**Other Abstract** | The Loess Plateau of China has been susceptible to ongoing severe soil erosion.
Among many controlling measures, vegetation restoration is the most economical and
efficient. Soil moisture is the most critical factor affecting vegetation restoration on the
Loess Plateau, and exerts major influences on vegetation growth, agricultural development,
soil erosion, and solute transport. Soil moisture is an integrated response to climate,
vegetation, topography, and soil properties, and is closely related to soil indexes such as
texture, saturated hydraulic conductivity and bulk density. Therefore, knowledge of the
spatial-temporal characteristics of soil moisture and related variables is of great importance
to soil water management and vegetation restoration.
In connection to the spatio-temporal issues of soil moisture and related soil indexes,
and based on a large number of in-situ measurement data and the use of classical statistics
and geostatistical methods, this dissertation mainly focuses on the following issues: the
temporal stability of soil water storage at the hillslope scale; the distribution of
spatio-temporal variability and temporal stability characteristics of water content within
soil profiles (0-300 cm); a feasibility analysis of the a priori prediction of temporal
stability locations; the scaling of temporal stability for surface soil moisture (0-6 cm); the
interpolation accuracy for seven soil properties at various sampling scales; and the spatial
scaling of soil saturated hydraulic conductivity in a small watershed. The investigations
were all carried out at the hillslope scale except for the last one. The main results were as
follows:
(1) The temporal stability of soil water storage in different soil layers (0-1, 1-2, and
2-3 m) was strong at the hillslope scale. The temporal stability was stronger with increases
in soil depth based on either the Spearman correlation coefficient or the standard deviation
of relative difference (SDRD) index. Furthermore, the closer two soil layers were within a
given profile and the deeper any two adjacent soil layers were, the more similar was the
temporal pattern. Using the relative difference method, representative locations were
indentified for each soil layer. More locations estimated the mean soil water storage of the study area accurately in deeper soil layers than in shallower layers. However, none of the
locations were able, individually, to represent the mean soil water storage for all three
layers. Temporal variability played a more important role than spatial variability in
determining the number of representative locations.
(2) The soil water storage during this study was more heterogeneously distributed on
the studied hillslope under wetter than under dryer conditions. A linear equation could
describe well the positive relationship between the mean soil water storage and its variance
(p < 0.01). Furthermore, this dependency increased with increasing soil depth. The
determination coefficients between mean soil water storage and their variance, based on
the full dataset, were 0.33, 0.91, and 0.97 for the soil layers of 0–1, 1–2, and 2–3 m,
respectively. The soil water content data were then analyzed at smaller sampling intervals:
10 cm increments between soil depths of 0 and 100 cm; and at 20 cm increments between
the 100 and 300 cm soil depths. The relationships between the mean soil water contents
and their variances were fitted slightly better by a power function than by a linear equation.
The coefficients of determination did not consistently increase down the 0-300 cm soil
profile, but followed the pattern of decreasing between 0 and 30 cm, increasing from 30 to
160 cm, and being relatively constant below 160 cm.
(3) Choosing 200 cm as the maximum soil sampling depth would be sufficient in
areas similar to the study area when the spatio-temporal characteristics of soil moisture are
to be studied. This was justified after identifying three soil sub-layers according to the
profile distribution of spatio-temporal variability and the temporal stability characteristics.
Layer 1, a complex layer (0-60 cm), was considered to be the active root-zone in which the
soil water within the layer was also subject to the strongest effects resulting from climatic
and topographical factors. The multiple influencing factors led to the diversity of the
spatio-temporal characteristics of the soil moisture. Layer 2 was the steadily changing
layer (30-160 cm), in which most of the spatio-temporal characteristics either increased or
decreased at an almost constant rate. This stable rate of change mainly occurred because of
the effects of vegetation and rainfall on soil moisture, which steadily decreased with
increasing soil depth. Layer 3 (160 to 300 cm) was the stable layer. In this soil layer,
vegetation and rainfall had almost no effect on soil moisture. Thus, the variability of soil
properties became the most important factor to the spatio-temporal characteristics of soil moisture in this layer. The loessial soils have homogeneous soil profiles, which leads to the
stability of the soil moisture spatio-temporal characteristics within this soil layer. Therefore,
when spatio-temporal variability and temporal stability characteristics in soil moisture are
investigated, it would be reasonable to choose 200 cm as the maximum soil sampling
depth.
