asimina canopy across all sample days for each patch (+SD, N=7–8 sensors per patch). The pairs of numbers above each of the four sets of bars are the all-patch mean and SD across the patches (N=8) in that set. The italicized numbers are the all-patch mean and SD for percent transmission from above to below the canopy, calculated separately for patch means and patch medians. The bars represent (left to right) patches A1, B8, B9, C1, C2, C3, C7, and C10.

Individual patch means above the canopy varied by 2.3-fold, from 15.8 to 36.0 lmol m^–2 sec^–1, and were over twice the magnitude of patch means below the canopy, which showed even greater relative variation (5.2–15.2 lmol m^–2 sec^–1, 2.6-fold) (Fig. 1). Patch medians showed about the same relative variation among patches above (6.2–12.54 lmol m^–2 sec^–1, 2.0-fold) and below (2.8–5.0 lmol m^–2 sec^–1, 1.8-fold) the canopy.

All-Patch Comparison on Days with Similar Conditions

Average PPFD across all patches differed significantly above versus below the A. triloba canopy for both patch means (Wilcoxon signed-rank test: z = –2.521, d.f. = 7, P = 0.008) and patch medians (paired t test: d.f. = 7, P < 0.0001) (Fig. 2). The eight patches differed significantly from each other in above-canopy mean PPFD (Kruskal–Wallis H= 23.23, d.f. = 7, P = 0.002), below-canopy mean PPFD (H = 23.65, d.f. = 7, P < 0.001), above-canopy median PPFD (H = 41.18, d.f. = 7, P < 0.001), and below-canopy median PPFD (H = 31.15, d.f. = 7, P < 0.001).

Fig. 2.

Average (± 1 SE, N = 7–8 sensors per patch) patch mean (a) and patch median (b) photosynthetic photon flux density (PPFD) above and below the A. triloba canopy for all patches on selected days with similar mean daily open-site PPFD and wind speed. Percent transmission from above to below the canopy is also plotted. Note y-axis scale differences for means versus medians.

Patch means above versus below the canopy differed significantly within every individual patch (paired t tests: t = 4.3–10.0, P < 0.001 to 0.004, d.f. = 7) except C2 (Wilcoxon signed-rank test, z = –1.40, P = 0.195), whereas patch medians differed above versus below in every patch (paired t tests: t = 5.1–11.0, P < 0.001 to 0.008, d.f. = 7). The transmission percentage averaged 43% (range 25–91%) for means and 41% (range 31–51%) for medians (Fig. 2).

Patches differed substantially in the degree and pattern of spatial variation among sensor locations (Fig. 3). There was more spatial variation in absolute terms above the canopy than below, both within and across patches, and this contrast was stronger for PPFD means than medians. Below values paralleled the above values in many patches (e.g., B8) but not in others (e.g., B9). Only two patches (B8, C7) showed complete separation of the mean and median data ranges, with no crossovers of above versus below means for any location. In contrast, four of the eight sensor towers in C2 showed mean PPFD below the canopy that was equal to or greater than mean PPFD above, which is consistent with its unusually high percent transmission for means (Fig. 2a). Patch C3 showed the highest incident PPFD overall, but it varied substantially across sensor locations.

Fig. 3.

Spatial variation among sensors (N = 8) within patches for patch mean photosynthetic photon flux density (PPFD; open symbols) and patch median PPFD (closed symbols) above (solid lines) and below (dashed lines) the Asimina triloba canopy, calculated for sample days with similar open-site daily mean PPFD and wind speed. Note different y-axis scales for patches C2 and C3 compared with the other patches.

Attenuation versus Foliar Density

Scatterplots of percent transmission versus average LAI or LAD across the eight patches were noisy, with a distinctive outlier (patch C3) (Fig. 4). Linear regressions for percent transmission based on patch means were insignificant for both LAI (P = 0.340, R² = 0.152, d.f. = 1,7) and LAD (P = 0.699, R² = 0.027, d.f.= 1,7). Removing the C3 outlier improved the correlation for mean transmission, but the relationship was still insignificant (P = 0.133, R² = 0.391, d.f. = 1,7). Scatterplots based on patch medians were generally similar, except that C3 was no longer an outlier (Fig. 4b). The regression for percent transmission versus LAD was very insignificant (P = 0.828, R² = 0.009, d.f. = 1,7); however, percent transmission versus LAI showed a significant negative slope with a moderate correlation (P = 0.030, R² = 0.570, d.f. = 1,7).

