In this study, a rural highway (Trishuli Highway) crossing a typical mountainous farmland area between Trishuli City and Kathmadu, the capital of Nepal, was selected (See Figure 1). In the area, the annual temperatures range from freezing point to over 30 °C. July, August and September are the monsoon months. The yearly average rainfall is 1,850 millimeters per year. The Trishuli Highway is an all-weather gravel/dirt (4.5 m) track with a 40 mph design speed and low traffic volume of ~1,569 vehicles per day. The vehicles are composed of 33% motorcycles, 15% cars, 27% buses, 24% trucks, and 1% tractors [32]. The highway’s altitude ranges from 730 m to 1,900 m. Due to the complex terrain of the study area, the roadside farmland is either upgrade or downgrade, and closely connected to the road edges.

According to the characteristics of the study site in mountainous areas in the Third Pole region, the analysis followed a 6 × 5 × 2 × 2 MANOVA design for analyzing whether the independent variables Altitude, Distance, Terrain, and Tree have a statistically significant effect on the concentration distribution of four heavy metals (Cu, Zn, Cd, and Pb) in the roadside farmland soil. Table 1 provides a detailed description of the four independent variables. Altitude is defined as the true elevation above mean sea level. Six altitude levels range from 800 m to 1,800 m with increments of 200 m. Distance is defined as the distance from the soil sampling location perpendicular to the road edge. The sampling distances to the highway edge are designated as 0 m, 10 m, 30 m, 50 m, and 100 m, assuming that 100 m samples represent the background heavy metal contents. Two types of roadside farmland terrains (upgrade or downgrade) are designated, assuming that drainage direction to the road corresponding to the roadside slope direction may affect the concentration distribution of heavy metals. Since trees intermittently grow along the Trishuli Highway, whether the trees have a protection effect on the heavy metal pollution was also investigated.

A total of 342 topsoil samples from the depth of 0–5 cm were collected under dry weather conditions along the highway at the six levels of altitudes. Table 2 lists the sample distribution cross-tabulation by the independent variables of Altitude × Distance × Tree × Terrain. It was planned to collect three samples for each cell in order to achieve a full factorial design. However, limited to the on-site conditions, the cases of 1,200 m altitude, downgrade terrain with trees could not be found; the sample was not available in the case of 1,400 m altitude, downgrade terrain without trees at 0 m distance; and the farmland soil was not available in the case of 1,600 m altitude, upgrade terrain with trees at 100 m distance.

As shown in Figure 2, at each sampling location, three sets of samples were collected in three sampling regions with spacing not less than 10 m to minimize their dependency. The sampling distances to the highway edge were designated as 0 m, 10 m, 30 m, 50 m, and 100 m. For each sample, 8–10 topsoil sub-samples were taken in an ‘S-shape’ pattern in a 10 m × 4 m plot and evenly mixed.

Each soil sample was air-dried in the laboratory, pulverized with an agate mortar and pestle, and then sieved through nylon sieve with diameter of ≤0.149 mm. 0.3 ± 0.0001 g sieved soil sample was then mixed with 6 mL HNO₃–3 mL HCl–0.25 mL H₂O₂ in a tetrafluoroethylene (PTFE) beaker and heated in a microwave digestion system (GEM mars). After that, the digested solution was diluted to 50 mL with ultra-pure water and filtered through 0.45 μm microporous membrane. Finally, 1.0 mL filtered solution was diluted to 10 mL for measurement of Pb, Cd, Cu and Zn by Inductively Coupled Plasma-Mass Spectrometry (ICP-MS, Thermo X Series 2). Besides, a series of standard samples were also prepared. Quality controls involved: (1) analysis of 14 random samples, 1 blank samples and 1 national standard samples each time; and (2) random selection of samples to ensure that the relative standard deviation is less than ~10%.

The basic statistical descriptions of heavy metal concentrations (mg/kg) in roadside soils are listed in Table 3. On average, the concentrations of Cu (19.99 mg/kg), Zn (76.30 mg/kg), Cd (0.36 mg/kg), and Pb (22.57 mg/kg) in the mountainous rural farmland topsoil are lower than the tolerable levels, which are considered as phytotoxically excessive: 100 mg/kg of Cu, 300 mg/kg of Zn, 3 mg/kg of Cd, and 100 mg/kg of Pb [33]. However, the maximum Zn, Cd, and Pb concentrations in a few sites are close to or higher than the tolerable levels. The observations indicate that although the average accumulations of heavy metals pose no hazard in the region, some spots with peak concentrations may be severely polluted. For example, the maximum Pb concentration at the edge of highway with 800 m altitude, upgrade terrain, and no tree protection is up to two times higher than the tolerable level.

