3.2. Effects of Land Use on Soil Properties
There were statistically significant differences in soil properties between the six land use types, except for bulk density (
Table 3). The mean clay content varied between 1.2% and 9.6%. Multiple comparisons revealed that the clay level under check-dam farmland was significantly higher than other land use types. However, the clay level under shrub land was significantly lower compared with other land use types. Contents of silt (15.7–67.8%) displayed similar patterns as clay in the multiple comparisons. High clay and silt contents in check-dam farmland may result from the deposition of sediments carried by overland flow. The high sand content of the shrub land can be explained due to its location on the top of the hill slopes where clay and silt are lost easily due to soil erosion. Furthermore, the high sand content may result from the influence of desertification to a certain degree. The spatial variation of erosion intensity was generally controlled by topography and increased from the top to the bottom of the slopes [
46]. There was no statistical difference in bulk density among different land use types. The coefficient of variation and spatial variation of bulk density was low in the study area, which may result in small differences in the average values of BD among land use types.
The mean SOM content varied between 1.55 and 7.53 g kg
−1. Multiple comparisons revealed that SOM levels under check-dam farmland was significantly higher than under other land use types, and the SOM content under shrub land was the lowest among all land use types. However, compared with other areas of Loess Plateau studied by Zheng [
47], the mean SOM content (4.47 g kg
−1) was lower in this catchment. This may result from a combination of serious soil erosion and low C inputs. In our prior study [
48], crop residues like millet and soybean stalks were fed to animals, while maize and sunflower stalks were used as fuel for cooking. A little amount of residue was returned to the cropland, resulting in low C inputs. Although TN exhibited a small coefficient of variation, it was significantly different among the six land use types. The TN content displayed the similar patterns as SOM in the multiple comparisons, and this similarity may be associated with SOM influencing nutrient retention and supply [
41].
The TP contents ranged from 0.61 to 1.30 g kg
−1. No significant difference in TP content was observed between terrace farmland and check-dam farmland, but the TP content of terrace farmland was significantly higher than elsewhere. The most important reason for higher TP in terrace farmland may be low runoff potential, which decreases loss of P in terrace farmland. On the other hand, this result can be attributed to better farming conditions (the local farmers prefer to use more fertilizer on terrace farmland). There was no significant difference in TP between the other of land use types. Most of P is held very firmly in crystal lattices in largely insoluble forms, such as various Ca, Fe and AlPO
4, and is also chemically bonded to the surface of clay minerals [
49], which may result in relatively small differences in the average values of TP among land use types.
For most of the soil properties, there were no significant differences between slope farmland, woodland and grassland. This result was not consistent with other studies [
50,
51], which found that woodland and grassland had higher SOM and TN than slope farmland. This may be the result of the change in land use policies since 1999. In order to return to the natural regulating functions of land resources, the Chinese government launched the Green-for-Grain program, aimed at returning barren slope farmland to woodland and grassland. In the study area, however, converting woodlands and grasslands from slope farmland is unable to effectively control soil and water loss because of the low survival ratio of the trees and the short growth period. Furthermore, the former woodlands are sparse, and few plant residues persist.
3.3. Soil Properties Associated with Landscape Position
Significant differences among landscape positions were observed for clay, BD, SOM and TN (
Table 4). Results indicated that clay, mean SOM and TN contents were higher on FV position than for other positions. The lowest clay content was for the CT position, and multiple comparisons of clay revealed that the clay content on the FV position was significantly higher than for the CT and LS position. This result may be attributed to surface run-off and soil erosion. On the one hand, the CT position is the uppermost position of the hillslope and a large amount of clay can be transported to downslope positions by surface run-off with the result that soil structure at the CT position becomes more compact. Therefore, the highest clay content appeared at the FV position. The highest BD occurred at the CT position and was significantly higher than other landscape positions except for the FV position. Similar results were detected by Malo [
52] and in general, the high soil bulk density is a signal of land degradation [
53].