Fig. 4.

Overall patch percent transmission plotted against patch leaf area index (LAI) or leaf area density (LAD) separately for patch means (a) and patch medians (b) on sample days with comparable open-site mean photosynthetic photon flux density and wind-speed conditions. Linear regressions were insignificant for all scatterplots except percent transmission of medians versus LAI in panel (b) (P = 0.030, R² = 0.57, d.f. = 1,7). The LAI and LAD values are nearly identical within two of the patches (overlapping symbols).

Effects of Sunny versus Cloudy Skies

Overall Patterns. Comparing the four patches on sunny versus cloudy days produced both expected and unexpected results. The contrast between sky conditions was substantial for mean PPFD values, but much less so for median PPFD (Fig. 5). The average of the four patch means above the canopy (25.1 lmol m^–2 sec^–1) was significantly greater than below the canopy (9.3 lmol m^–2 sec^–1) on clear days (paired t test: t = 4.26, d.f. = 3, P = 0.020), but there was no difference on cloudy days (8.0 versus 3.5 lmol m^–2 sec^–1; Wilcoxon signed-rank test: z = 1.83, d.f. = 3, P = 0.125). In contrast, the average of the four patch medians differed above versus below the canopy on both sunny days (paired t test: t = 7.77, d.f. = 3, P = 0.004) and cloudy days (paired t test: t= 3.19, d.f. = 3, P = 0.050).

Fig. 5.

Average (± 1 SE, N = 8 sensors per patch) mean (a) and median (b) photosynthetic photon flux density (PPFD) above and below the A. triloba canopy for four patches that could be contrasted on sunny days (left of the vertical dashed line) versus cloudy days (right of the dashed line). The percent transmission from above to below the canopy is also plotted for each patch.

Calculations of percentile distributions showed that 95% of the readings on cloudy days were < 20 lmol m^–2 sec^–1 above the canopy and < 10 lmol m^–2 sec^–1 below. Photosynthetic photon flux density means were higher on sunny days, as expected, but 65% of the readings above the canopy and 95% below were still < 10 lmol m^–2 sec^–1.

Percent transmission ranged from 20% to 60% across all patches and days and showed no consistent contrasts, either between sky conditions or between means and medians (Fig. 5). Mean percent transmission across the four patches was nearly the same on sunny days (41.5%) versus cloudy days (42.9%), with somewhat higher patch-to-patch variation for sunny skies (SD = 10.5% versus 8.7%, respectively) As a result, average percent transmission across patches did not differ significantly between sunny versus cloudy days for either mean PPFD (paired t test: t = 0.83, d.f. = 3, P = 0.467) or median PPFD (paired t test: t = 2.54, d.f. = 3, P = 0.085). The four patches showed a similar pattern in the rank order of their transmission percentages for both sky conditions and for both means and medians.

Diurnal Variation. Mean PPFD both above and below the A. triloba canopy showed substantial variation over time on sunny days, particularly during midday hours, due to scattered cloud occurrence or to heterogenous overstory alteration of incident PPFD on the A. triloba canopy (Fig. 6). Patch-level percent transmission also fluctuated considerably in patterns that differed among patches and sample days. Cloudy days contrasted dramatically in these patterns, with very strong parallels for above and below means and nearly constant percent transmission between 30% and 40%.

Fig. 6.

Examples of diurnal time courses on sunny versus cloudy days for mean photosynthetic photon flux density (PPFD) above the canopy (open circles), below the canopy (closed circles), and for percent transmission (triangles). The input data are based on sensor averages calculated across half-hourly intervals, and the plotted data points are patch means across the eight sensors at each measurement height. Panels (a) through (c) correspond to patch B8 and panels (d) through (f) to patch C1. For each patch, the upper graph is for the sunny day, the middle graph is for the cloudy day, and the third shows percent transmission on both days. The diurnal patterns for patches A1 and C7 were similar to those presented here for C1 and B8, respectively.

