Modelling of root reinforcement and erosion control by ‘Veronese’ poplar on pastoral hill country in New Zealand
 M. Schwarz^{1, 2}Email author,
 C. Phillips^{3},
 M. Marden^{4},
 I. R. McIvor^{5},
 G. B. Douglas^{6} and
 A. Watson^{7}
DOI: 10.1186/s4049001600604
© Schwarz et al. 2016
Received: 19 April 2015
Accepted: 19 January 2016
Published: 3 February 2016
Abstract
Background
The control of erosion processes is an important issue worldwide. In New Zealand, previous studies have shown the benefits of reforestation or bioengineering measures to control erosion. The impetus for this work focuses on linking recent research to the needs of practitioners by formulating quantitative guidelines for planning and evaluation of ground bioengineering stabilisation measures.
Methods
Two root distribution datasets of ‘Veronese’ poplar (Populus deltoides x nigra) were used to calibrate a root distribution model for application on single root systems and to interacting root systems at the hillslope scale. The root distribution model results were then used for slope stability calculations in order to quantitatively evaluate the mechanical stabilisation effects of spaced trees on pastoral hillslopes.
Results
This study shows that root distribution data are important inputs for quantifying root reinforcement at the hillslope scale, and that root distribution strongly depends on local environmental conditions and on the tree planting density. The results also show that the combination of soil mechanical properties (soil angle of internal friction and cohesion) and topographic conditions (slope inclination) are the major parameters to define how much root reinforcement is needed to stabilise a specific slope, and thus the spacing of the trees to achieve this.
Conclusions
For the worst scenarios, effective root reinforcement (>2 kPa) is reached for tree spacing ranging from 2500 stems per hectare (sph) for 0.1 m stem diameter at breast height (DBH) to 300 sph for 0.3 m stem DBH. In ideal growing conditions, tree spacing less than 100 sph is sufficient for stem DBH greater than 0.15 m. New quantitative information gained from this study can provide a basis for evaluating planting strategies using poplar trees for erosion control on pastoral hill country in New Zealand.
Keywords
Shallow landslides Land use Root reinforcement BioengineeringBackground
Globally, many different forms of land use, driven from different socioeconomic needs, produce a variety of anthropised landscapes that in some cases lead to significant changes in geomorphological processes. While some traditional land uses reduce the intensity of processes such as floods or shallow landslides (Sidle and Ochiai, 2006), other types of land use lead to an acceleration of such processes (e.g. Phillips and Marden, 2005; Marden and Rowan 2015). Pastures established on steep (20–35°) hillslopes in New Zealand for livestock grazing is an example of land use which over a long period has caused major issues from a landscape management perspective (Blaschke et al. 1992). In particular, shallow landslides cause considerable losses of productive soil and represent a challenge for risk management in such systems (Heaphy et al. 2014). Furthermore, landslide scars may contribute to further erosion from other processes that may cause continuing on and offfarm effects including enhanced loss of soil productivity, water holding capacity, and decline in water quality of streams and rivers. It is estimated that for each highmagnitude storm event, with estimated return periods in excess of 20–50 years, regionally, up to 10 % of soil area on steep pastoral hillslopes may be lost and as high as 36 % lost on individual properties (Marden et al. 1995; Dymond et al. 2006; Rosser and Ross 2011). Studies have also shown that the recovery of soil productivity on shallow landslideeroded areas takes many years, with annual pasture production unlikely to attain more than 80 % of that of uneroded ground over many decades (Douglas et al. 1986; Lambert et al. 1984; Rosser and Ross 2011).
The benefits of reforestation or bioengineering measures to control erosion have also been demonstrated (e.g. Pearce et al. 1987; Phillips et al. 1990; Blaschke et al. 2008; McIvor et al. 2011; Marden 2012; Phillips et al. 2012). In New Zealand pastoral hill country, widespaced trees of poplar (Populus spp.) and willow (Salix spp.) have been the primary means of erosion control for 50+ years (Wilkinson, 1999). These species are established using 2.5 to 3.0mlong poles (vegetative cuttings), with plastic sleeve protectors, and planted on slopes at a spacing of 10–15 m depending on erosion potential and practitioner experience. These trees add value to hill pastoral systems through providing shade, shelter, quality fodder (summer/autumn), and carbon sequestration (Wall et al. 1997; Betteridge et al. 2012; Basher 2013), increasing their utilisation and further adoption by landowners. A mix of clones is usually planted, and they are managed for at least the first 5 years after planting. In some regions, treatment of erosionprone pastoral areas has shifted from being farmerled and voluntary to becoming a requirement promoted (and generally incentivised) or regulated by the local authorities.
Some knowledge and understanding of how widespaced trees contribute to slope stabilisation is provided by limited data on the root distribution of young (<12 years) poplar and willow trees on slopes (McIvor et al. 2008; McIvor et al. 2009; Douglas et al. 2010). However, although significant advances in understanding have been made in the last decade, and are being used to refine planting recommendations, a key gap is linking the mechanical quantification of root reinforcement with its effects on slope stability. The framework proposed by Schwarz et al. (2010a) provides an opportunity to upscale root distribution data to the stand scale, calculate root reinforcement, and thus discuss its effects on slope stability.

The model uses a strainstep loading approach to calculate the forcedisplacement behaviour of each single root of a bundle while considering the progressive failure of the roots as a function of displacement.

The model uses a Weibull survival function that describes the range of probability that a single root failed before or after the fitted value of maximum tensile force estimated based on its diameter class. This function considers the mechanical variability of root strength due to root material properties, root geometry, and soil mechanical conditions.
The RBMw allows the calculation of root reinforcement in terms of force and displacement for different root distributions which are spatially distributed at the hillslope scale. Such an approach allows an upscaling of root reinforcement in order to quantify the mechanism of slope stabilisation due to plant roots. Schwarz et al. (2010a) showed how root distribution is influenced by the forest structure and how it is possible to quantify root reinforcement as a function of forest structure (tree stem density and dimensions).
The impetus for this work focuses on linking research results to the needs of practitioners by providing supporting evidence to enable quantitative guidelines for planning and evaluation of ground bioengineering stabilisation measures to be formulated, as well as balancing the positive effect of slope stabilisation and carbon storage with the negative effect on pasture production. Available root distribution data provide the necessary foundation to calculate how factors such as location and morphology influence the stabilising effects of vegetation through root reinforcement. In this specific case, root distribution data and mechanical parameters of roots are combined to characterise the dynamic and the spatial distribution of root reinforcement at the slope scale. Root distribution data are rare and difficult to obtain. The comparison of data from two different datasets, as presented in this work, provides a rare opportunity to discuss the effect of environmental conditions and measuring methods on the results of root distribution.
