MATERIALS AND METHODS

Study area

The study area is located in northeastern China, in a temperate monsoon climate which is characterized by four distinct seasonal conditions. The annual mean temperature of the area is approximately 2.54oC. The average coldest monthly temperature is -18.42oC (January), and the average hottest monthly temperature is 20.54oC (July). The mean annual precipitation is 622 mm, the rainy season is from June to September.

Forest plot network

During the summer months of 2017 and 2018, an extensive network of forest plots was established. The network includes 397 circular field plots each covering an area of 0.1 ha (radius: 17.85 m; Figure 2). All the natural temperate forest types in the region are included. The plots are located between 40°47.134′ to 53°22.053′ N, and 120°3.546′ to 134°1.008′ E. The elevations of the plots range from 79 to 1,255 m. The distances between the individual plots range from 24 to 60 km.
Each circular plot was divided into four subplots in north-south and east-west directions. All trees with DBH ≥ 5 cm were assessed. The x/y coordinate, species, DBH, height, and crown width of each tree was recorded. 30,539 individual trees were recorded, belonging to 62 species, 33 genera, and 16 families.

Spatial scales

On the basis of the published vegetation regionalization map of China (Zhang 2007), the region in this study was divided into 3 zones, 9 areas, and 17 districts (Fig. 3). The species compositions and environmental conditions are similar within each habitat, but dissimilar among different habitats. Five different spatial scales were used: zones within the region (region-zone), areas within the zones (area-zone), districts within areas (district-area), plots within the districts (plot-district), and the within-plot scales. These divisions are effective in detecting the signatures of particular ecological processes at each spatial scale.

Beta Diversity Calculation

Following Legendre & De Cáceres (2013), the β-diversity was calculated as the total variance of the Hellinger-transformed community matrix, i.e. YHel, for each forest plot as follows:
BDTotal = Var (YHel) = SS Total/ (n −1)(1)
Where SSTotal indicates the sum of the squared deviations from the column means of the entire YHel matrix; and n is the number of subplots. Since the Hellinger distance was bounded between 0 and \(\sqrt{}2\), the BDTotal had varied between 0 and 1, which indicated that all of the subplots had identical composition and each subplot contained a unique set of species, respectively.

Null Models and Partitioning of Ecological Effects

Individual-based randomizations were used to partition the effects of the environmental filtering and spatial aggregation processes on community compositions. At the region-zone scale, all plots were pooled and the individuals were randomly shuffled 1,000 times. This process removed the effects of both the environmental filtering among the zones and the spatial aggregation within each zone. For each shuffling process, the number of individuals in each plot and the number of species, as well as the abundance of each species, were preserved. Then, the β-diversity of each plot was calculated after each shuffle. The mean value of the 1,000 iterations was used to represent the expected β-diversity at the regional level and referred to as\(\overset{\overline{}}{\beta}\)exp_region. In addition, in order to remove the effects of the spatial aggregation, the plots within each zone were pooled and the individuals were randomly shuffled for 1,000 times. For each shuffling process, the number of individuals within each plot and the number of species, as well as the abundance of each species within each zone, were preserved. The mean value of the β-diversity values calculated after 1,000 iterations was used to represent the expected β-diversity at the zone level, and referred to as\(\overset{\overline{}}{\beta}\)exp_zone. Subsequently, the effects of environmental filtering on each plot were calculated as\(\overset{\overline{}}{\beta}\)exp_region -\(\overset{\overline{}}{\beta}\)exp_zone. The mean value of \(\overset{\overline{}}{\beta}\)exp_region -\(\overset{\overline{}}{\beta}\)exp_zone for all the plots was used to represent the effects of the environmental filtering processes at the region-zone scale. Similarly, at the zone-area, area-district, and district-plot scales, the plots were pooled within 9 areas and 17 districts, respectively. Individuals were randomly shuffled 1,000 times within each area, district, and plot, respectively, on an independent basis. For each shuffling process, the number of individuals within each plot and the number of species, as well as the abundance of each species within each area, district, and plot, were preserved, respectively. The expected β-diversities at the area, district, and plot levels (i.e. \(\overset{\overline{}}{\beta}\)exp_area,\(\overset{\overline{}}{\beta}\)exp_district, and\(\overset{\overline{}}{\beta}\)exp_plot) were calculated as the mean values of the 1,000 iterations. The effects of the environmental filtering on each plot at the zone-area, area-district, and district-plot scales were calculated as\(\overset{\overline{}}{\beta}\)exp_zone -\(\overset{\overline{}}{\beta}\)exp_area,\(\overset{\overline{}}{\beta}\)exp_area -\(\overset{\overline{}}{\beta}\)exp_districtand\(\ \overset{\overline{}}{\beta}\)exp_district -\(\overset{\overline{}}{\beta}\)exp_plot. The mean values of \(\overset{\overline{}}{\beta}\)exp_zone -\(\overset{\overline{}}{\beta}\)exp_area,\(\overset{\overline{}}{\beta}\)exp_area -\(\overset{\overline{}}{\beta}\)exp_district, and\(\ \overset{\overline{}}{\beta}\)exp_district -\(\overset{\overline{}}{\beta}\)exp_plot of all the plots was used to represent the effects of the environmental filtering effects at the zone-area, area-district, and district-plot scales. We calculated the observed β-diversity of each forest plot. Then,\(\overset{\overline{}}{\beta}\)obs -\(\overset{\overline{}}{\beta}\)exp_plot was used to measure the effects of the spatial aggregation processes on each forest plot. The mean value of\(\overset{\overline{}}{\beta}\)obs -\(\overset{\overline{}}{\beta}\)exp_plot of all the examined plots was calculated to represent the effects of the spatial aggregation processes at the within-plot scale. A summary of the models used to measure the effects of environmental filtering and spatial aggregation are presented in Table 1.

Key environmental factors identification

A multiple stepwise regression method was used to identify the key factors of the environmental filtering processes. The candidate environmental factors included 21 climate variables and 5 plot attributes. The climate data were collected from WorldClim Version 2 (Fick & Hijmans 2017). The 5 plot attributes were obtained from field observations, including elevation, slope, aspect, soil depth, and litter thickness. The statistical information of these environmental variables is shown in Table S1.1 of Appendix S1. To mitigate the inherent collinearity among the climate variables, the 11 temperature variables and the 8 precipitation variables, respectively, were subjected to principle component analysis (PCA) based on their correlation matrixes. The first two axes were determined to account for over 92% and 95%, respectively, of the variances in the temperature and precipitation variables, as detailed in Tables S1.2 and S1.3 of Appendix S1.
In this study, all of the calculations and statistical analyses were conducted using the statistical program R 3.5.1.