The 30-year impact of post-windthrow management on the forest regeneration process in northern Japan

We targeted the University of Tokyo Hokkaido Forest (UTHF) in Furano, Hokkaido, Japan (Fig. 1a; 43° 10′–21′ N, 142° 23′–41′ E). Various forest management experiments have been conducted and recorded for this area since 1958. The available aerial photos before and after windthrow and recent LiDAR data that can be found for that area enable this research. The total forest area is 22,717 ha, with a large elevational difference from 190 to 1459 m. From 2011 to 2020, the average temperature at the arboretum (230 m) was 6.6 °C, and the annual precipitation was nearly 1196 mm (The University of Tokyo Hokkaido Forest 2021). From the end of November to the beginning of April, the forest is covered by snow, and the maximum snow depth is 85.6 cm. Abies sachalinensis (F. Schmidt) Mast (Sakhalin fir) dominantly covers the UTHF site from elevations of 200 m to approximately 1200 (Jayathunga et al. 2018), and dwarf bamboo (Sasa senanensis (Franch. Et Sav.) Rehder and S. kurilensis (Rupr.) Makino et Shibata occupy the forest floor (Owari et al. 2011). A strong typhoon (Typhoon Thad) affected the UTHF on the 23rd of August, 1981. The wind reached 26.3 ms⁻¹ and caused severe windthrow within most of the forest area (Fig. 1b). After this typhoon, a recovery plan including 4 forms of post-windthrow management (Table 1) was set up and implemented in March 1986.

To reveal the combined influences of the post-windthrow management projects, pre-windthrow attributes and topographic features on forest structure recovery, we selected 48 vegetation plots (Fig. 1c and Table 1) with sizes of 50 m × 50 m based on a wind disturbance map (Fig. 1b) created in ArcGIS (ArcMap 10.7.1, Esri Japan Corp., Japan) by referring to management records from 1981 to 1986. All plots were focused on the seriously damaged area determined by assessing aerial photos taken in 1981. We also certified that no other major artificial or natural disturbances occurred in the study plots after the post-management project by checking the records from 1986 to 2015 provided by UTHF. The reference (RF) in our study was represented by a stand of forests that has undergone no major disturbances in the last 30 years. We considered this reference plot to represent the possible climax stage of the UTHF plots following a major wind disturbance event on a long-term scale.

In this study, we used aerial photos taken in 1977 to quantify the complexity of the canopy layers before the windthrow event and aerial photos taken in 2017 to quantify the complexity of the recent canopy layers 30 years following windthrow. Light detection and ranging (LiDAR) data (Table S1.1) collected from 2015 to 2019 were used to generate the leaf area index (LAI) data to determine the complexity of the current forest structure, including the shrub layers. Although a temporal difference was found between the acquisitions of the aerial photos taken in 2017 and the LiDAR data, we supposed that any influence on the results was negligible because (1) no major disturbance occurred in the study sites from 2015 to 2020 and (2) a 5-year gap is insufficient to produce any remarkable changes in hemiboreal forests. We collected aerial photographs (Table S1.2) targeting the study sites from the Ministry of Agriculture, Forestry and Fisheries in Japan (MAFF) and the Geospatial Information Authority of Japan (GIA). We quantified the average canopy tree height in 1977 as the pre-windthrow attributes and the average canopy tree height, number of canopy trees and forest cover ratio in 2017 as the forest structure 30 years after the management projects using Stereo Viewer Pro (PHOTEC Co., Ltd., Japan). Stereo Viewer Pro allows the user to perform stereoscopic measurements based on aerial photographs. It is an established method for measuring canopy tree height and forest cover ratio, and the validity of data using this method has been demonstrated in several studies (St-Onge et al. 2004; Korpela 2004; Korpela and Tokola 2006; Shimizu et al. 2014). In Stereo Viewer Pro, we measured the height of all visible vegetation in each plot (Fig. S1.1). Each measured vegetation was set as a 3D point with X, Y, and Z coordinates in Stereo Viewer Pro; these points could then be opened in ArcGIS or exported as a text file. The polygon of each crown was created in Stereo Viewer Pro and transferred into ArcGIS for the following analyses. In ArcGIS, we first merged the crown polygons to represent the forested area in each plot and used the field calculator function to calculate the average canopy tree height, number of canopy trees and forest cover ratio in each plot (Table 2). We confirmed the accuracy of tree height estimated by Stereo Viewer Pro by comparing with tree height estimated by LiDAR (Fig. S1.2). We found a strong consistency between them. Further, we compared the estimated canopy tree density with stem density derived from field census data, provided by UTHF, for 11 plots (Fig. S1.3). The stem density from remote sensing was roughly consistent with the density of stems with DBH ≥ 11 cm from the field data.