The objectives of this paper are to characterise the lateral root distribution of poplar trees using datasets from two different locations in order to compare the influence of environmental conditions, calibrate the root distribution model from Schwarz et al. (2010a), and calculate the spatial distribution of root reinforcement on slopes with different tree spacing. Finally, it was also the aim to simulate the stabilisation effects of root reinforcement on slopes prone to failure in order to formulate general guidelines for planning and evaluation of ground bioengineering stabilisation measures.
Methods
General modelling process
Detailed root distribution datasets were used to calibrate the rootdistribution model proposed by Schwarz et al. (2010a) and then combined with the mechanical parameterisation of root reinforcement implemented using unpublished^{1} and published data for New Zealand poplars (McIvor et al. 2011). The results of the model simulations using different combinations of parameter values, tree density, and tree dimension were then used to quantify the minimum planting setup needed to assure minimum levels of lateral root reinforcement (2, 5, 10, and 15 kPa, respectively). Finally, a threedimensional forcebalance approach was used to calculate the lateral root reinforcement needed for hillslope stabilisation as a function of slope inclination and mechanical properties of soil, as illustrated by Schwarz et al. (2010a). The details of the applied methods are explained in the following sections.
Study area and data collection

The Gisborne Dataset, G, is from measurements of two single, but complete, root systems of poplar (Populus deltoides Marshall x P. nigra L. clone ‘Veronese’) trees grown from poles: one tree aged 1 year and one tree aged 2 years. Trees were part of a field trial established in September 2009 near Gisborne on the east coast of the North Island (see Phillips et al. 2014 for details). The trial was established on a lowlying, evensurfaced alluvial flat terrace adjacent to the Taraheru River, at the site of previous ‘plant growth performance’ trials to allow comparisons between species (e.g. Marden et al., 2005). The soil is a free draining ‘Te Hapara’ Typic Sandy Brown Soil (Hewitt, 1998). The site (50 m by 50 m) was tilled before planting, weed mat laid to reduce competition from weeds, and trickle irrigation installed (to deal with summer dry periods). Poles (3 m unrooted stem cuttings approximately 50–60 mm in diameter at the base) were pushed/rammed vertically into the soil to 0.7 m depth on a 5 m × 5 m grid. Root distribution was measured by the complete excavation of the root systems. The number and the dimension of roots were measured along regularly spaced concentric circles centred on the middle of the stem (0.5, 1.5, 2.5, 3.5,…m). Root diameters were measured manually with a pair of callipers. The number of roots was calculated for root diameter classes of 1 mm interval (0–1.5, 1.6–2.5, 2.6–3.5,…). Fine roots are defined as having a diameter less than 1.5 mm.

The Palmerston North Dataset, PN (3 m poplar poles, densities 89, 160, 210, and 237 stems per hectare (sph)). This dataset was obtained from a study on the effect of tree density on root distribution of hybrid poplar (Populus deltoides Marshall x P. nigra L. clones ‘Veronese’ and ‘Tasman’) growing on two nearby (<1 km apart) pastoral hill country sites at AgResearch’s Ballantrae Hill Country Research Station near Palmerston North (Douglas et al. 2010). Plantings at the two sites were on slope angles of 0–30° but most were on angles of 5–20°. Soils were silt loams with limited natural drainage, grey subsoil, and various mottles and concretions. The trees were aged 9–11 years, and most were arranged in a Nelder radial planting design established in 1996. Maximum average tree height was 15.2 m, and mean diameter at breast height (DBH; 1.4 m above ground on the upslope side of trees) ranged from 0.17 to 0.30 m. Roots between adjacent trees at densities of 89 (10.6 m × 10.6 m), 160 (7.9 m × 7.9 m), 210 (6.9 m × 6.9 m), and 237 (6.5 m × 6.5 m) sph were assessed in trenches to determine root number and root diameter distribution. A total of 20 trenches were analysed in this study, 5 for each tree density (89, 160, 210, and 237 sph). Each trench was 1.2 m long, 1.0 m deep, and 0.4 m wide, and measurements were made by placing a 0.9 m × 0.9 m steel reinforcing mesh on the smoothed trench faces. The mesh comprised 36 cells, each 0.15 m × 0.15 m, arranged in six rows of six cells per row. The number and dimension of roots were collected for each cell. Root diameters were measured manually using a pair of callipers. The technique measured roots growing horizontally or at an angle to the trench face and excluded roots growing vertically (tap roots, sinker roots). The minimum distance of trench faces from trees was 0.9 m so that information on roots in close proximity to the original planting pole was not recorded. The values used as input for the model simulations for the two datasets are summarised in Table 1.Table 1
Data inputs for the rootdistribution model (DBH and distance of each tree, T, from each soil trench, P), in the four transects (89, 160, 210, and 237 sph) (also see Fig. 4)
Tree number and spacing
DBH (m)
Distance (m)
P1
P2
P3
P4
P5
89 sph
T1
0.19
5.35
13.4
16.1
18.8
26.7
T2
0.26
5.35
2.7
5.4
8.1
16.0
T3
0.26
16.15
8.1
5.4
2.7
5.2
T4
0.2
26.55
18.5
15.8
13.1
5.2
160 sph
T1
0.27
3.75
9.55
11.60
13.65
19.77
T2
0.29
3.75
2.05
4.10
6.15
12.27
T3
0.25
11.95
6.15
4.10
2.05
4.07
T4
0.22
20.1
14.30
12.25
10.2
4.07
210 sph
T1
0.31
3.35
8.45
10.2
11.95
17.2
T2
0.31
3.35
1.75
3.5
5.25
10.5
T3
0.31
10.35
5.25
3.5
1.75
3.5
T4
0.28
17.35
12.25
10.5
8.75
3.5
237 sph
T1
0.26
3.05
7.77
9.45
11.12
16.15
T2
0.3
3.05
1.67
3.35
5.02
10.05
T3
0.23
9.75
5.02
3.35
1.67
3.35
T4
0.21
16.15
11.72
10.05
8.37
3.35
Root distribution modelling
The root distribution data were used to calibrate the root distribution model described by Schwarz et al. (2010b). This modelling approach involves a static fractalbranching model (Tobin et al. 2007) based on simple morphogenetic parameters, similar to those used by other models (Diggle 1988; Lynch et al. 1997; OzierLafontaine et al. 1999; Pages et al. 2004).
where μ is the empirical pipe coefficient (N° m^{−2}) and DBH is the tree stem diameter at 1.4 m height (m).
where δ is a dimensionless empirical coefficient.
where d _{frf} is a dimensionless factor for the estimation of the scaling factor of the survival function (=d _{frf} ⋅ DBH), and w is the shape factor of the survival function.
