Selection of tree species by principal component analysis for abandoned farmland in southeastern Horqin Sandy Land, China

Peng Meng ·Jing Liu ·Xuefeng Bai

Abstract With changes in global climate and land use, the area of desertified farmland in southeastern Horqin Sandy Land (HSL) has increased in recent years, and farmlands are being abandoned. These abandoned farmlands (AFs) negatively impact the local ecology. Therefore, the aim of the present study was to select suitable trees and shrubs for those AFs to prevent and control the desertification tendency. In this study, three AFs were fenced for 2 years, then 37 arbor and shrub species or varieties of 21 families were planted in the fenced AFs and grown for 10 years. The ecological adaptability of the species was evaluated and ranked using a principal component analysis. The results showed that the biodiversity of the AFs significantly improved after 2 years of fencing; the Shannon-Wiener index and species richness of perennial grasses and forbs were 1.45 and 3.6 times higher, respectively, than for the unfenced AF. Among all species planted in fenced AFs, nine tree species had positive comprehensive F (CF) values; Pinus sylvestris (Russian Shira steppe provenance), Populus alba ‘Berolinensis’ and Gleditsia triacanthos had CF greater than 1, and the first (PC1), second (PC2) and third (PC3) principal component values (F1, F2, F3) were all positive. Among the shrubs, only Lespedeza bicolor and Rosa xanthina f. normalis had CF greater than 0. All these results suggest that fencing improves biodiversity and that planting trees and shrubs that have higher CF values on the basis of fencing is an effective way to green and beautify AFs in HSL.

Keywords Horqin Sandy Land·Fenced abandoned farmland·Principal component analysis·Tree species selection

Introduction

Land desertification is not only a global threat (Uzuner and Dengiz 2020), but also an important factor leading to poverty and restricting the sustainable economic development, so controlling desertification has received more attention from ecologists and forest scientists (Su et al. 2007). The Three-North Shelter/Protective Forest Program in China, the largest and longest ecological restoration program, has lasted for more than 40 years, and the desertification trend in the southeastern Horqin Sandy Land (HSL) incorporated into this program has been controlled (Zhang et al. 2012), and better sand prevention and control methods have also been devised (Jiao and Xing 2004). Nevertheless, in recent years, the area of desertified farmlands, especially abandoned farmlands (AFs), has increased in southeastern HSL The main causes are the decline of original vegetation and farmland shelterbelt tree species (such asPinus sylvestrisvar.mongolica) (Zhu et al. 2008), and soil wind erosion after overgrazing (Yan et al. 2018). The AF is an important landscape type of HSL, and its soil characteristics are very similar to that of semi-fixed sandy land, resulting in low productivity and eventual abandonment (Wu et al. 2008). Longterm remote sensing data of groundwater level in southeastern HSL shows that, from the point of view of water balance, the area of farm land should be reduced and transformed into forest or shrub land (Zheng et al. 2012). Although some species have been introduced and cultivated in HSL, few were suitable for extension in AFs due to lack of longer-term and systematic evaluation (Wang et al. 2014; Chi 2017). Furthermore, poor nutrient status (Gao and Huang 2020) and intense human interference (Li et al. 2017) have made tree selection and afforestation in AFs more difficult.

Edges play an important role in species diversity and landscape ecology in the forest-grassland mosaic (Erd?s et al. 2019). Fencing can create edges, permit the selfrepairing capacities of natural systems to effectively reduce the intensity of human interference, and promote nutrient accumulation in the ecosystem (Golodets et al. 2009). Fencing has a positive effect on soil fertility in HSL, and the total phosphorus, total nitrogen, available phosphorus and organic matter in fenced areas are trending upward (Zhao et al. 2017). Since fencing for a certain time can improve soil nutrients and thus be beneficial for tree species selection and afforestation of AFs.

For selecting suitable tree species, a reasonable ecological evaluation index (EEI) system needs to be established. Sapling growth and survival are considered direct and necessary growth indices to determine plant suitability (Campbell and Waser 2007). As phenological indices, the overwintering dry shoot rate (DSR) and length of the growing period (LGP) can reflect cold or drought tolerance in winter and the growth potential of a certain species, respectively (Voltas et al. 2013; Hurdebise et al. 2019). For photosynthetic physiology, photosynthetic indices such as photosynthetic rate (Pn) and transpiration rate (Tr) can reflect drought resistance and productivity of plants, so they are often used for screening drought-tolerant plant varieties (Wang and Chen 2005; Liu et al. 2007). The application of chlorophyll (Chl)afluorescence parameters to ecological adaptability evaluation and species screening has also been reported (Zheng and Shangguan 2006; Qiu et al. 2012). But until now, studies on constructing an EEI system using indices extracted from all the above aspects for tree species selection are still rare.