In order to assess roadside soil quality for heavy metals, pollution index (Pi) is calculated for comparing the heavy metal concentrations of roadside samples within 50 m distance (Ci) to the average concentrations of soil samples taken from 100 m for each altitude as the local background values (Xa). The pollution index (Pi) is calculated as Ci divided by Xa. Table 4 summarizes the descriptive statistical results of heavy metal concentrations comparison between roadside samples and local background values. On average, the Pi values of Cu, Zn, Cd, and Pb fluctuate around one, which do not indicate serious pollution owing to traffic emission. However, the maximum values of Pi are much higher than 1. The maximum roadside concentrations of Cu, Zn, Cd, and Pb are up to 1.99, 5.69, 12.52, and 7.75 times of the local background values. It is found that there is a clear trend that both average and maximum Pi values of Pb are decreasing with the roadside distance. Additionally, the Pi values of heavy metals in the samples without tree protection tend to be higher than those with tree protection, except for Cd.

Except for vehicle emissions, the concentrations of heavy metals in soil can be influenced by other local factors, such as the use of agricultural fertilizers and pesticides, climate and anthropogenic activities. The correlation analysis of the typical soil heavy metals associated with traffic activities will contribute to understand the homology of pollution source. As shown in Table 5, the correlation analysis of the dependent variables shows that either Cu or Cd content in roadside soil is significantly correlated with Zn and Pb content while there is no correlation between Cu and Cd. This finding can be reasonably explained by vehicular heavy metal emission processes. Almost 100% of Cu emission comes from brake wear while 83% of Cd emission comes from engine oil consumption [12]. However, Pb and Zn emissions are rather equally distributed in the mechanisms of fuel consumption, engine oil consumption, brake wear, or tire wear. The correlation pattern indicates that the heavy metal concentrations in roadside soils are associated with traffic contamination although the daily traffic volume in the Trishuli Highway is very low.

In the subsequent statistical analyses, a MANOVA is used to investigate differences between factors (see Table 6). The hypothesis testing in the following analysis is based on a 0.05 significance level. The result indicates that the independent factors are complicatedly associated with the concentrations of Cu, Zn, and Pb. The Cu concentration is significantly influenced by Altitude (p < 0.001), Tree (p < 0.001), and their two-way interaction (p < 0.001); the Zn concentration is significantly influenced by Altitude (p < 0.001), two-way interaction between Altitude and Tree (p < 0.001), and two-way interaction between Terrain and Tree (p = 0.025); and the Pb concentration is significantly influenced by Altitude (p < 0.001), Terrain (p < 0.001), Tree (p = 0.015), two-way interaction between Altitude and Distance (p = 0.012), two-way interaction between Altitude and Terrain (p < 0.001), two-way interaction between Altitude and Tree (p < 0.001), and two-way interaction between Tree and Distance (p = 0.004). However, except for Altitude (p = 0.039), no other factors have a significant effect on concentration of Cd. Meanwhile Cd is the least correlated with other heavy metals. Comparing the sums of correlation coefficient values among the heavy metals, Cd (0.601) is lower than Cu (0.726), Zn (0.794), and Pb (0.783), which suggests that the Cd distribution in Kathmandu region might be affected by some other resources. For example, the phosphorus fertilizer use can add heavy metals to farmland soil [34].

The percentage of oxygen in the atmosphere decreasing with the increment of altitude can influence the efficiency of gas consumption and vehicular emission mechanism. Previous research focusing on the effect of altitude on vehicle on-road emissions indicated that vehicular emissions at high altitude can be much higher than observed at sea level [35,36,37]. As shown in Figure 3, among the six levels of altitude, the Cu concentrations at 1,400 m (M = 13.73 mg/kg; S.D. = 6.05 mg/kg) and 1,200 m (M = 17.08 mg/kg; S.D. = 4.20 mg/kg) are significantly lower than the other levels; the Zn concentration at 1,400 m (M = 38.78 mg/kg; S.D. = 20.17 mg/kg) is almost half of the other levels; and there is a similar trend for Pb distribution by altitude also. It should be noted that the highway segments for sampling at 1,400 m and 1,200 m have large elevation variations. Especially for the 1,400 m segment, it increases 70 m in height within a length of 1,000 m. Such a topographic feature can cause rainfall runoff to carry more heavy metal contaminants down to the lower highway segments or out of the highway through its drainage system. The results indicate that altitude is an important block factor in identifying the other factors effects. However, because the sampling region in this study is fully covered by vegetation, there is no significant difference in percentage of oxygen between the sampling locations with increasing altitudes. Therefore, the assumed pattern that the increasing altitudes may lead to consistently increasing roadside heavy metal concentration level cannot be captured in this study.