Sunfleck Regimes

A total of 727 sunfleck events were identified across the 25 sensor days using a threshold of 20 lmol m^–2 sec^–1, with more than twice as many (N= 522) above the canopy as below (N = 205). Locations below the canopy experience an average of 17 flecks per day that collectively represent 5.3% of the 14 daytime hours but 51% (0.253 mol day^–1) of the total daily fluence (Table 2). Statistics calculated across all recorded flecks below the canopy show that mean fleck duration is just over 2.5 min (155 sec), with an average of the mean PPFD within individual flecks of just 54.4 lmol m^–2 sec^–1 and an average of the maximum PPFD within individual flecks of 115 lmol m^–2 sec^–1. Frequency distributions for these three variables below the canopy are skewed toward shorter, dimmer flecks, however, with 75% lasting less than 3.1 min and corresponding mean and maximum PPFDs of 57.5 and 122 lmol m^–2 sec^–1, respectively.

Table 2.

Descriptive statistics and paired t test results for sunflecks above and below the A. triloba canopy in four patches (A1, B8, C1, C7) on individual sunny days using a threshold of 20 lmol m^–2 sec^–1. Each patch was represented by one input value, so N=4 and d.f.=3 in all cases. The eight input values were based on 727 total fleck events across the four patches (522 above, 205 below). Bold-faced P values are significant on the basis of two-tailed tests. One-tailed tests for mean duration, mean photosynthetic photon flux density (PPFD), and mean fluence of flecks produced the same outcomes. See Methods for explanations of data processing for individual fleck events and for aggregation to single patch-level input values used for calculations and tests below.

For fleck-event statistics, only the average maximum duration (P = 0.031, 1.6-fold) and average maximum PPFD (P = 0.021, 1.9-fold) differed between sample heights (Table 2, upper half). The averages for fleck duration, mean PPFD, and fluence were also greater above the canopy by 9, 16, and 64%, respectively, but these differences were not significant. The means for all daily total variables were significantly greater above the canopy, as expected (Table 2, lower half). This occurred despite the modest sample size (N = 4) because of the highly consistent pattern across individual patches. Every patch showed greater mean values above than below the canopy for all variables except mean fleck duration. The percentage of total daily fluence in flecks was 38% higher above the canopy, and the contrast for the remaining five variables ranged from 2.4-fold (mean number of flecks) to 3.8-fold (total fleck fluence). The same pattern of significant versus insignificant statistical test outcomes occurred for all 10 variables described above when the 40 lmol m^–2 sec^–1 sunfleck threshold was used instead of 20 mol m^–2 sec^–1 (data not shown).

PPFD Comparison Inside versus Outside A. triloba Patches. Mean PPFD below the A. triloba canopy was consistently lower by 31–45% than outside patches for both the mean and for all percentiles from the 25th to the 85th (Fig. 7). Only the 95th percentile appeared to show an insignificant difference, most likely due to the large variance around the below-canopy average. Mean PPFD was 0.45% full sun below the A. triloba foliage (9.0 lmol m^–2 sec^–1) compared with 0.65% outside patches (13.1 lmol m^–2 sec^–1), whereas medians were 0.25% (5.0 lmol m^–2 sec^–1) and 0.48% (9.5 lmol m^–2 sec^–1) full sun, respectively.

Fig. 7.

Mean (± 1 SE) photosynthetic photon flux density (PPFD) at 0.25 m below Asimina triloba patch canopies in 2016 versus comparable statistics for PPFD measured at 0.5 m in sensor arrays outside A. triloba patches across stands A, B, and C in 2012. Variables include the PPFD mean and values corresponding to the 25th, 50th (median), 75th, 85th, and 95th percentiles. N = 8 for patches and N = 9 for the arrays.