Based on the above analysis, this study was designed to answer two questions: (1) Based on the present ecological degradation condition of the AFs, can the AFs be restored by short-term fencing to make them afforestation easier and more successful? (2) Can we select tree species suitable for fenced AFs in the southeastern HSL through the rational construction of an EEI system and appropriate evaluation methods?

Toward these goals, we fenced newly formed AFs for 2 years and measured richness and diversity to evaluate the impact of short-term fencing on the ecological environment of the AFs, since these metrics often serve as measures of the degree of restoration of degraded lands (Miao et al. 2015). Then we introduced and selected 37 tree species in the fenced AFs and constructed an EEI system that included 10 indices. After long-term monitoring (10 years) of the EEI system, a principal component analysis (PCA) was used to rank and select tree species to provide technical support for controlling desertification in the study area.

Materials and methods

Study area

The experiment was carried out in fenced AFs in southeastern HSL in Zhanggutai (42° 43′ N, 122° 29′ E, altitude 226.5 m a.s.l.), Zhangwu County, Liaoning Province, China. The average annual wind speed in this area is 3.9 m s-1, the maximum wind speed is 28 m s-1, the main wind directions are northwest (NW) and southwest (SW). Mean annual temperature is 6.7 °C, mean annual precipitation is 433 mm, mean pan evaporation is 1570 mm, and the ratio of evaporation to precipitation is 3.43-6.43 (Jiao 2015). The soil is mainly sandy soil with a small amount of chestnut, salinealkali and peat soil.

Fencing and vegetation investigation

In 2008, four newly formed adjacent AFs were selected; three were fenced with cement rods and barbed wire, and one was unfenced (check [CK]). The experimental AFs had been abandoned for 1 year and used to grow peanuts before being abandoned. One year later, six standard plots (30 × 30 m, two in one AF) were set in the fenced area, and 4 quadrats (1 m × 1 m) were set in each plot to obtain vegetation data, and the same number of plots and quadrats were set in the CK (Fig. 1). Three times a year for two consecutive years (2009-2010), plant species, species abundance (number of individuals of each species), species richness and diversity index were recorded, and all grass and forb species were classified by life form as annuals (annual species) or as perennials (biennial and perennial species).

Fig. 1 Location of study area and map of experimental layout. Numbers 1-3 indicate fenced AFs with an area of 0.5 hm2, and number 4 is unfenced AF (control) with an area of 1.5 hm2. Squares and solid squares in each experimental site represent vegetation survey standard plots and sample quadrats, respectively. The base map was redrawn based on the Fifth National Desertification Monitoring Atlas of the State Forestry Administration, with pink representing desertified farmlands (2014)

The diversity index was calculated as:

whereSis the total number of species andPiis the relative frequency of the speciesI, ∑Pi= 1.

Cultivation of tree species

Between 2008 and 2010, tree species from China and elsewhere were introduced, and better-performing native tree species in southeastern HSL were also selected, and finally seeds or vegetative propagules of 37 species (or varieties) from 21 families were sown or cultured in the same nursery with the same water and fertilizer management (Table 1).

In spring 2011, saplings of the 37 species (or varieties) were cultivated independently in the three fenced AFs. To minimize any influence of random factors such as soil heterogeneity on the test results, 37 plots each AF were randomly set, and each plot included only one species with 30 seedlings planted at a 2 m × 1.5 m spacing. The natural growth environment was simulated by excluding any irrigation, fertilization or pest control during study period except for one watering on the day of planting. Since the first half of the experiment (2008-2014) was a period when the saplings were sensitive to local climate and soil factors, the relevant data during this period are shown in Table 2.