Soil erosion is an important process influencing nutrient loss from ecosystems [
19], and surface runoff is the major transport mechanism for soil nutrient losses in the Loess Hilly area [
54]. Soil organic matter content varied with position on the slope [
55], as the lowest SOM and TN occur at the LS position, which was significantly lower than SOM content at the FV position. The highest value of SOM and TN occurred at the FV position. Although differences for silt and TP among landscape positions were not statistically significant, further observation suggested that they tended to have higher value at the FV position and lower value at the CT position, with similar patterns existing for clay, SOM and TN. The higher values of soil properties occurring at the FV position may result from deposition of soil organic matter, and nutrients eroded from upslope positions.
3.4. Relationship between Soil Properties and Terrain Attributes and Key Environmental Variables in Impacting Spatial Distribution of Soil Properties
The correlation analysis (
Table 5) showed that clay and silt had positive correlations with
β and
Kh, and negative correlations with
CTI.
Kh is a measure of convergence of substance flows, and soil moisture and lateral intrasoil flow increase if
Kh < 0 and decrease if
Kh > 0 [
24,
56]. This result led to positive correlations of clay and silt with
Kh. A negative correlation was found between BD and
β, but BD had a positive correlation with
CTI. SOM had a negative correlation with
CTI,
SPI and
STI and TN had a negative correlation with
STI. The secondary indices,
CTI,
SPI and
STI are parameters related to surface and subsurface water and sediment transport processes. They reflect the spatial distribution of zones of surface saturation and soil water content in landscape and describe potential flow erosion and related landscape processes. All these comparisons resulted in positive or negative correlations between the secondary indices and some soil properties. There was no significant correlation between TP and most topographic attributes, except a negative correlation with
β. Slope gradients influence infiltration, drainage and runoff, and steeper slopes may exhibit lower soil moisture owing to lower infiltration rates, rapid subsurface drainage and higher surface runoff [
57]. These factors influenced redistribution of soil properties, especially for TP, because soil phosphorus loss is mainly induced by runoff.
The correlations between terrain attributes and soil properties indicated that the soil developed in response to the way water flowed through and over the landscape. On the whole, the correlations between terrain attributes and soil properties were relatively weak, and this result was unexpected. Surface soil properties were mostly modified by land management and as McKenzie and Austin [
36] noted, relationships between landform and soils are more strongly expressed in younger alluvial units than in older landscape units.
To explore the influence of environmental variables on soil properties, land use types and topographic attributes were used to explain the variation of measured soil properties using multiple stepwise linear regression. The regression equations presented in
Table 6 explain 13% to 63% of the variability of measured soil properties (
Table 6). Regression models for clay had the highest R
2 value, followed by TN, silt, BD, SOM and TP. With higher resolution, larger scale digital elevation models and more detailed environmental variables, it may be possible to explain a higher proportion of the variance. However, because of high spatial variability of soil properties and unique physical conditions of the loess plateau, it is unrealistic to expect that the methods employed could explain more variance.
Further observations suggested that land use was the dominant factor affecting soil chemical properties. For example, land use and topographic attributes explained 24.7% and 9.7% of the variability for SOM, respectively. For TN, land use and topographic attributes explained 45.6% and 5.6% of the variability, respectively. For soil physical properties, topography was a dominant predictive factor, especially for BD. Land use and topographic attributes explained 23.2% and 39.5% of the variability for clay, respectively. For silt, land use and topographic attributes explained 29.8% and 15.2% of the variability, respectively. These results imply that change of soil physical properties is a long-term process, but soil chemical properties can be changed by human activities in the short term.
The mean prediction error (MPE) and the root mean square prediction error (RMSPE) measuring the bias or precision of prediction models should be as small as possible for unbiased and precise prediction. For all soil properties except TP, MPE and RMSPE were relatively small. A comparison of observed and predicted values of the linear regression model showed that the regression model was precise for soil bulk density (
Figure 2) with the MPE and RMSPE being 0.009 and 0.127, respectively. For other soil properties, there was a high smoothing effect on the predicted values, and the variation between the predicted and observed values was larger compared to soil bulk density. This observation may be attributed to less spatial variability of bulk density. For TP, the predicted result was very poor and only TEL entered the regression model which explained 12.8% of the variability. This result indicates that some other stochastic factors such as human activity may affect TP to some extent, and therefore needs further study.