Overall PPFD Levels and Attenuation

Asimina triloba patches of the size and foliar density typical of this mature upland forest significantly and substantially reduced available light in the lower 3 m by 50–60% to levels below 1% or even 0.5% full sun. This strong impact occurs in nearly all patches, across sunny and cloudy days, and holds across numerous categories of analysis.

Mean percent transmission varied considerably among patches on similar days and within patches under sunny skies. However, fairly consistent background attenuation levels were discernible within patches, especially on cloudy days but also on sunny days, where a baseline transmission level was overlain by higher and lower increases in response to direct beam dynamics. The background attenuation levels presumably correspond to the interception of diffuse radiation on both types of days.

Regressions of patch percent transmission using PPFD means or medians versus LAI or LAD produced negative relationships overall, but only median percent transmission versus LAI was significant. An inverse relationship between LAI/ LAD and transmission is generally expected, but has not always been found since variation in leaf vertical distributions, orientations, and optical properties among taxa can influence interception by plant canopies with similar LAI (Pearcy et al. 2005). The relationship is usually stronger under the diffuse conditions of cloudy skies because clear skies generate a complex interaction between solar movement (azimuth and elevation), the resulting solar beam path angles and lengths through the heterogeneous plant community, and the three-dimensional foliar structure closer to the point(s) of PPFD measurement. The 4 days used for testing this relationship across A. triloba patches at the AMW study site were chosen because of their similar mean daily PPFD, but they had variable cloud cover within and across days. The 2-m-wide belt transects were large enough to quantify patch LAI and LAD well, given the moderate sizes of most of the patches and thus the fraction of total patch area represented by the transects (mean = 22%, range = 6–31%). However, foliage both elsewhere inside and outside the transect would influence PPFD in a specific 2 ｘ 2 m quadrat, particularly at the low 0.25-m measurement height. This lateral attenuation impact increases with patch stature and is more pronounced in morning or early evening because of lower solar path angles, especially under clear skies. A larger number of patches measured simultaneously under differing sky conditions would be required to determine with greater confidence the relationship between transmission and LAI/LAD for A. triloba patches.

Comparison with Other Studies of Shrub Impacts on Light

The few published papers reporting the effects of shrubs or small trees on understory PPFD patterns in temperate mesic forests differ considerably in their study systems, objectives, and methods, so direct comparison with A. triloba effects at AMW is challenging. The results of the present study are mostly applicable to undisturbed, mature forests on upland sites with moderate-quality soils. Asimina triloba can produce a much taller, denser, nearly monospecific stratum in moister, more fertile uplands, on lower slopes, or in bottomlands, particularly in sites that are younger successionally, have experienced canopy disturbance, or in which the mature overstory is dominated by less shade-tolerant species. This type of system is represented in controlled experiments by Cole and Weltzin (2005) on the competitive impacts of A. triloba on the invasive grass Microstegium vimineum (Trin.) A. Camus in a mixed-deciduous floodplain forest in eastern Tennessee. They measured PPFD with an integrating line sensor at 10 locations inside and outside each of three A. triloba patches on one clear day in mid-May to determine PPFD levels to use in the experiments. Photosynthetic photon flux density averaged 249 ± 132 lmol m^–2 sec^–1 outside patches and 15 ± 1 lmol m^–2 s^–1 inside patches, representing 6% transmission overall. The PPFD levels inside their patches are generally consistent with the AMW results, but their understory outside patches is brighter than at AMW by over an order of magnitude.