Table 1 List of experimental tree species

Table 2 Climate and soil factors in the study area

Construction of EEI system

The cultivated trees were all in the sapling stage, and wholeyear growth and phenological assessments were done from spring 2012 to autumn 2019. Three growth indices included survival rate (SR), sapling annual height growth (HG) and annual ground diameter growth (GDG). Phenological indices (2013-2015) included bud bursting period (BBP), leaf-expansion period, leaf-expansion peak period, shoot elongation cessation period (SECP), ground diameter growth cessation period (GDGCP), leaf fall period (only for deciduous tree species), LGP, and DSR. Among them, LGP for broad-leaved trees is the number of days between BBP and SECP, and for coniferous tree is the number of days between BBP and GDGCP.

Pn, Tr and Chl a fluorescence parameters for each species were determined using a Li-6400 photosynthetic apparatus (Li-Cor, Lincoln, NE, USA) and pocket plant efficiency analyzer (PEA) (Pocket PEA, Hansatech Instrument Ltd., Pentney, UK) at 9:30-10:30 h on three sunny days in July. Chlafluorescence parameters included minimal and maximal fluorescence yield in the dark-adapted state, maximum variable fluorescence, maximum quantum yield of primary PSII photochemistry (Fv/Fm), performance index on absorption basis (PIABS) (Stirbet et al. 2018), potential activity of PSII (Fv/Fo), density of active reaction centers on absorption basis.

On the basis of various reports (Wang and Chen 2005; Zheng and Shangguan 2006; Campbell and Waser 2007; Liu et al. 2007; Qiu et al. 2012; Voltas et al. 2013; Hurdebise et al. 2019), we selected 10 indices from the above measured indices to construct an EEI system used to evaluate and select species suitable for AF using a PCA: SR, HG, GDG, LGP, DSR, Pn, Tr,Fv/Fm, PIABSandFv/Fo.

Tree species selection based on PCA method

All the data for the IEE systems were checked, and the reverse data such as DSR and Tr were processed using “maximum value minus original value” method. These data were entered in the Descriptive Statistics module of SPSS V19.0 software (IBM, Armonk, NY, USA) and standardized. Standardized data for each index were represented byX1(SR),X2(HG),X3(GDG),X4(LGP),X5(DSR),X6(Pn),X7(Tr),X8(Fv/Fm),X9(PIABS) andX10(Fv/Fo). Pearson’s correlation matrix and PCA were carried out by running the Dimensionality Reduction module in SPSS 19.0 software, and the use of bootstrapping in this process allowed the estimation of 95% probability confidence intervals for the response function coefficients. According to the extracting principle of the principal component (PC), PCs with initial eigenvalue (IE) greater than 1 can be extracted. The initial factor load value (LV) of 10 indices in each PC can be obtained from the software, so the eigenvector value (A) corresponding to the 10 indices can also be calculated by formula (3). By multiplying the obtained eigenvectors with the standardized data of each index, the correspondingFvalue of each PC was obtained by formula (4). A proportion of IE corresponding to each PC to the sum of the total IEs of the PCs was taken as the weight to calculate the comprehensiveFvalue (CF) according to formula (5), and all species were sorted according to their CFs. The arbors and shrubs with bigger CF value could be selected out as suitable species for planting in AF.

wherenandbare the serial numbers of each PC and indicator, respectively.Xrepresents standardized data for each index.

Statistical analyses

All data were analyzed using SPSS V19.0 software, and data between the fenced AF and the control for two years after fencing were subjected to a one-way ANOVA and comparisons of ten indicators from 30 species for four to eight years after planting were subjected to Fisher’s least significant difference (LSD) multiple comparison test (p＜ 0.05).

Results

Vegetation change in AF

Ten grass and forb species were found to coexist in the fenced and unfenced AF (control, CK); most were annuals. In the fenced plots, more than 10 perennials appeared, such asLeymus chinensis,Calamagrostis epigeiosandPotentilla aiscolor, which significantly increased the species richness of perennials in the fenced area, 3.6 times that of CK. Although no significant difference in species richness of annuals were observed between fenced and unfenced treatment, anthropochory annuals (i.e.Digitaria sanguinalis) often appeared in unfenced AF. Furthermore, fencing resulted in a significant increase in the biodiversity index (Table 3) with λ and H′ values of 1.20 and 1.45 times those of CK, respectively. After 2 years of fencing, aforementioned tree species were planted in the fenced AFs.