The density of coarse roots as a function of distance from the stem is deduced by assuming a constant value of branching distance (BD) (OzierLafontaine et al. 1999; Van Noordwijk and de Willigen 1994). At each branching point, a coarse root may split into smaller diameter roots, maintaining a constant proportionality factor (red_coeff) between prebranching crosssectional areas of a coarse root and the sum of crosssectional areas of finer roots after branching (Van Noordwijk and de Willigen 1994; Wang et al. 2006). For a given maximum lateral root system extent and for a given fineroot density, the diameter of coarse roots at each branching point is computed defining root diameter as a function of the distance from the stem. Hence, for each distance from a tree, a maximum root diameter (RDmax) is computed as an upper bound for root diameter distribution. The number of roots as a function of root diameter for diameter values between fineroot size and maximum diameter is calculated on the basis of empirical root distribution data using an exponential function (parameter exp_distr) (see Schwarz et al. 2010b).
Calibration of the model was conducted by maximising the likelihood value obtained by the sum of normal probability function of the residuals (mean = 0 and standard_deviation = (error variance)^(0.5)).
Scaling root reinforcement
Root reinforcement at the hillslope scale is estimated assuming the minimum value of tensile forces calculated for the bundles of roots between neighbouring trees. The tensile forces of the root bundles are calculated using the RBMw (Schwarz et al. 2013). The RBMw requires a set of parameters to quantify the mechanical behaviour of roots and a set of root distribution data (as the results of measurements or modelling, as shown in the previous section).
The mechanical behaviour of each class of root diameter is calculated using the equation of elasticity theory where maximum tensile force and Young’s modulus are functions of the root diameter. The failure of roots approaching a critical value of maximum tensile force (breakage or slipout) is calculated using a survival function. The form of the survival function (in our case a Weibull function) is used to characterise the variability of the root mechanical properties.
where d is underbark root diameter (mm). In order to account for the root diameter underbark used by Watson et al.^{1} and the root diameter with bark considered in the datasets G and PN, the conversion proposed by Watson et al.^{1} was used (under_bark_diameter = 0.72 diameter_with_bark − 0.35; R ^{2} = 0.95). No data for the calibration of the secant Young’s modulus equation and the breaking survival function were found for the implementation in the RBMw so the values reported by Schwarz et al. (2013) were used.
The spatial distribution of root reinforcement is characterised in terms of maximum tensile force of the root bundle crossing 1 m of landslide scarp.
Slope stability calculations
Calculations of slope stability are made within the limit equilibrium framework considering a unique slice corresponding to the dimensions of the considered shallow landslide. The equilibrium of forces is calculated considering the threedimensional shape of the shallow landslide (Schwarz et al. 2010a). The shear strength at the failure surface of the shallow landslide was quantified using the MohrCoulomb criterion. The inclusion of lateral root reinforcement in slope stability calculations is achieved by considering an additional stabilising force proportional to the upper scarp length and to the lateral root reinforcement calculated for a defined tree density (Schwarz et al. 2010a). In order to compare the efficiency of the lateral root reinforcement versus the basal root reinforcement (Schwarz et al. 2014), 10 % of the lateral root reinforcement was considered in the calculations acting on the landslide slip surface. The value of 10 % is obtained considering the reviewed data of vertical root distribution reported by Glenz (2005) and Phillips et al. (2014).
where A _{L} is the landslide area [km ^{2}], ϱ is the parameter primarily controlling powerlaw decay for medium and large values in the threeparameter inversegamma probability distribution, a is the parameter primarily controlling location of the maximum in the threeparameter inversegamma probability distribution, s is the parameter primarily controlling exponential rollover for small values in the threeparameter inversegamma probability distribution, and Γ(ϱ) is the gamma function of ϱ. The set of parameters fitted by Malamud et al. (2004) (ϱ = 1.4, a = 1.28 10^{−3}, and s = −1.32 10^{−4}) was applied, and a sample of 10,000 hypothetical landslides generated using a random number generator function in the R environment (www.rproject.org ).
Variability of soil depth, soil effective friction angle, and soil cohesion were considered assuming a normal distribution. For simplicity, fully saturated conditions were assumed representative of an extreme triggering rainfall event (i.e. pore water pressure = soil depth). Assuming saturated conditions of the soil trench implies that possible additional suction stress due to capillary forces are not present, soil specific weight for fully saturated conditions are considered, and that effective normal stress is given by the normal stress minus the pore water pressure at the shear surface (which is proportional to the soil depth). Based on inventories of shallow landslides (De Rose et al. 1995; Rickli and Graf 2009 ), the mean soil depth was set to 1.2 m with a standard deviation of 0.1 m. The range of soil mechanical parameters and their variability were estimated according to data reported by Claessens et al. (2007), assuming the values to be normally distributed. Three classes of effective friction angles and cohesion were defined (24°, 29°, and 34° for the friction angle and 0, 6, and 12 kPa for the soil effective cohesion). Soil density was fixed to 1.4 [t m^{−3}] (Claessens et al. 2007). In order to confine the analysis of sediment balance only to shallow landslides (defined as landslides with maximum soil thickness equal to 2 m), the relationship found by KaldaronAsael et al. (2008) was used to estimate the threshold of maximum shallow landslide area. Assuming a coefficient of 0.03, a threshold of about 5000 m^{2} was obtained. Based on the event analysis of Rickli and Graf (2009), an elliptical shape of the landslides is assumed with a width/length relationship equal to 0.5.
The approach described above is essentially the basis of the SlideforNET tool, which was developed to quantitatively assess the effect of lateral root reinforcement on shallow slope stability. The detailed description of this formulation can be found in Schwarz et al. (2010a, 2012b, 2014).
Results
Root distribution
Calibration with the G dataset
Values of the fitted parameters of the root distribution model for the two trees of the G trial and all the trenchs of the PN trial
Parameter  Pole (1 year old)  Pole (2 years old)  Trench 

Pipe coefficient (pipe_coef)  0.029  0.045  0.077 
Exponent coarse root distribution (exp_distr)  −0.5  −0.49  −1.38 
Distance peak fine root frequency (d_frf)  2.5  2.1  6.4 
Weibull coef. for fine root distribution (w.coef)  2.4  2.1  6.2 
Coarse root proportionality factor (red_coef)  2.5  3.4  0.18 
Maximum root spread coefficient (max_dist_coef)  74  88  14 
Error variance (var)  0.17  0.08  4.25 
Calibration with the PN dataset
The calibrated values of the model parameters for the two datasets are summarised in Table 2. The two datasets exhibit considerable differences with the pipe coefficient, the root distribution exponent, and the maximum root spread coefficient in particular.