Furthermore, we aimed to compensate for the limitation whereby aerial photos might neglect trees in the intermediate or suppressed layers by estimating the LAI using LiDAR data. The LiDAR datasets used herein were acquired in 2015, 2017, 2018 and 2019 using an Optech Airborne Laser Terrain Mapper (ALTM) Orion M300 sensor (Teledyne Technologies, Waterloo, ON, Canada) and a GNSS Trimble 5700・R3 (Trimble Inc., USA) mounted on an AS350B helicopter (Eurocopter Group, USA) (Moe et al. 2020). Table 4 shows the flight parameters of the LiDAR campaigns each year. The LiDAR data were initially processed and delivered in LAS format by Hokkaido Aero Asahi Corp. (Japan). The accuracy of these LiDAR data was also verified by Jayathunga et al. (2018) and Moe et al. (2020). The ALS Forest function in LiDAR360 V4.1 (GreenValley International Corp., USA) enabled us to calculate the LAI in each plot from the LiDAR data provided by the UTHF. In this process, we first normalized the point cloud data by using the ‘Normalize by DEM’ function in LiDAR360. Then, we used the ‘ALS Forest’ function to calculate the LAI by testing several parameters, including the grid size, height break and leaf angle distribution. We finally chose the recommended value of 0.5 as the leaf angle distribution (Li et al. 2016; Jin et al. 2020) and 20 m as the grid size. As the maximum individual crown in our field measured by Stereo Viewer Pro was larger than 15 and less than 20 m, we chose 20 m instead of the recommended 15 m when running the function. The height break was a parameter that allowed us to avoid the influence of vegetation on the forest floor. The final value we chose for this parameter was 3 m, as we considered any vegetation lower than 3 m to be representative of the shrub layer when processing the aerial photos previously.

We performed a field survey on the 3rd and 4th of August 2021 to identify the dominant species in the canopy layer. We first visually separated the identified trees into 237 groups using aerial photos based on texture and color and labeled these groups manually. However, considering the time required for this visual identification process, we selected only 6 plots under each management scheme for conducting the species classification process. In this process, we selected one representative tree from each of the 237 groups and circled these trees in both the aerial photos and digital canopy height model (DCM; resolution of 0.1 m) provided by UTHF, thus collecting the crown images and the GNSS data. We brought the prepared data to the field to verify the species of the targeted trees using aerial photos and a handy GNSS unit (GPSMAP^® 62SCJ, Garmin Ltd.). Twenty-one species represented by 237 trees were recorded in the field (Table 3). The subsequent analyses were performed at the species level basically. But four species of Acer, two species of Betula, and two species of Salix were analyzed at the genus level because they sometimes appear to be the same in aerial photographs. We also classified the species into three regeneration types based on Hanada et al. (2006): pioneer species, gap species, and shade-tolerant species.

R statistical software (v. 4.0.3, R Core Team 2020) was used for the statistical analysis conducted in this study.

