The Hulugou Watershed (38˚12'N - 38˚17'N, 99˚50'E - 99˚54'E) is regarded as the most typical watershed in the arid and semi-arid Qilian Mountains [33] [34] , it is located at the Qilian Alpine Ecology and Hydrology Research Station of the Cold and Arid Regions Environmental and Engineering Research Institute (CAS), occupies 23.1 km² in area, spans from 2960 m to 4820 m in altitude, and belongs to alpine periglacial landform influenced by the continental alpine climate with the annual temperature from −18.4˚C to 19.0˚C with a mean of 0.2˚C and annual precipitation from 300 mm to 400 mm. The aeolian silty loess is the dominant parental materials of soils, mixed with various glacial debris. Vegetation is mainly alpine shrubs and grasses. The dominant herb species of the profiles are Carex melantha, Kobresia capillifolia and Polygonum viviparum. Soils are relative young and mostly poorly developed, Entisols, Inceptisols and Mollisols are the dominant soil orders [35] , Gelisols only appear above 3600 m and Histosols are sporadically distributed in depressions [1] .

The watershed DEM map (Figure 1) was formed by using the 90 m resolution SRTM3 DEM data, assisted with the information of the parental material, vegetation, terrain and foot accessibility. Eighteen typical profiles of alpine grassland with the same parental material were selected in an altitude sequence (red points in Figure 1).

Sampling was conducted from June to July in 2013. The longitude and latitude of each profile was obtained by portable GPS (Mode: Garmin GPS map 629sc). Each profile was dug to the appearance of parental rocks. Profile thickness and gravel (>2 mm in diameter) volumetric percentage (%) were estimated visually in the field. Soil samples of all genetic horizons were collected and ground to pass 0.149 mm diameter sieve for measuring CaCO₃ content by the calcimeter method. The cutting ring samples of 17 horizons with no or little gravels or grass roots were collected to measure bulk density by the oven drying method [36] , bulk densities of other 37 horizons without cutting ring due to dense gravels or roots were inferred from the correlation between bulk density (y) and organic matter content (x) of 50 genetic horizon samples sampled within and near the watershed (, , p < 0.01, n = 50).

The weighted mean content of CaCO₃ (X) of each profile was calculated as:

where Hi is the thickness of i horizon, Xi is the measured CaCO₃ content (g・kg⁻¹)

of i horizon, H is the profile thickness.

CaCO₃ density (CaCO_3D) of each profile was calculated as:

where Bi is the bulk density of i horizon, Gi is the volumetric percentage of gravels (>2 mm in diameter), k is the horizon number of a profile.

According to the information the longitude and latitude of each profile, the values of the terrain attributes including the altitude, slope, aspect, plane curvature, profile curvature and terrain wetness index (TWI) were obtained from the 90 m resolution SRTM3 DEM data on ArcGIS 9.3 platform (Table 1).

IBM SPSS Statistics 20 was used to test for curve estimation. Origin Lab 8.0 was used to depicted the illuviation mode of CaCO₃.

Results of Curve Estimation between terrain attributes showed that a significant quadratic correlation existed between slope (y), TWI and altitude (x)

(, R² = 0.429, p < 0.05;

, R² = 0.564, p < 0.01;

, R² = 0.355, p < 0.05), significant exponential correlation between slope and TWI (, R² = 0.623, p < 0.01). The above results indicated the complex and irregular spatial distribution pattern of terrain attributes in the Hulugou Watershed.

PT, GV(%) and BD are the three parameters in calculating CaCO₃ density. Table 2 showed the statistical information of the three parameters of the 18 profiles. PT ranged from 10 cm to 135 cm with a mean of 62 cm, GV(%) ranged from 5% to 95% with a mean of 40%, and BD ranged from 0.30 g/cm³ to 1.73 g/cm³ with a mean of 1.02 g/cm³.