Root reinforcement
Discussion
Root distribution and reinforcement
Differences in measured root distribution in the two datasets (G and PN) analysed may be due to a range of factors. Phillips et al. (2014) found that root growth rate at the G trial site was greater than that reported in comparable studies reported in the literature, arguing that the uniform sandy loam alluvial soil, warm temperatures, available moisture, and weed suppression were all factors likely to have contributed to particularly favourable root growing conditions. In contrast, the denser, clayrich, heterogeneous soils such as those at the PN site may have restricted root growth and performance of trees at that site. The different tree ages at the time of sampling (1–2 years in G and 9–11 years in PN) represent an issue when comparing data on root distribution. However, in the context of the modelling framework, the physiologicalrelated parameters such as the pipe coefficient or maximum rooting distance coefficient represent a useful term of comparison between the growing conditions of the two field sites. Differences in the root distribution within the same field site may be explained by the variability of the initial planting material and local variation of environmental factors. For instance, it is recognised that the size and volume of planting materials influence the initial growth rate of roots and shoots (Phillips et al. 2014, 2015; Sulaiman 2006), due to the effect of rhizocauline (Schiechtl 1992) present in the cambium cells. The higher survival and growth rates of larger diameter poles may compensate for their extra startup cost during field establishment (Sulaiman, 2006). Another reason for the differences in the root distribution between the two datasets could be related to the different sampling methods used to obtain the field data. The timeconsuming excavation of entire root systems tends to limit measurements to only a few trees, but the trench sampling used to obtain the PN dataset provides a lower accuracy in terms of spatial characterisation of roots but enables the analysis of variability among more trees. In the case of asymmetrical root systems, the trenching method may lead to underestimation or exaggeration of root number depending on the direction of the trench position. In this study, how the proposed quantitative framework could be applied to both types of datasets has been demonstrated, using either a single tree (e.g. the G dataset) or multiple trees (e.g. the PN dataset) for model calibration. Although comparing the root distribution of the two datasets is difficult because of different field trial characteristics, age of trees, and sampling methods, the application of a modelling framework does allow a general comparison of the datasets in terms of calibrated model parameters. Moreover, the possibility to crosscalibrate the model using the best existing datasets for ‘Veronese’ poplar allows for some discussion on the variability of root distribution as influenced by environmental conditions.
One of the most important and difficult to measure parameters of a tree’s root distribution is the maximum lateral root spread. Only difficult and timeconsuming measurements, as performed for example in the G dataset, lead to an accurate assessment of the maximal lateral spread of a root system (Phillips et al. 2014). Nevertheless, the results shown in Fig. 4 support the calibrated model coefficient value of 14 (shown in Table 2) for the PN dataset. This number is based on a mean DBH of 0.3 m. From this, a maximum lateral root spread of 4.2 m was obtained, which means that no roots are expected to cross soil trenches at the middle distance between trees with density lower than 160 sph. Analogously, the calculated maximum lateral root spread of 13.3 m for the 2yearold poplar of the G dataset is comparable to the values reported in Phillips et al. (2014) for growth of another ‘Veronese’ poplar at the G field site.
Another difficult parameter to be determined is the pipe coefficient. An exact estimation of this parameter is possible only if all the fine roots of the root system are measured, as in the case of the G dataset; otherwise, it can be estimated fitting the distribution of fine roots at different distances from the tree stem (as shown in Fig. 4). Not surprisingly, the results of this study show that the estimated pipe coefficient is considerably different for the two datasets. Considering that the pipe coefficient is an index of how many fine roots are actively contributing to metabolism of the plant (assimilating water and nutrients), it could be argued that increased concurrence and unfavourable local stand characteristics (i.e. soil type) would lead to a reduction in metabolism and thus to a reduction in fineroot number and growth. However, the value of the pipe coefficient results is actually higher for the PN dataset than the G dataset because the total number of fine roots is concentrated in a smaller area occupied by the root system, resulting in low total number of fine roots.
Fitted parameters in the root distribution model such as the ‘Exponent of coarse root distribution (exp_distr)’ and the ‘Coarse root proportionality factor (red_coef)’ reflect the characteristics of the coarse roots in terms of frequency and maximum root diameter, respectively. The ‘Peak distance of fine root frequency (d_frf)’ and ‘Weibull coef.’ for ‘fineroot distribution (w.coef)’ parameters were used to characterise the fineroot distribution as a function of distance from tree stem. The range of d_frf values indicated that the maximum density of fine roots was expected at a distance equal to 2 and 6 times the tree DBH for the G and PN datasets, respectively. The importance of fineroot density in determining the magnitude of root bundle tensile strength is shown in Fig. 5 in comparison to the results of Fig. 2b, where the maximum root reinforcement for a tree with 0.3 m DBH is expected at about 1.9 m, exactly at the distance where the maximum intensity of fineroot density is reached. Overall, the two sets of parameters calibrated for the G dataset are consistent, showing that the tree stem diameter has a major influence on the different results of root distribution.
Seasonal changes in fineroot frequency may also be an important factor influencing lateral root reinforcement. McIvor et al. (2011) found that ‘Veronese’ poplar had a significant reduction in fineroot length density during the dormant season, with little change in coarse root distribution observed. From the modelling perspective, it may be assumed that only root diameters greater than 2 mm contribute to lateral root reinforcement constantly during the year; this assumption may have a strong effect on the estimation of root reinforcement since fine roots in some cases determine the overall mechanical behaviour of the root bundle. For instance, the calculation of lateral root reinforcement for the root distribution measured at trench number 4 (P4) on the 160 sph plot of the PN dataset produced a value of 10.4 kN m^{−1}, whereas assuming zero root for diameter classes <2 mm, a value of 8.85 kN m^{−1} lateral root reinforcement was obtained. In this case, the contribution of fine roots to total lateral root reinforcement is relatively small (about 20 %), but in the case of a root bundle far from the tree stem, this contribution may be higher and lead to values of root reinforcement equal to 0.
Root systems on moderate to steep slopes may develop more asymmetric growth patterns than on flat land, particularly in the upslopedownslope direction (Chiatante et al. 2003), depending on factors including species, soil texture, plant age, mechanical effects (windsnow), and nutrientwater availability (Stokes et al., 2009). A simple overlapping pattern of root systems was used in the calculation of the minimum distance between trees presented in Fig. 7. This simplification excluded the possible effect of root system concurrence and asymmetry due either to hillslope inclination or the concurrence of neighbouring trees. Even if this simplification is unrealistic, it seems that when considering the effective tree distance for slope stabilisation, then the concurrence effects of neighbouring root systems on the root distribution are weak and thus may be excluded (Schwarz et al. 2010b). Also under these conditions, the effects of grafting or mechanical interaction between roots have no influence on the calculation of rootbundle tensile forces, as discussed in Giadrossich et al. (2012). Moreover, in the case of poplar species growing on gentle slopes, no effect of direction of slope was found by McIvor et al. (2005), which is in accord with the assumption made in the model tested here. Further research is needed to clarify these aspects for other locations or other tree species.