Nilsen et al. (2001) measured light and soil resources inside and outside areas with a 3- to 4-m-tall R. maximum L. subcanopy in a mature, mixed-deciduous forest dominated in the overstory by Q. rubra L. on a north-facing slope in western North Carolina. The motivation for their research was similar to that of the present study, which was to quantify the impact of the shrub canopy on juvenile tree performance. They measured PPFD at 0.5 m using fixed sensors over an extended period and with spatial arrays of eight sensors on single days. Total daily fluence inside and outside the R. maximum subcanopy averaged 0.729 and 0.141 mol m^–2, respectively, for an average transmission of 19.3%. Sunflecks, which they defined as PPFD occurrence above a threshold of 100 lmol m^–2 sec^–1, averaged 18.3/day (total duration ∼ 3 hr) outside R. maximum and just 0.8/day inside (total duration ∼ 8 min). Average maximum daily PPFD measured by the fixed sensors and recorded as 10-min means generally ranged from 10 to 15 lmol m^–2 sec^–1 outside R. maximum and 2 to 4 lmol m^–2 s^–1 inside during the growing season (∼ 24% transmission). No information was provided on the structure of the R. maximum subcanopy in their site, but their PPFD results suggest important differences compared with A. triloba patches at AMW. Average patch transmission levels (∼ 35–45%) were about twice as high compared with the R. maximum subcanopy (∼ 18–24%), and total daily fluence below the A. triloba patches on sunny days was approximately three times greater than under R. maximum (0.445 versus 0.141 mol day^–1). Lei et al. (2006) extended the analysis of light regimes in the same study site, particularly through hemispherical photo analysis across seasons. They confirmed the very low mean PPFD levels inside R. maximum thickets and stressed the year-round impact of the evergreen foliage on tree seedling carbon gain, which contrasts with the seasonal influence of deciduous A. triloba.

Aubin et al. (2000) measured percent open-site PPFD at several heights between 5 m and 0.05 m, calculated light extinction coefficients across the profile, and related extinction to vegetative cover by five shrub and herbaceous community components in six coniferous, deciduous, and mixed stands in the southern boreal forest of Quebec. The open-site PPFD averaged 15.3% (range 7.1–38.7) above the understory strata and 3.3% (range 0.8–6.6) below across the stands, both of which are considerably higher than for A. triloba patches at AMW. The most important connection between the studies is the impact of Acer spicatum Lam. (mountain maple) on light transmission in their system, which parallels A. triloba's role at AMW. Acer spicatum typically produces multiple shoots from the base and has limited clonality (by lower branch layering). However, its structure and ecology are otherwise similar to A. triloba in several other ways, including shrub/small-tree variability in growth form, stature, plagiotropic branching and leaf orientation, overall shade tolerance, the ability to form dense thickets, the ability to respond quickly to canopy gaps, and its preference for fertile soils on slopes or stream bottoms. Aubin et al. (2000) concluded that Ac. spicatum played the dominant role in lower-strata light extinction in three of their six forest types. This is linked to the impact of Ac. spicatum on forest regeneration (Bourgeois et al. 2004; Raymond et al. 2018), a characteristic it also shares with A. triloba (Baumer and Runkle 2010).

Potential Impacts on Juvenile Trees and Regeneration Processes

Forest composition measurements in AMW during 2012 included seedling surveys and showed that seedling population densities were extremely low. A census of all seedlings in four height classes up to 1.0 m tall in 130 1 ｘ 1 m plots distributed randomly across the three stands (24–72 per stand) showed only 1,088 total seedlings (83,692/ha), of which 76% were < 10 cm tall and only 4 were ≥ 25 cm. The populations were very young (mean = 2.7 yr) and very small (mean = 8.4 cm tall) on average. Increased deer browsing had affected the juvenile tree populations by 2012, but it is likely that increasing abundance of Ac. saccharum and F. grandifolia in the sapling and mid-sized strata was also playing a role by attenuating light (Sipe and Yamulla 2021).

Asimina triloba patches appear to reduce near-ground light levels even further compared with PPFD measured under variable sapling densities outside patches, and this holds true for nearly the entire PPFD range. The overall difference between patch and array grand means is only 4.1 lmol m^–2 sec^–1 (9.0 versus 13.1, 31%), but the effects of this apparently small absolute reduction on juvenile tree survival may be magnified at such low PPFD levels overall and depend on species-specific photosynthetic efficiencies below 0.5% full sun, the ability to use sunflecks, or both.