The PCA results (Fig. 2) revealed the forest structure of each study plot 30 years after the windthrow event. Only the first two axes were presented and interpreted due to these axes corresponding to a meaningful proportion (91.98%) of the explained variance. The first principal component (Comp. 1) was correlated with all indexes (average canopy tree height, number of canopy trees, forest cover ratio and LAI) informing the forest structure, whereas the second principal component (Comp. 2) was also correlated with all of these indexes except average canopy tree height. All reference forest (RF) plots were intensively plotted at the rightmost ordination and at the top of the y-axis. We considered higher average canopy tree height, forest cover ratio and LAI values and a lower number of canopy tree values to represent the RF. The windthrow sites mostly plot to the left of the RF. This implies that the windthrow sites correspond to lower canopy tree height and LAI values compared with the RF sites. The 95% confidence ellipse of NS was the nearest ellipse to that of RF. This result means that NS had the forest structure most similar to that of RF among all of the groups. On the other hand, SL_P(Pg) was the farthest from RF, showing a significantly different forest structure from that of RF.

The species composition at various post-windthrow management sites among the 6 analyzed groups is shown in Fig. S2.1. Both conifer species and broadleaf species covered nearly 50% of the canopy layer in the RF sites (Fig. 3a, b). Compared with the RF site, significant increases were found in conifer species in SL_P(As) and SL_P(Pg), while dramatic decreases in conifer species were found in SL and SL_S. No significant difference was observed between RF and NS in terms of the conifer or broadleaf cover ratio. On the other hand, shade-tolerant species dominated the canopy layer of RF, which was related to the low pioneer species cover ratio (Fig. 3c, d). Among RF, NS, SL_P(As) and SL_P(Pg), the coverage ratios of pioneer species were almost the same, while the coverage ratio of shade-tolerant species was slightly lower at the NS site than at the other three sites. At the SL site, the number of pioneer species was higher, while that of shade-tolerant species was significantly lower than the corresponding values in the R and NS plots. In contrast, shade-tolerant species covered only 25% of the canopy layer, and the coverage ratio of pioneer species significantly increased in SL_S compared with the other sites.

The NMDS results (Fig. 4) show the canopy-layer species composition at different post-windthrow management groups. As the stress value was 0.120, which was lower than 0.200, we assumed two dimensions would be acceptable when running the NMDS. The plots of the RF sites were located to the left of the x-axis, and the NS sites were adjacent to the x-axis. Although a significant difference in species composition was shown between the two groups, NS was more similar to RF than SL, SL_S or SL_P(Pg) (Table S2.1). The results also explained that shade-tolerant conifer species, such as P. jezoensis and A. sachalinensis, tended to occur more in RF (Fig. S2.1). SL_S was intensively plotted above the y-axis at the same location as the pioneer species Betula. SL occurred to the bottom left and was located between NS and SL_S. Although the species composition among NS, SL and SL_S varied from group to group, SL was more similar to NS and SL_P(As) but differed from SL_S and RF (Table S2.1). The coverage of broadleaf species, especially T. japonica, was higher in SL (Figs. 4, S2.1). SL_P (As) plots were separately plotted at the bottom left of the ordinate. A. sachalinensis appeared to occur in this area, as shown by the NMDS results. SL_P(Pg) was the only group located to the right of the x-axis, and P. glehnii tended to occur more frequently at this site.

Analyzing the canopy tree height derived in 2017 demonstrated that slope aspect significantly affected canopy tree height recovery (Table 5). The predicted average canopy tree heights on the south side of the study plot were significantly higher than those on the east side, as revealed by the post hoc test (Fig. 5, Table S4.1). We also found that forests with higher average canopy tree heights before the windthrow event had lower average canopy tree heights 30 years after the windthrow event (Table 5). However, as the CIs of the post-windthrow management and the slope angle included 0, our results cannot support any significant influence of the analyzed post-windthrow management schemes (NS and SL) or the slope angle on the average canopy tree height recovery after the windthrow event in the natural succession or salvaged sites.