The differences in root diameter distribution for the two datasets (G and PN) are reflected in a considerable difference in the distribution of maximum root pullout forces as shown in Fig. 5. Root distribution is an important factor in determining root reinforcement behaviour (maximal pullout forces and stiffness), as well as the mechanical and geometrical properties of roots (Schwarz et al. 2010a; Cohen et al. 2011). Root number as well as the distribution of diameter classes of roots also has a major influence on the mechanical properties of a bundle of roots. The peak of root reinforcement at 2 m distance from a 0.3 m DBH tree modelled for the PN dataset is due to the peak of root frequency calculated for this distance. This result shows that, at this distance, root reinforcement is dominated by roots with diameter classes less than 2 mm. While the number of roots mainly influences the magnitude of the reinforcement, the distribution of roots in different diameter classes influences the magnitude of reinforcement, with a factor ranging from 1 to 3 (Cohen et al., 2011), and the stiffness of the reinforcement (Schwarz et al. 2013). This last aspect has been discussed less often in the literature, but it represents an important factor in conditions where the stabilisation effects of roots are due mainly to lateral roots (Schwarz and Cohen 2011; Schwarz et al. 2012a, Schwarz et al. 2015). Although the results show a good prediction of the root distribution model (Fig. 4), the prediction of root reinforcement in Fig. 6 shows how sensitive this calculation is to slightly different distribution of roots, especially in the case of the PN dataset. Vice versa, the model prediction of root reinforcement based on the G dataset fits the data better (Fig. 5). This difference may be explained by two main reasons. First, the simulation conducted for only a single tree is not a representative of an entire stand and thus shows lower variability. Second, the calibration of the root distribution model based on data of a complete excavated root system with a greater number of trench distances (one every metre) allows for the whole root system to provide a better prediction of root distribution.
The forcediameter relationship is also important because it strongly influences the calculation of root reinforcement using the RBMw. Forcediameter relationships are usually quantified by laboratory tensile tests of root diameters up to a few millimetres, but difficulties exist in extrapolating values for larger root diameters. The unpublished poplar data of Watson et al.^{1} include root diameters up to about 8 mm (see Additional file 1), which increases the plausibility of results from the RBMw for a root bundle dominated by roots smaller than 8 mm. The variability of measured root tensile strength is considered in the RBMw using a survival function with a Weibull shape, which describes the probability of a root breaking before or after the fitted value of maximum tensile force (Schwarz et al. 2013). Even if the unpublished forcediameter data of ‘Veronese’ poplar roots from Watson et al.^{1} used in the calculation were obtained for root material from a different location than PN and G, it was assumed that these data were, on average, representative for the studied field conditions. In fact, it is not possible to make a direct comparison of these data with other data from the literature for the same clone, or even the same species. Further studies are needed to assess the variability of root mechanical properties of poplar, also both as a function of local environmental conditions and of the development stage of the plant. For instance, Hathaway and Penny (1975) discussed how root strength depends on season, showing that the root tensile strength of 1yearold poplar clone ‘I488’ was greater during the winter, reaching a peak in September (early spring), in parallel to a peak in lignin content. In view of such probable root mechanical seasonal variability, the results presented in this study aimed to quantify the possible range of lateral root reinforcement due to ‘Veronese’ poplar roots only as a function of different root distribution datasets.
Stabilisation effect of lateral root reinforcement and application of results for erosion control
Rooting depth reported in the literature for ‘Veronese’ poplar ranges between 0.5 and 1 m, indicating that no basal root reinforcement may be expected in shallow landslides with failure surfaces commonly between 1 and 2 m soil depth. According to Schwarz et al. (2010a), lateral root reinforcement has been shown to contribute to slope stability. The values of estimated lateral root reinforcement in our results confirm the finding of Schwarz et al. (2010a). The use of the ‘Slidefor^{NET}’ approach allows for a stochastic characterisation of the effects of lateral root reinforcement on slope stability in terms of reduced partial landslide probability (Fig. 8), or in terms of stabilised landslide volume (Fig. 9).
The stabilisation effects of lateral roots are related to their spatial distribution, which is related to the architecture of each root system and the position of the trees on the slope (intertree distance). Interlocking root systems with a high number of roots allow a wider redistribution of destabilising forces, increasing the probability that unstable zones are linked to stable ones, assuring the stability of the whole slope. As discussed in the literature, roots under tension contribute to slope stabilisation, but those under compression may also contribute to the overall reinforcement and may play a major role (Schwarz and Cohen 2011). The efficiency of root reinforcement under compression may eventually depend on the geometry of the tree layout (squared, triangular, radial, etc.); further studies are needed to analyse this aspect.
The results in Fig. 7 show that the minimal lateral root reinforcement has a realistic range between 2 and 15 kN m^{−1}, which in terms of mean distances between trees (see Fig. 5) corresponds to 5.5 and 18 m for poplar trees with 0.3 m DBH (tree density of 330 and 30 sph, respectively). Comparing results obtained from the two datasets (G and PN) enables the possible range of lateral root reinforcement that could be expected for different tree spacings to be defined. As discussed previously, the difference in the two datasets results in an important difference in calculated lateral root reinforcement, which in turn gives an idea of the variability of slope stabilisation efficiency of a spaced tree population. Data on the estimated minimal effective tree distance derived from empirical studies (Douglas et al. 2013), 8 × 8 m − >160 sph, falls in the calculated range of values obtained in this study giving further weight to the approach developed here. However, it is important to emphasise that root distribution is highly variable and may not be represented by single values, but by a range of values.
The importance of trees to contribute to soil carbon has been widely recognised and therefore soil conservation trees have both environmental and a political relevance. For instance, requirements for claiming carbon credits are under discussion within the New Zealand Emissions Trading Scheme (ETS). A tree canopy cover of 30 % is considered to provide a balance between carbon sequestration and the reduction of annual pasture production due to trees (estimated to be 10 % less than from no trees). Using the data reported by McIvor and Douglas (2012) for a 0.3m DBH poplar tree, about 70 sph is needed to achieve this balance. This corresponds to a mean stem spacing of about 12 m. Comparing these data with the results of the present work, such a tree spacing would lead to lateral root reinforcement ranging between 0 kN m^{−1}for the PN dataset and 16 kN m^{−1} for the G dataset. These results show that it is important to have complete information on root distribution characteristics in order to be able to perform quantitative analysis of the effect of spaced trees on the stability of hillslopes.