Preliminary tree species selection

After another 2 years of preliminary cultivation, all saplings of 7 species had died, which wereHaloxylon ammodendron,Swida alba,Tamarix chinensis,Elaeagnus angustifolia,Amygdalus persica,Ulmus laevisandUlmus pumila‘Zuantian’, therefore, only the remaining 30 tree species were evaluated by monitoring IEE system.

Index correlation

Pearson correlation matrix of 10 indices obtained from 30 species was given in Table 4, and the data showed very significant positive correlation between the two growth indices (HG, GDG), and between the three chlorophyll fluorescence parameteres (Fv/Fm, PIABSandFv/Fo). Note that the significant positive correlation between growth indices SR, HG or GDG and Chlafluorescence variables,r∈ [0.293, 0.653], which indicated that fluorescence variables were sensitive and could be used for interspecific selection. In addition, Pn was positively correlated toFv/FmandFv/Fo.

Table 3 Indices for grass and forb species in fenced and unfenced AF for two years after fencing

Table 4 Pearson’s correlation coefficient (r) matrix of 10 indices obtained from 30 species by using bootstrapping with the estimation of 95%probability confidence intervals

When tree selection was based on the two growth indices as shown in Fig. 2a, the tree species with better performance wereGleditsia triacanthos,Populus alba’Berolinensis’ andLespedeza bicolor, but based on Pn andFv/Fm,Rosa xanthinaf.normalis,Lespedeza bicolorandPhiladelphus incanuswere better performers (Fig. 2b). These results showed that the selection based on different positive correlation index pairs led to inconsistent selection results. For overcoming the one-sidedness of single factor selection, PCA methods should be used for comprehensive evaluation.

Fig. 2 Scatter diagram of HG vs GDG (a: mean ± SE, n = 7 ～ 30) and Pn vs. Fv/Fm (b: mean ± SE, n = 10 for Pn, n = 5 for Fv/Fm) for each species. The negative HG was mainly due to overwintering dry shoot, and negative GDG (i.e., Sambucus williamsii) was mainly due to the determination of new tree stem from tillers after death of the determined tree stem last year. See abbreviations list at start of article

Tree species selection by PCA

As shown in Table 5, the IE of PC1 was 3.820, and its variance ratio was 39.197%; the IE of PC2 was 1.745, and its variance ratio was 17.449%; and the IE PC3 was 1.350, and its variance ratio was 13.498%. Therefore, the above three PCs should be extracted, and the cumulative variance of three PCs was 70.145%, which was representative.

The component matrix refers to the initial factor load matrix, and each load value (LV) represents the correlation coefficient between the PC and the corresponding index. As shown in Table 6, SR, GDG, LGP,Fv/Fm, PIABS,Fv/Fohad higher load (absolute value) on PC1, and Pn, Tr had higher load (absolute value) on PC2, and HG, DSR had higher load on PC3. The extraction of these three PCs reflected information of all the EEIs in this study, so these three new variables could replace the original 10 variables, thus achieving the purpose of dimension reduction. It could also be seen from the above data that, in growth indices, GDG was located in PC1, while HG was located in PC3, suggestting that GDG occupied the main position in the evaluation and screening of tree species. We also found that all Chlafluorescence indices were in PC1, while photosynthetic indices were in PC2, also indicating the importance of the Chlafluorescence index for evaluating tree species. The coefficients corresponding to each index in the three PCs were obtained by formula (3), those were the eigenvectoresA1,A2andA3(Table 7).

Table 5 Initial eigenvalues and variance proportions for the principal components (PCs) of the standardized data for the IEE systems used to evaluate and select species suitable for AFs

Table 6 Component matrix for the three components extracted according to initial eigenvalues and variance proportions showing the correlation coefficient between the PC and the corresponding index of the IEE systems

The correspondingFvalue of each PC and CF for each species was calculated (Table 8), and nine trees species had a CF greater than 0, indicating they were suitable for cultivating in the fenced AFs. Of these nine,Pinus sylvestris(Russian Shira steppe provenance),Populus alba‘Berolinensis’, andGleditsia triacanthos, had CF values ＞ 1, and three had PC values (F1,F2,F3) that were all positive, suggesting that they should be the most suitable species for growing and popularizing in HSL. Based on CF values, only two shrubs,Lespedeza bicolorandRosa xanthinaf.normalis, hadFvalues greater than 0 (Fig. 3e, f).