The results of Model 2 suggested that post-windthrow management had a strong influence on the regeneration process related to the average canopy tree height (Table 6). The predicted values derived from the GLM and post hoc processes by estimating the marginal means (Fig. 6) showed no significant difference between NS and SL, while surprisingly higher stem densities were estimated by the GLM (Fig. 6b, Table S6.1) in SL_S, SL_P(As) and SL_P(Pg) compared with those of NS and SL. However, no clear impact was found on the average canopy tree height or LAI due to NS, SL, SL_S or SL_P(As) according to the GLM results (Fig. 6a, d). Although the estimated forest coverage ratios were the same in NS, SL and SL_S, these ratios were significantly lower in SL_P(As) and SL_P(Pg) (Fig. 6c, Table S7.1). The LAI values predicted by the GLM also substantially decreased in SL_P (Pg) compared with those derived for the other sites (Fig. 6d, Table S8.1).

At the same time, the number of canopy trees, forest cover ratio and LAI were also related to the post-windthrow management scheme (including scarification, planting and seeding). Our results revealed that the average canopy tree height was more likely to decrease at high elevations (Table 6). Similarly, the forest coverage ratio and LAI in 2017 were negatively influenced by elevation. However, although elevation was selected in the final model corresponding to the number of canopy trees, no significant influence was found, as the 95% confidence CI ranged from − 1.11E−06 to 1.44E−06. We concluded that the slope angle did not affect the average canopy tree height, forest cover ratio or LAI for the same reason, but the results implied that higher stem densities could be found more easily in flat landscapes than in steep landscapes (Table 6).

Our study suggested that the forest cover ratio underwent a full recovery approximately 30 years after the windthrow event. However, the LAI and average canopy tree height (Fig. 2) values of the disturbed areas were still significantly lower than those of the reference area. From these results, we believe that the current forests are still in the recovery process during which they are developing stems (Oliver 1980) even 30 years after the windthrow event. In other research (Götmark and Kiffer 2014) conducted in a hemiboreal zone in Sweden, the average recorded canopy tree height was nearly the same as that reported in this work even 40 years after a windthrow event. Although no significant differences were found in the coverage ratio of the functional types, the nonsalvaged area tended to have a higher coverage ratio of pioneers and broadleaved trees but a lower coverage ratio of shade-tolerant and coniferous trees than the reference (Fig. 3). In addition, we found a divergence of species composition between the nonsalvaged area and reference (Fig. 4). Conifers such as Picea spp. and A. sachalinensis usually dominate undisturbed forests in Hokkaido, northern Japan, while broadleaved trees such as Betula spp. become dominant in windthrow areas due to their fast growth rate in the gap environments (Oliver 1980). Furthermore, conifers could be more severely damaged by wind disturbance because conifers are more susceptible to wind disturbance than broadleaves (Dhubháin and Farrelly 2018). Most conifers might be destroyed during the wind disturbance and recover slowly due to the gap environments created by windthrow, which cause the divergence of species composition between the nonsalvaged and reference areas.

The forest structure recovery was not evidently changed after 30 years of salvage logging (Fig. 6), which was similar to several previous studies (Royo et al. 2010; Morimoto et al. 2011). However, we found a significant decrease in conifer species but an increase in pioneer species at the salvaged sites compared with the nonsalvaged area (Fig. 3). Similar results were also reported by Orczewska et al. (2019), who observed that the number and coverage of ancient forest indicator species decreased following salvage logging. Salvage logging often severely destroys residual vegetation, including tree saplings (Morimoto et al. 2011), which can largely affect the subsequent recovery process. Indeed, Kurahashi et al. (1986) reported that approximately 25% of saplings were destroyed by salvage logging after windthrow at our study site.