Terrain displays a complex impact on Pb concentration in the soil but have no effect on the other metals, except that there is an interaction effect with Tree on Zn, as shown in Table 6. It was found that the Pb concentration in the downgrade roadside soil (M = 18.23 mg/kg; S.D. = 13.97 mg/kg) is significantly lower than that in the upgrade soil (M = 26.62 mg/kg; S.D. = 29.71 mg/kg), as shown in Figure 4(a). This finding is consistent with the previous conclusion that near roadways terrain morphology affects the distribution of Pb contamination and higher concentration values in the upgrade roadside soils as a consequence of obstructed dispersion [38]. This trend is especially significant at the altitude levels of 800 m, 1,000 m, and 1,200 m, as shown in Figure 4(b). The concentration of Pb in the top layer of soils differs between upgrade and downgrade farmland possibly because of the deposition and accumulation of atmospheric particulates from vehicle emission. Road emission was reported as the largest resource for Pb contamination for European countries and lead concentrations in topsoil are spatially heterogeneous [39]. The upgrade roadside soil would serve as a windward slope exposed to the deposition of atmospheric particulates generated from vehicle emission. This effect is further illustrated by Figure 4(c), which indicates that the Pb concentration decreases with the distance from road in the upgrade side of farmland, but it is uniformly distributed in the downgrade side. The rain runoff can also cause lead to transfer from further roadside soil to closer roadside soil since lead is immobilized by the soil [40].

On the contrary, Zn concentration in the downgrade roadside soil (M = 85.26 mg/kg; S.D. = 63.28 mg/kg) is marginally higher than that in the upgrade soil (M = 67.94 mg/kg; S.D. = 38.36 mg/kg), as shown in Table 6 (p = 0.051). Zn contaminants due to vehicle emissions are mainly from engine oil, tire wear, and brake wear [12], and their deposition to roadside soils are greatly influenced by the road surface runoff process. The road surface and roadside runoff process in mountainous terrain is significantly different from that in the flat highways [41]. Therefore, the rain runoff may carry more Zn contaminants to the downgrade farmland soil than the upgrade farmland soil. Intuitively, roadside trees can block more contaminants transferring the contaminants through water to the upgrade farmland, as shown in Figure 4(d).

It is found that roadside trees show a positive effect on the heavy metal concentration control, and concentrations of Cu and Pb in the roadside soils are statistically lower than those without tree protection, as shown in Figure 5. The interactive effects between altitude and tree for Cu, Zn, and Pb show that for some cases, the heavy metal concentrations with roadside trees may be equivalent to or higher than without trees, as shown in Figure 6, which might be caused by unknown local environmental factors that were not covered by this experiment. Planting trees can effectively prevent the pollution particles from depositing on roadside farmland so that more heavy metal contaminants can be expelled into drainage facilities. In addition, trees can be used to remove, transfer, or stabilize heavy metal soil contaminants to render them harmless [42]. In recent studies, phytoremediation has been considered as a promising new countermeasure for in situ cleanup of heavy-metal contaminated soils [43,44].

Generally, the heavy metal content in roadside soil has a belt-shaped distribution in terms of distance to road edge, decreasing exponentially with increment of roadside distance [26]. However, in this study such a distribution pattern by roadside distance is not available, as shown in Figure 7. Even graphically checking the samples site by site, no more than 20 percent of cases display a gradually decreasing distribution by roadside distance. The phenomenon can be explained by several reasons. Firstly, frequent farming activities such as irrigation, plough, and fertilization may mix the farmland top soil spatially and disturb the roadside heavy metal distance-distribution pattern. Secondly, the crops in roadside farmland may have various abilities to assimilate heavy metal soil contaminants. Thirdly, the complex local terrain and environments, such as rain runoff and drainage, wind direction and speed, and other non-crop plants might change the heavy metal contaminants’ distribution patterns in terms of roadside distance. Very few previous studies on roadside farmland soil contamination explored the relationship between heavy metal concentrations and roadside distance. According to the recent research focusing on Pb concentrations in roadside farmland soils, the Pb accumulation due to traffic activities is restricted to within 10 m of the motorway and 3 m of the minor road [45]. At distances greater than these, the other sources make the dominant contribution to the Pb concentration in farmland soil. Therefore, selecting 100 m as the maximum roadside sampling distance in this study would sufficiently cover the vehicle pollution’s scope. However, in nonagricultural land soils, the influential roadside distance of traffic pollution is generally less than 50 m but may be up to 100 m [27,28]. Even, Pb as the most adapted tracer of highway contamination may have an impact on roadside soil up to 320 m [13].