Table 7 Eigenvector calculated by formula (3) for the three components extracted

Discussion

Our study showed that, in the southeastern HSL, the species richness of perennials in AF obviously increased in the fenced area. Previous studies in HSL also led to similar conclusions that perennials can prevail in relatively stable ecosystems (Frank 1968; Katoh et al. 1998). Thus, fencing can improve ecosystem stability. However another study showed the opposite result; fencing led to a decrease in species richness, mainly because fencing led to the emergence of various trees such as willows, which inhibited the survival of many grass species (Takeuchi et al. 1995). But in our study area, some trees had been removed during the first few years of cultivation of the AFs (Fig. 3a).

AlthoughLeymus chinensiscan endure heavy grazing and extreme drought by increasing abscisic acid content and altering crucial functional traits (Yue et al. 2019; Zhang et al. 2020), it did not appear in the unfenced AF, indicating very serious land degradation. Annuals are widely adaptable, so not surprisingly, we found no significant difference in species richness of annuals between the fenced and unfenced AF.Digitaria sanguinalishas a transient surface seedbank (Oreja et al. 2020), so it always relies on unconscious spread by humans. The appearance ofD. sanguinalisindicated that the unfenced AF was seriously disturbed by human activities. The Simpson index (λ) and Shannon-Weiner index (H′) in the fenced AF increased significantly, indicating that the distribution of individuals among various species within the fenced community was more homogeneous and that each species had higher relative abundance (Bhattacharya et al. 2019).

In our study, 37 species of 21 families were planted in fenced AF, and seven species had died after 2 years. Among the remaining 30 species, 10 species had negative HG (Fig. 2a), indicating poor afforestation conditions in the study area. As shown in Table 2, the experimental forest received only 354 mm total precipitation in the year that it was planted, and the mean wind speeds in the early years after planting (2.3-2.9 m s-1) were nearly double those before planting, reflecting the harsh conditions. Some authors reported that conditions at their site were suitable forHaloxylon ammodendron(Wang et al. 2015), but they were not suitable in our study area, thus, its requirements need further study. Poor drought resistance ofSwida albawas considered to be the responsible for its previous introduction failure in HSL; it always died of physiological dehydration (Feng et al. 2013). Some studies also showed that the adaptability and growth performance of different varieties ofTamarix chinensis,Elaeagnus angustifoliaandAmygdalus persicawere not consistent when introduced in arid areas (Xun et al. 2007; Jiao et al. 2018); only a few varieties survived, so different provenances and varieties need to be tested in future studies. We also found that although the nativeUlmus pumilasurvived (Table 8),U. laevisandU. pumila‘Zuantian’ were not successful introductions.

Fig. 3 One of the new AFs with severe desertification a. Fenced AF after 1 year of afforestation. The degree of desertification has been reduced, and herbaceous vegetation was restored b. Several typical v

It is difficult to screen tree species simple using IEE indices because the results from selection based on different IEE indices are inconsistent. The use of PCA can overcome this shortfall by transforming multiple indices into a few independent indices reflecting the overall information. Using PCA method, Du et al. (2009) selectedPlatanus acerifolia,Salix babylonicaandFraxinus chinensisas most suitable as street trees in northern China, and Reubens et al. (2011) selectedCordia africanaandDodonaea angustifoliafor land rehabilitation in Ethiopia. Using PCA scores, we obtained a CF value of 1.715 forGleditsia triacanthos, which ranked third, while the CF forG. melanacanthawas negative, and it only ranked 17th (Table 8). Yang et al. (2008) suggested that the strong water absorption ability ofG. triacanthosled to its higher survival in arid sandy areas. We also found that the average Tr value forG. triacanthoswas only 1.12 mmol m-2s-1, and its average DSR value was 0, also indicating its strong drought and cold tolerance.G. triacanthosdoes not have spines, and its pods are edible and have medicinal value, so local residents are about to planting it (Cao et al. 2020). At the same time, the successful introduction ofG. triacanthosalso enriched the genetic resources of the localGleditsiaspecies. AlthoughG. triacanthoshas been reported as an invasive tree species after its introduction (Tognetti et al. 2019), the mean annual temperature range of the areas invaded is generally 13-19℃, and the annual precipitation is 650-1400 mm (Fernandez et al. 2017). Obviously, HSL does not have suitable conditions for the plant to be invasive. In fact,G. triacanthosdid not show any invasive characteristics in the study area.