In this study, the results obtained for different dimensions of ‘Veronese’ poplar trees were compared, ranging from 66 mm DBH for the 1yearold pole in the G dataset to 300 mm DBH for poles in the PN dataset in the 237 sph density plot. This range in DBH corresponds to the dimensions that poles of this clone may reach in 11 years of growth on a gentle hillslope. Assuming an ageDBH correlation, it would be possible to predict the temporal development of root reinforcement. This type of information would enable management strategies for spaced tree populations on slopes to be developed that would ensure constant minimal root reinforcement over a long period of time. For instance, if a high tree density was planted (300–400 sph), later thinning and pruning may optimise the density of trees in later years in view of growth rates and root distribution. A future valuable initiative would be to determine how thinning and pruning measures influence root distribution over time.
Even in the case where the mechanical effects of lateral roots on the stabilisation of shallow landslide are small or negligible, the hydrological effects of vegetation may still contribute to slope stabilisation. In fact, depending on the hydromechanical condition of the slope, vegetation may increase directly or indirectly the matric suction and the permeability of the soil, enhancing slope stability. However, considering the typical hydrological conditions for the triggering of shallow landslides in New Zealand (Douglas et al. 2006; Wilkinson et al. 2002; Hawke and McConchie 2011), the hydrological effects of vegetation are strongly seasonally dependent and the stabilisation effects during intense and prolonged rainfall are likely to be limited.
The inclusion of poplar root distribution and tree size data in slope stability models in this study has provided valuable guidance on potential implications for pastoral hillside stability. The approach is believed to be the first for widespaced tree plantings. Although results have been restricted to young trees with maximum DBH of 0.3 m, pastoral slopes with establishing trees are often the most vulnerable to shallow landslides and other erosion processes mainly because trees have small root systems compared with older trees (McIvor et al. 2008). The results support a strategy of planting at high density to hasten slope stabilisation and thinning later as canopies and roots develop, to facilitate growth of the pasture understorey.

Root distribution mapping of trenches on a representative ecological range of hillslopes, performing pullout tests of roots with diameters greater than 8 mm (eventually in different seasons),

Quantifying the performance of different dimensions of planting material in terms of temporal development of lateral root reinforcement, and

Performing event analysis for shallow landslides in view of a better stochastic characterisation of the frequencymagnitude relationship of such events.
Conclusions
This paper describes a unique quantitative approach that combines information about root distribution and root mechanical data in order to calculate the spatial distribution of root reinforcement as a function of tree dimension and distance between ‘Veronese’ poplar trees. The results were used to formulate guidelines for the planning of bioengineering measures with the aim of reducing erosion on pastoral hill slopes in New Zealand. The calculations show that the definition of effective planting density for the same poplar clone ‘Veronese’ is a function of the local root growing condition, the slope inclination, and the soil mechanical properties (effective friction angle and cohesion). Generally, it can be concluded that planting density that ranges between 330 and 160 sph (corresponding to stem distance of 5.5 and 8 m, respectively) would assure significant root reinforcement for slope stabilisation (>2 kN m^{−1}) and reduce the volume of triggered shallow landslides by up to 100 %. In ideal growing conditions, tree spacing starting from 100 sph is sufficient for stem DBH larger than 0.15 m to assure enough root reinforcement. A lower planting density would lead to less lateral root reinforcement, and the contribution of vegetation to slope stability would probably be limited to the hydrological effects.
Watson, A, McIvor, IR, Douglas, GB. Live rootwood tensile strength of Populus x euramericana ‘Veronese’ poplar. Unpublished Landcare Research report prepared for FRST Contract CO2X0405 in 2007. http://www.poplarandwillow.org.nz/documents/veronesepoplarpaperliverootwoodstrength.pdf. See Additional file 1.
Declarations
Open AccessThis article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.
Authors’ Affiliations
References
 Ammer, C., & Wagner, S. (2005). An approach for modelling the mean fineroot biomass of Norway spruce stands. Trees, 19, 145–153.View ArticleGoogle Scholar
 Basher, L. R. (2013). Erosion processes and their control in New Zealand. In J. R. Dymond (Ed.), Ecosystem services in New Zealand—conditions and trends (pp. 363–374). Lincoln, New Zealand: Manaaki Whenua Press.Google Scholar
 Betteridge, K., Costall, D., Martin, S., Reidy, B., Stead, A., & Millner, I. (2012). Impact of shade trees on Angus cow behaviour and physiology in summer dry hill country: grazing activity, skin temperature and nutrient transfer issues. In L. D. Currie & C. L. Christensen (Eds.), Advanced nutrient management: gains from the past—goals for the future. [Occasional Report No. 25]. Palmerston North: Massey University Fertilizer and Lime Research Centre..
 Blaschke, P. M., Trustrum, N. A., & DeRose, R. C. (1992). Ecosystem processes and sustainable land use in New Zealand steeplands. Agriculture, Ecosystems & Environment, 41, 153–178.View ArticleGoogle Scholar
 Blaschke, P, Hicks, D, Meister, A (2008). Quantification of the flood and erosion reduction benefits, and costs, of climate change mitigation measures in New Zealand. [Report prepared by Blaschke and Rutherford Environmental Consultants for MfE]. Wellington: Ministry for the Environment
 Chiatante, D., Scippa, G. S., Di Iorio, A., & Sarnataro, M. (2003). The influence of steep slope on root system development. Journal of Plant Growth Regulation, 21, 247–260.View ArticleGoogle Scholar
 Claessens, L., Schoorl, J. M., & Veldkamp, A. (2007). Modeling the location and shallow landslides and their effects on landscape dynamics in large watersheds: an application for Northern New Zealand. Geomorphology, 87, 16–27.View ArticleGoogle Scholar
 Cohen, D., Schwarz, M., & Or, D. (2011). An analytical fiber bundle model for pullout mechanics of root bundles. Journal of Geophysical Research, 116, F03010.