There are no data on comparative seedling abundance, growth, or survival inside versus outside the A. triloba patches at AMW. The potential impact of PPFD attenuation by A. triloba can be explored by comparing published data on leaf-level photosynthetic light compensation points (LCPs) with below-patch PPFD values at AMW. Light compensation point is only one of many variables that affect survival in low light, but its role in leaf carbon economy makes it a key reference point for interpreting the consequences of understory PPFD measurements. Light compensation point data collated by Walters and Reich (1999) from experimental studies include 17 temperate tree species of varying shade tolerance, of which 16 are native to North America. The average LCP across these taxa is 10.5 lmol m^–2 sec^–1 (SD = 7.4, range 1.1–28.2). Nine of these species, including representatives from across the entire shade tolerance spectrum, had LCPs lower than the below-canopy mean PPFD at AMW not only for all patches on all days (9.0 lmol m^–2 sec^–1) but also for four patches on sunny days (9.3 lmol m^–2 sec^–1). These mean daily PPFD values mask the differing impacts of diffuse background light and sunflecks, however. The former show midday peaks averaging just 6.0 lmol m^–2 sec^–1 (range 3.0–11.0) across four patches on sunny days, and the diffuse levels are considerably lower than this most of the time. Only five of the 17 species had LCPs below the 6.0 lmol m^–2 sec^–1 midday diffuse mean, and even these species would show negative net photosynthesis much of the day.

Thus one of the key impacts of A. triloba patches on juvenile tree performance is to increase the importance of species responses to sunflecks for net daily carbon gain, as is also the case for R. maximum thickets (Lei et al. 2006; Nilsen et al. 2009). It is unclear whether the more common modest flecks or the rarer, longer-duration brighter flecks under the A. triloba canopy would produce greater contributions to total daily carbon gain by juvenile trees. The answer depends on species-specific sunfleck induction/induction loss dynamics and on fully induced saturation irradiances (which set the upper limit on photosynthetic conversion during flecks) in the field, for which considerably more research like that reported by Nilsen et al. (2009) is needed.

Impact of A. triloba versus Other Tree Species

The increasing dominance of Ac. saccharum and F. grandifolia at AMW in the sapling and midsized tree strata over the last 50 yr represents an altered competitive environment not only for juvenile trees but for shrub and herb taxa as well, including A. triloba. Selective logging at AMW during 1890–1920 opened up stands A and C and likely favored some A. triloba expansion along with enhanced regeneration of many tree taxa. However, substantial losses of saplings of nearly all tree taxa have occurred since 1960 for many species in stands B and C during a time when both Ac. saccharum and F. grandifolia have increased their dominance (Sipe and Yamulla 2021). The small-tree losses include major reductions in three shade-tolerant understory species: Carpinus caroliniana Walt. (American hornbeam), Cornus florida L. (flowering dogwood), and Ostrya virginiana (Mill.) K. Koch. (ironwood). Some of the decline for Co. florida may be due to disease (dogwood anthracnose, Discula destructiva Redlin), but the simultaneous losses of Ca. caroliniana and O. virginiana suggest an impact of increasing shade caused by Ac. saccharum and F. grandifolia. There is no reason to suspect that A. triloba has not been affected as well.

Daily mean incident PPFD on the A. triloba canopies at 3 m is low overall, generally in the 10–40 lmol m^–2 sec^–1 range among patches and individual sample days. Grand patch means across all sample days above the canopy ranged from 15.8 to 23.3 lmol m^–2 sec^–1, with an outlier (patch C3) at 36.0 and an overall mean of 22.0 lmol m^–2 sec^–1. The corresponding values on days with similar sky/ wind conditions were nearly the same (range 15.0 to 25.5, overall mean 23.0). This is consistent with previous 2- to 7-m vertical PPFD profile measurements at AMW (Sipe and Yamulla 2021) and reflects substantial attenuation by the overstory trees, mid-sized trees, and larger saplings (> 2 m tall). Published data on A. triloba photosynthetic performance in the field includes the maximum photosynthetic rate but not full light response curves (Augspurger et al. 2005), so A. triloba LCPs are unknown. The greater number, duration, and fluence of sunflecks above the canopy may offset the low background PPFD levels somewhat. However, it is likely the A. triloba canopy itself is often light limited, and additional intraspecific self-shading may constrain total foliage production and LAI across much of the study site. This in turn will affect the impact of A. triloba PPFD attenuation on juvenile trees and regeneration patterns.