(4) Elevation and clay content of the soil were the dominant factors affecting the
temporal stability characteristics of soil water in the shallow soil layer (0-60 cm). However,
the a priori selection of representative locations based solely on soil properties and
elevation was determined to be infeasible at the present time since predicted locations
differed greatly from those identified by measurement. Therefore, it is necessary to
introduce more variables or to use a more advanced method to obtain more reliable
predictions of the relationships between the indexes of temporal stability and the selected
variables. Furthermore, the relationships between soil moisture and correlated variables
varied in time and space, which limited the application of these empirical models.
Therefore, we concluded that the a priori identification of representative locations is
presently infeasible, and that more work is needed.
(5) The temporal stability characteristics of surface soil moisture (0-6 cm) at the
hillslope scale were scale-dependent. Sampling extent had a stronger effect on the temporal
stability of soil moisture than sampling spacing.For most of the parameters, a logarithmic
equation could express well the relationships between these parameters and sampling
scales. The parameters changed at a greater rate when sampling spacing or sampling extent
was smaller. However, the specific patterns of scaling differed among parameters. For
example, the mean values of the Spearman rank correlation coefficient did not significantly
change with sampling spacing (p > 0.05), but they increased significantly with increasing
sampling extent (p < 0.01). The ratio of the number of sites under diverse dates with
significant temporal stability, at both the 0.01 and 0.05 probability levels, to the total
number of datasets decreased with increasing sampling spacing or decreasing sampling
extent; the range of mean relative difference (MRD) decreased linearly with the increase in
sampling spacing (p < 0.01), and increased logarithmically with the increase in sampling
extent (p < 0.01); the mean values of the SDRD increased logarithmically with the increase in both sampling spacing and sampling extent (p < 0.01), but the increase was more
sensitive to changes in sampling extent.
（6）Sampling scaling had an important effect on the data distribution types and
interpolation accuracy, as defined by G values. The interpolation accuracy was predicted
better by the scaling index than by the classic index or by the geo-statistic index. For the
seven soil properties (clay, silt and sand contents, bulk density, saturated hydraulic
conductivity (KS), surface soil moisture content and soil organic carbon content) the
smaller the sampling extent or the greater the sampling spacing, the greater the probability
that the sample distribution would be normal or log-normal. For all the studied soil
properties, the interpolation accuracy increased with either increasing sampling extent or
decreasing sampling spacing. However, the mean interpolation accuracy varied greatly
among the seven investigated soil properties. To obtain the greatest contribution rate (the
ratio of the G value to the number of samples) under the same sampling extent, sand
content required the fewest number of samples while soil organic carbon content required
the most, and about the same number of samples was required for the other five soil
properties.
（7）The statistical parameters (variance, correlation length and nugget-sill ratio) for
soil saturated hydraulic conductivity were scale-dependent in a small watershed, and
depended differently on the scale triplet, in terms of sampling spacing, sampling extent and
sampling support. With increases in sampling spacing, apparent variance tended to
decrease in a non-significant linear relationship (p = 0.137); as sampling spacing increased
below 1.1 times the “true” correlation length (i.e. below 80 m), the apparent correlation
length decreased slightly but, as spacing increased above 80 m, it notably increased; the
nugget-sill ratio decreased logarithmically with the increase in spacing (p < 0.01). The
three parameters all increased with increasing sampling extent but with different patterns.
When the sampling support increased, apparent variance and nugget-sill ratio decreased
and correlation length increased. The mean coefficient of determination of the fitted
models between the three parameters and sampling spacing, sampling extent and sampling
support were 0.53, 0.96 and 0.83, respectively. Thus, for the soil property, KS, upscaling or
downscaling was more reliable when based on sampling extent than on spacing or support
in this study. Consequently, distributing limited sample locations in a sub-area of the main study area at a higher sampling density is an alternative sampling method, especially in a
more homogeneous study area.
Based on a large number of field measurements of soil moisture and related variables,
a series of issues concerning the temporal stability of soil moisture and the spatial scaling
of related variables were explored in a small watershed on the Loess Plateau. The findings
presented in this dissertation add to the knowledge about the spatio-temporal
characteristics of soil moisture and related variables in semi-arid environments. They are of
benefit to the application of the temporal stability concept to ecological construction and
agricultural production in the Loess Plateau region. They can also add to the data related to
spatio-temporal variability at multiple scales. Moreover, the findings can also be useful
when designing optimal sampling strategies for similar research work.
Keywords: Soil moisture; Temporal stability; Representative location; Interpolation
accuracy; Sampling scale |

## Edit Comment