 De Rose, R. C., Trustrum, N. A., Thomson, N. A., & Roberts, A. H. C. (1995). Effect of landslide erosion on Taranaki hill pasture production and composition. New Zealand Journal of Agricultural Research, 38(4), 457–471.View ArticleGoogle Scholar
 Diggle, A. J. (1988). ROOTMAP—a model in threedimensional coordinates of the growth and structure of fibrous root systems. Plant and Soil, 105, 169–178.View ArticleGoogle Scholar
 Douglas, G. B., Trustrum, N. A., & Brown, I. C. (1986). Effect of soil slip erosion on Wairoa hill pasture production and composition. New Zealand Journal of Agricultural Research, 29(2), 183–192.View ArticleGoogle Scholar
 Douglas, G. B., Walcroft, A. S., Hurst, S. E., Potter, J. F., Foote, A. G., Fung, L. E., et al. (2006). Interactions between widely spaced young poplars (Populus spp.) and the understorey environment. Agroforest Systems, 67, 177–186.View ArticleGoogle Scholar
 Douglas, G. B., McIvor, I. R., Potter, J. F., & Foote, A. G. (2010). Root distribution of poplar at varying densities on pastoral hill country. Plant and Soil, 333, 147–161.View ArticleGoogle Scholar
 Douglas, G. B., McIvor, I. R., Manderson, A. K., Koolaard, J. P., Todd, M., Braaksma, S., et al. (2013). Reducing shallow landslide occurrence in pastoral hill country using widespaced trees. Land Degradation and Development, 24, 103–114. doi:10.1002/ldr.1106.View ArticleGoogle Scholar
 Dymond, J. R., Ausseil, A. G., Shepherd, J. D., & Buettner, L. (2006). Validation of a regionwide model of landslide susceptibility in the ManawatuWanganui region of New Zealand. Geomorphology, 74(1–4), 70–79.View ArticleGoogle Scholar
 Giadrossich, F., Schwarz, M., Cohen, D., Preti, F., & Or, D. (2012). Mechanical interactions between neighbouring roots during pullout tests. Plant and Soil. doi:10.1007/s1110401214751.Google Scholar
 Glenz, C. (2005). Processbased, spatiallyexplicit modelling of riparian forest dynamics in Central Europe: tool for decisionmaking in river restoration. [PhD Thesis]. École polytechnique fédérale de Lausanne (EPFL). doi:10.5075/epflthesis3223.
 Hathaway, R. L., & Penny, D. (1975). Root strength in some Populus and Salix clones. New Zealand Journal of Botany, 13, 333–344.View ArticleGoogle Scholar
 Hawke, R., & McConchie, J. (2011). In situ measurement of soil moisture and porewater pressures in an ‘incipient’ landslide: Lake Tutira, New Zealand. Journal of Environmental Management, 92(2), 266–274. doi:10.1016/j.jenvman.2009.05.035.View ArticlePubMedGoogle Scholar
 Hawley, J. G., & Dymond, J. R. (1988). How much do trees reduce landsliding? Journal of Soil and Water Conservation, 43, 495–498.Google Scholar
 Heaphy, M., Lowe, D., Palmer, D., Jones, H., Gielen, G., Oliver, G., et al. (2014). Assessing drivers of plantation forest productivity on eroded and noneroded soils in hilly land, eastern North Island, New Zealand. New Zealand Journal of Forestry Science, 44: 24.View ArticleGoogle Scholar
 Hewitt, A.E. (1998). New Zealand soil classification. (2nd ed.). Lincoln, New Zealand: Manaaki Whenua–Landcare Research.
 KaldaronAsael , B, Katz, O, Aharonov, E, Marco, S (2008). Modeling the relation between area and volume of landslides. [Report GSI/06/2008]. Jerusalem: Ministry of National Infrastructures Geological Survey of Israel.
 Lambert, M. G., Trustrum, N. A., & Costall, D. A. (1984). Effect of soil slip erosion on seasonally dry Wairarapa hill pastures. New Zealand Journal of Agricultural Research, 27(1), 57–64.View ArticleGoogle Scholar
 Lynch, J. P., Nielsen, K. L., Davis, R. D., & Jablokow, A. G. (1997). SimRoot: modelling and visualisation of root systems. Plant and Soil, 188, 139–151.View ArticleGoogle Scholar
 Malamud, B. D., Turcotte, D. L., Guzzetti, F., & Reichenbach, P. (2004). Landslide inventories and their statistical properties. Earth Surface Processes and Landforms, 29, 687–711.View ArticleGoogle Scholar
 Marden, M., Arnold, G., Gomez, B., & Rowan, D. (2005). Pre and postreforestation gully development in Mangatu Forest, East Coast, North Island, New Zealand. River Research and Applications, 21, 757–771.
 Marden, M. (2012). Effectiveness of reforestation in erosion mitigation and implications for future sediment yields, East Coast catchments, New Zealand: a review. New Zealand Geographer, 68, 24–35.View ArticleGoogle Scholar
 Marden, M., & Rowan, D. (2015). The effect of land use on slope failure and sediment generation in the Coromandel region of New Zealand following a major storm in 1995. New Zealand Journal of Forestry Science, 45: 10.View ArticleGoogle Scholar
 Marden, M., Rowan, D., & Phillips, C. J. (1995). Impact of cycloneinduced landsliding on plantation forests and farmland in the East Coast of New Zealand: a lesson in risk management. In K. Sassa (Ed.), Proceedings of XX IUFRO World Congress, Tampere Finland 1995 (pp. 133–145).Google Scholar
 McIvor, IR, & Douglas, GB (2012). Poplars and willows in hill country  stabilising soils and storing carbon. In L.D. Currie and C.L. Christensen (Eds.), Advanced Nutrient Management: Gains from the Past  Goals for the Future. ). [Occasional Report No. 25]. Palmerston North, New Zealand: Massey University Fertilizer and Lime Research Centre.
 McIvor, I. R., Metral, B., & Douglas, G. B. (2005). Variation in root density of poplar trees at different plant densities. Proceedings of the Agronomy Society of New Zealand, 35, 66–73.Google Scholar
 McIvor, I. R., Douglas, G. B., Hurst, S. E., Hussain, Z., & Foote, A. G. (2008). Structural root growth of young Veronese poplars on erodible slopes in the southern North Island, New Zealand. Agroforestry Systems, 72(1), 75–86. doi:10.1007/s1045700790905.View ArticleGoogle Scholar
 McIvor, I. R., Douglas, G. B., & Benavides, R. (2009). Coarse root growth of Veronese poplar trees varies with position on an erodible slope in New Zealand. Agroforestry Systems, 76(1), 251–264. doi:10.1007/s104570099209y.View ArticleGoogle Scholar
 McIvor, IR, Douglas, GB, Dymond, J, Eyles, G, Marden, M (2011). Pastoral hill slope erosion in New Zealand and the role of poplar and willow trees in its reduction. In D. Godone and S. Silvia (Eds.), Soil Erosion Issues in Agriculture. (pp. 257–278). InTech  Open Access Publisher http://www.intechopen.com/articles/show/title/pastoralhillslopeerosioninnewzealandandtheroleofpoplarandwillowtreesinitsreduction.