Results of Curve Estimation between PT, GV(%), BD and terrain attributes were presented in Table 3, which indicated that altitude, gradient and aspect can be used to predict the PT, GV(%) and BD, while altitude had a higher prediction efficiency in comparison with gradient and aspect. The significant correlation with altitude can mainly be due to the geological erosion process caused by water, gravity and wind. All the 18 profiles actually are located on the hillsides with

slope ranged from 8˚ to 31˚ (Table 1), thus, erosion is inevitable even with dense grass coverage, besides, comparatively soil particles are easier to be eroded and move fast from the high place to low place than the gravels.

The illuviation depth of CaCO₃ in the profiles were ranged from 12 cm to 54 cm in the profiles (20 cm, 12 cm, 15 cm, 38 cm, 40 cm, 28 cm, 54 cm and 35 cm for P8, P9, P11, P13, P15, P16, P17 and P18, respectively, Figure 2(a) and Figure 2(b)). These 18 profiles can be categorized into three kinds of illuviation modes of CaCO₃ according to CaCO₃ contents of different horizons (Figure 2): 1) No illuviation mode, for examples of the profiles above 3600 m in altitude only had one horizon (P1 to P7) or profiles where CaCO₃ leached out intensively from the soil ( −1 in CaCO ₃ content, P10, P12 and P14); 2) Middle illuviation ( Figure 2(a)), CaCO ₃ leached down but accumulated in a certain middle horizon, which were 12 cm - 31 cm, 38 cm - 74 cm and 54 cm - 76 cm in P9, P13 and P17, respectively. CaCO ₃ content in the accumulated horizon were about 7.0, 1.9 and 2.9 times as much as the mean contents of up and down horizons for P9, P13 and P17, respectively); 3) Down illuviation mode ( Figure 2(b)), CaCO ₃

content increased with the increase of profile depth (P8, P11, P15, P16 and P18), the initial depth of illuviation was 20 cm, 15 cm, 40 cm, 28 cm and 18 cm for P8, P11, P15, P16 and P18, respectively, CaCO₃ content was about 9 - 29.5 times as much as the up horizons.

For the three kinds of illuviation modes of CaCO₃, the results of one-way ANOVA LSD tests (Table 4) showed that a significant difference existed only in profile curvature between no illuviation mode and down illuviation mode of CaCO₃ (p < 0.05), which indicated that it was difficult to predict the illuviation mode of CaCO₃ by terrain attributes in the Hulugou Watershed.

Results of Curve Estimation between the illuviation depth and terrain attributes were presented in Table 5, which indicated that altitude and TWI can be used to predicate the illuviation depth of CaCO₃ in the profiles in this area, while TWI is better than altitude and aspect in the accuracy of prediction.

CaCO₃ illuviation depth significantly correlated with altitude and TWI, which can be attributed to the significant correlation between altitude with precipitation and TWI. Three models are available to contact CaCO₃ illuviation depth (D, cm) with precipitation (P, mm), i.e., Retallack model [5] : p = 137.240 + 6.450D + 0.013D² (R = 0.52, n = 807), Zhao model [4] : p = 3.057D + 168.5 (R = 0.96, n = 15), and Pan and Huang model [8] : p = 68.4990 + 12.063D − 0.0693D² (R = 0.725, n = 48). Because no measured precipitation data is available for the 18 profiles, we used the rough information of the lowest and highest precipitation (300 mm and 400 mm) in the Hulugou Watershed to calculate the theoretical CaCO₃ illuviation depth, which was 24 cm - 37 cm, 43 cm - 76 cm, and 22 cm - 34 cm for Retallack model, Zhao model and Pan and Huang model, respectively. Compared with real 12 cm - 54 cm of CaCO₃ illuviation depth of the our profiles, the Retallack model and Pan and Huang model can be regarded more feasible than the Zhao Model which overestimated CaCO₃ illuviation depth. We

*No illuviation: P10, P12 and P14; Middle illuviation: P9, P13 and P17; Down illuviation: P8, P11, P15, P16 and P18; Std.: standard deviation, A, AB and B stands for different significance.

attributed it to the least or insufficient number of soil profiles used in Zhao model.