However, several geographic features and the canopy tree height before windthrow actually affect the forest regeneration rate. We found that the slope aspect affected the growth of individual trees in the naturally regenerating area (Table 5). The south-facing aspect, which corresponds to increased light availability in the Northern Hemisphere, facilitates the growth of light-demanding species in disturbed areas (Vodde et al. 2010). The slope aspect can also affect the abundance of advanced regeneration in forest floors; similarly, Owari (2013) showed that A. sachaliensis saplings were abundant on southwestern slopes. Regeneration from saplings enables fast recovery of canopy height after wind disturbance, which could cause greater heights on south-facing slopes (Fig. 5). Limited resources and low temperatures at high elevations (Coomes and Allen 2007; Potapov et al. 2020) can be the main reasons for the slow regeneration process observed at high elevations, as seed erosion occurring on steep slopes (Barij et al. 2007) accounted for a lower number of canopy trees compared with flat areas (Table 6). Furthermore, we found that the canopy tree height in 1977 had a negative influence on the canopy tree height in 2017 (Table 5). This result indicates that the average post-windthrow canopy heights tended to be higher in the forests where the average canopy tree height was lower before the windthrow event. As the windthrow risk increases with the growth of the canopy tree height (Oliver 1980; Dhubháin and Farrelly 2018), the forests in which canopy heights were lower have the advantage of high wind resistance. The stands where more surviving trees remain due to the lower windthrow risk will attain a rapid recovery of average canopy tree height.

Clear increases in the coverage and abundance of pioneer species, such as Betula spp., were found in the sites that underwent scarification (Figs. 3, 4, S2.1). This is a natural result of scarification, which thoroughly removes residual vegetation, including dwarf bamboo and tree saplings, and thus creates homogeneous open sites. Betula spp. thus occur frequently, dominating secondary forests and forming almost pure stands following scarification due to their high growth rates and reproductive abilities (Yamazaki and Yoshida 2020). A clear increase in pioneer species such as Betula spp. was also found in the sites with Q. crispula seeding, similar to the sites with scarification (Figs. 3, 4). This indicates that the establishment of Q. crispula after scarification was clearly unsuccessful at our study site. The capacity of oak to become established might be reduced because scarification creates an increased light intensity and dry conditions (Wetzel and Burgess 2001; Asada et al. 2017), while Betula spp. gain an advantage as light-demanding species. Although previous research mentioned that scarification contributed to the successful establishment of oak following direct seeding (Birkedal et al. 2010), seeding Q. crispula after scarification might be unsuitable at our study site.

Planting directly can lead to an identically high number of canopy trees recovering (Fig. 6b), while it might also hamper the natural regeneration process (Thorn et al. 2017). Planting can result in a huge decrease in species diversity (Newbold et al. 2015). Species composition of the canopy layer was shifted in the P. glehnii planting plots, which were mostly dominated by P. glehnii (Fig. 4) and the A. sachalinensis planting plots were dominated by planted A. sachalinensis, whereas the species composition of SL_P(As) was more similar to the reference and natural succession sites (Table S2.1). The coverage of unplanted species and LAI in the A. sachalinensis planting area also tended to be slightly higher than those in the P. glehnii planting area (Figs. 6d, S2.1). Okamura et al. (1995) reported that broadleaved species mostly regenerated on piles of debris, where no trees were planted, in the P. glehnii planting area, whereas many broadleaved species invaded among planted trees in the A. sachalinensis planting area, suggesting that P. glehnii plantations are exclusive against other species. Although structure indicators except the number of canopy trees attained similar levels to NS and SL in the A. sachalinensis planting plots, those in the P. glehnii planting plots were much lower than those in the other plots (Fig. 6a, c, d). This can be because of the slow growth rate of P. glehnii.

Among all geographic factors, elevation in response to a harsh environment negatively affected the recovery rate after scarification and planting after salvage logging (Table 6). However, unlike in the natural succession area, the slope aspect was not selected as a significant factor in Model 2 (Table 6). The possible reason for the difference is that the scarification has thoroughly removed residual vegetation, including advanced regeneration. Such homogeneous conditions created by scarification might make the effects of the slope aspect unclear.