 OzierLafontaine, H., Lecompte, F., & Sillon, J. F. (1999). Fractal analysis of the root architecture of Gliricidia sepium for the spatial prediction of root branching, size and mass: model development and evaluation in agroforestry. Plant and Soil, 209, 167–180.View ArticleGoogle Scholar
 Pages, L., Vercambre, G., Drouet, J. L., Lecompte, F., Collet, C., & Le Bot, J. (2004). Root Typ: a generic model to predict and analyse the root system architecture. Plant and Soil, 258, 103–119.View ArticleGoogle Scholar
 Pearce, AJ, O'Loughlin, CL, Jackson, RJ, Zhang, XB (1987). Reforestation: onsite effects on hydrology and erosion, eastern Raukumara Range, New Zealand. Forest Hydrology and Watershed Management. In Proceedings of the Vancouver Symposium (pp. 489–497). [IAHSASIH Publication No. 167]
 Phillips, C. J., & Marden, M. (2005). Reforestation schemes to manage regional landslide risk. In T. Glade, M. Anderson, & M. J. Crozier (Eds.), Landslide hazard and risk (pp. 517–547). Chichester, UK: John Wiley.Google Scholar
 Phillips, C. J., Marden, M., & Pearce, A. (1990). Effectiveness of reforestation in prevention and control of landsliding during large cyclonic storms. In Proceedings XIX World IUFRO Congress, Montreal, Canada.Google Scholar
 Phillips, C. J., Marden, M., & Basher, L. (2012). Plantation forest harvesting and landscape response—what we know and what we need to know. New Zealand Journal of Forestry, 56(4), 4–12.Google Scholar
 Phillips, C. J., Marden, M., & Lambie, S. M. (2014). Observations of root growth of young poplar and willow planting types. New Zealand Journal of Forestry Science, 44:15.
 Phillips, C. J., Marden, M., & Lambie, S. M. (2015). Observations of “coarse” root development in young trees of nine exotic species from a New Zealand plot trial. New Zealand Journal of Forestry Science, 45: 13.View ArticleGoogle Scholar
 Pollen, N., & Simon, A. (2005). Estimating the mechanical effects of riparian vegetation on streambank stability using a fiber bundle model. Water Resources Research, 41, W07025.
 Rickli, C., & Graf, F. (2009). Effects of forests on shallow landslides – case studies in Switzerland. Forest Snow and Landscape Research, 82(1), 33–44.
 Roering, J. J., Schmidt, K. M., Stock, J. D., Dietrich, W. E., & Montgomery, D. R. (2003). Shallow landsliding, root reinforcement, and the spatial distribution of trees in the Oregon Coast Range. Canadian Geotechnical Journal, 40, 237–253.View ArticleGoogle Scholar
 Rosser, B. J., & Ross, C. W. (2011). Recovery of pasture production and soil properties on soil slip scars in erodible siltstone hill country, Wairarapa, New Zealand. New Zealand Journal of Agricultural Research, 54, 23–44.View ArticleGoogle Scholar
 Schiechtl, H. M. (1992). Weiden in der Praxis: Die Weiden Mitteleuropas und ihre Verwendung und ihre Bestimmung. Berlin: Patzer Verlag. 130 S.
 Schwarz, M., & Cohen, D. (2011). Influence of root distribution and compressibility of rooted soil on the triggering mechanism of shallow landslides. [Geophysical Research Abstracts 13, EGU 2011–4817], EGU General Assembly 2011, Vienna, Austria.
 Schwarz, M., Preti, F., Giadrossich, F., Lehmann, P., & Or, D. (2010a). Quantifying the role of vegetation in slope stability: a case study in Tuscany (Italy). Ecological Engineering, 36, 285–291.View ArticleGoogle Scholar
 Schwarz, M., Lehmann, P., & Or, D. (2010b). Quantifying lateral root reinforcement in steep slopes: from a bundle of roots to tree stands. Earth Surface Processes and Landform, 35, 354–367.View ArticleGoogle Scholar
 Schwarz, M., Cohen, D., & Or, D. (2012a). Spatial characterization of root reinforcement at stand scale: theory and case study. Geomorphology, 171–172, 190–200. http://dx.doi.org/10.1016/j.geomorph.2012.05.020.View ArticleGoogle Scholar
 Schwarz, M., Thormann, J.J. , Zürcher, K., & Feller, K. (2012b). Quantifying root reinforcement in protection forests: implication for slope stability and forest management. Proceedings of the Interpraevent 2012, Grenoble, France.
 Schwarz, M., Feller, K. & Thormann, J.J. (2013). Entwicklung und Validierung einer neuen Methode für die Beurteilung und Planung der minimalen Schutzwaldpflege auf rutschgefährdeten Hängen. [Final report to the Wald und Holzforschungsfonds]. Bern: Swiss Federal Office for the Environment FOEN.
 Schwarz, M., Dorren, L., & Thormann, J.J. (2014). SLIDEFORNET: a web tool for assessing the effect of root reinforcement on shallow landslides. In International Conference, "Analysis and Management of Changing Risks for Natural Hazards", 18–19 November 2014 l Padua, Italy
 Schwarz, M., Rist, A., Cohen, D., Giadrossich, F., Egorov, P., Büttner, D., et al. (2015). Root reinforcement of soils under compression. Journal of Geophysical Research: Earth Surface , 120, 2103–2120. doi:10.1002/2015JF003632.
 Sidle, R.C., & Ochiai, H. (2006). Landslides: Processes, Prediction, and Land Use. Washington DC: American Geophysical Union.
 Stokes, A., Atger, C., Bengough, A. G., Fourcaud, T., & Sidle, R. C. (2009). Desirable plant root traits for protecting natural and engineered slopes against landslides. Plant and Soil, 324, 1–30.
 Sulaiman, Z. (2006). Establishment and silvopastoral aspects of willow and poplar. [PhD Thesis], Institute of Natural Resources, Massey University, Palmerston North, New Zealand
 Tobin, B., Cermak, J., Chiatante, D., Danjon, F., Di Orio, A., Dupuy, L., et al. (2007). Towards developmental modelling of tree root systems. Plant Biosystems, 141, 481–501.View ArticleGoogle Scholar
 Van Noordwijk, S. L. Y., & de Willigen, P. (1994). Proximal root diameter as predictor of total root size for fractal branching models. Plant and Soil, 164, 107–117.View ArticleGoogle Scholar
 Wall, A. J., Mackay, A. D., Kemp, P. D., Gillingham, A. G., & Edwards, W. R. (1997). The impact of widely spaced soil conservation trees on hill pastoral systems. Proceedings of the New Zealand Grassland Association, 59, 171–177.Google Scholar
 Wang, Z., Guo, D., Wang, X., Gu, J., & Mei, L. (2006). Fine root architecture, morphology, and biomass of different branch orders of two Chinese temperate tree species. Plant and Soil, 288, 155–171.View ArticleGoogle Scholar
 Wilkinson, A. G. (1999). Poplars and willows for soil erosion control in New Zealand. Biomass and Bioenergy, 16, 263274.
 Wilkinson, P. L., Anderson, M. G., & Lloyd, D. M. (2002). An integrated hydrological model for raininduced landslide prediction. Earth Surfaces, Processes and Landforms, 27(12), 1285–1297.View ArticleGoogle Scholar