Table 8 CF and F values of PCs calculated by formula (4) and (5) used to sort species suitable for AFs

In 1981, several provenances ofP. sylvestrisfrom Sweden, Finland and Austria were introduced into HSL, but their SR were less than 15% (You et al. 2007). Summer drought may account for the low SR, and Jorge et al. (2005) discussed the effect of summer drought onP. sylvestrisin the Mediterranean mountains (the southernmost distribution border of this species in Europe), and Irvine et al. (1998) explored the relationship between summer drought and the death ofP. sylvestrisby measuring the resistance of stem flow, and both studies reached similar conclusions. The poor introduction result for western and orthern European provenances suggests that the eastern European and northern Asian provenances, which are from physical settings similar to the studied introduction sites should be selected for further tests. HSL is considered to be in the forest-steppe transition zone, so we introducedP. sylvestrisfrom the Shira steppe in Russia and found that its SR was more than 90% after 10 years of planting and that it was healthy in the fenced AF in HSL.

We also found thatPopulus alba’Berolinensis’ was suitable for fenced AFs and had no pests or diseases. It also has strong vegetative propagation ability via tillering and cutting (Yang and Mo 2020), tissue culture (Qi et al. 2012), and it grew fast in our study (Figs. 2a and 3d) with SR above 90%.P. alba’Berolinensis’ is a hybrid, withP.×berolinensisas its male parent, and it is closely related toP.×xiaozhuanicaandP. simonii, which are native species of HSL. The introduction ofP. alba’Berolinensis’ enriched the genotype and increased genetic diversity of poplar in HSL. In addition,P. alba’Berolinensis’ is a male clone and does not produce catkins.

Some studies have shown that shrubs play a more direct effect in reducing desertification (Yan et al. 2015). Our PCA method showed that only two shrub species had CF values greater than 0:Lespedeza bicolorandRosa xanthinaf.normalis.Lespedeza bicoloris a leguminous tree species native to HSL, and its developed main roots and numerous root nodules on lateral roots fix nitrogen fixation and improve the soil.Leymus chinensissymbiotically enhances nitrogen fixation byLespedeza bicolorw (Xing et al., unpublished data). Stimulatory relationships between grasses and legumes have been reported in some plant combinations, such as maize and faba bean; yields of both increase when they are intercropped because root exudates of maize can promote faba bean nodulation and nitrogen fixation (Li et al. 2016). In our study, fencing promoted the emergence ofLeymus chinensis, which can form a good symbiotic relationship withLespedeza bicolor, and increase its SR to 95.7%, its average Pn to 14.37 μmol m-2s-1, placing it fourth among the studied species (Table 8, Fig. 2b), and decrease its DSR to 0.

Rosa xanthinaf.normalisis a typical spring flowering shrub 1 ～ 3 m tall. For the first time, this species was introduced into HSL, and it grew well in the fenced AFs (Fig. 3e). Its beautiful single-petal, yellow flowers and red fruits will beautify the sandy land and promote ecotourism. Previous authors have found a positive correlation betweenLespedeza bicolorandRosa xanthinaf.normalis, which can form stable communities (Yan et al. 2006). Our study also found that both had positive CF values in the fenced AF, suggesting that mixed planting of the two shrub species might be an effective model for sand control.

Conclusions

This study revealed that the biodiversity in the AF of HSL was significantly improved by fencing; the Shannon-Wiener index and perennial richness in the fenced AF was 1.45 and 3.6 times higher, respectively, than those of the CK. We also selected some excellent tree and shrub species such asPinus sylvestris(Russian Shira steppe provenance),Populus alba‘Berolinensis’,Gleditsia triacanthos,Lespedeza bicolorand

Rosa xanthinaf.normalisbased on the PCA and showed that the use of these improved varieties for afforestation is a feasible way to promote vegetation restoration in AF of HSL. A mixed forest generated in this way can effectively prevent erosion and control the sand. In addition, the evergreenP. sylvestris, straight and tallP. alba‘Berolinensis’, beautifully shapedG. triacanthos, and stunning yellow flowers orR. xanthinaf.normalishave important complementary roles in beautifying sandy areas for developing ecotourism.