Zhao [4] pointed out when two, three layers or unusually thick CaCO₃ illuvial horizon exists under the same paleosol or weathering profile, it indicates that there are two or more soil-forming periods and corresponding climate change at that time, but so far, there is no information or model of terrain attributes available on the CaCO₃ illuviation mode in soil profile. Our study also found it is hard to indicate CaCO₃ illuviation mode by the terrain attributes in the irregular and complex alpine region.

For 18 profiles, CaCO₃ mainly exists in the form of mottles and powders in the profiles. No pseudomycelium or concretion was found, which means it is hard for CaCO₃ to crystallize under the semiarid-arid climate of the Qilian Mountains. Table 6 showed that CaCO₃ weighted content of the profiles ranged from 1.30 g・kg⁻¹ to 93.09 g・kg⁻¹ with mean of 23.53 g・kg⁻¹ and standard deviation of 24.19 g・kg⁻¹, while CaCO₃ density ranged from 0.05 kg/m² to 75.69 kg/m² with a mean of 9.79 kg/m² and standard deviation of 18.05 kg/m². Results of Curve Estimation between the weighted content and density of CaCO₃ with terrain attributes were presented in Table 7, which revealed that slope, plane curvature, profile curvature and TWI can be used to predicate weighted content of CaCO₃ in these profiles. Besides, altitude, aspect, gradient and TWI can be used to predicate the

CaCO₃ density. Obviously, TWI had the highest prediction efficiency in comparison with other terrain attributes.

CaCO₃ weighted content had significant negative correlation with profile curvature, it is because the higher of profile curvature, the higher of the change rate of gradient, which may cause more CaCO₃ to be moved away from soil through water.

CaCO₃ density had significant negative correlation with gradient, usually the increase of gradient accelerates the movement of moisture, which promotes the eluviation of CaCO₃. In general, precipitation is higher in the shady slope than the sunny slope, so CaCO₃ density in shady slope should be lower than that of sunny slope. But our study showed a negative correlation between CaCO₃ and aspect, which means that the shady slope contains higher CaCO₃ than the sunny slope. To some extent it can be attributed to the significant negative correlation between the profile thickness and aspect. Significant nonlinear correlation existed between CaCO₃ density and altitude, which is not only attributed to PT, GV(%) and BD, but also to the changes of precipitation and TWI with altitude.

Liu et al. [37] showed that CaCO₃ density of 1 m depth profile ranged from 1.0 kg/m² to 137.6 kg/m² in the alpine grassland of Qinghai Province (the same province as the Hulugou Watershed). Zhang [10] showed CaCO₃ density in 1 m depth profiles in the Loess Plateau ranged from 8.3 kg/m² to 291.7 kg/m². CaCO₃ density in our profiles ranged from 0.05 kg/m² to 75.69 kg/m², which can be regarded within the above ranges.

CaCO₃ illuviation depth is related to altitude and TWI. CaCO₃ weighted content is related to slope, plane curvature, profile curvature and TWI, CaCO₃ density related to altitude, gradient, aspect and TWI, which indicated the data of the terrain attributes are feasible in predicting CaCO₃ content and density in the alpine region. But almost all these correlations are nonlinear and some even with low value of R². This can be attributed to the irregular and complex spatial distribution of terrain attributes. In general, the factors influencing CaCO₃ behaviors in soil is complex, beside precipitation, other factors include topography, parental materials, etc. [8] [38] [39] [40] . Usually precipitation increases with altitude increases, but Niu et al. [41] found that in the north slope of the Qilian Mountains (similar as the Hulugou watershed), the average annual rainfall increases about 17.41 mm from 1700 m to 3300 m in altitude; while decreases about 30.21 mm when the altitude increases 100 m from 3300 m to 3800 mm. It is due to the increase of the wind speed and the decrease of the temperature which cause the rainfall mainly in solid phase, thus, the precipitation actually reduces. Thus using only precipitation it is difficult to predict CaCO₃ characteristics in the alpine region because of the irregular and complex spatial distribution of terrain attributes.

Moreover, the CaCO₃ data of each profile in our study only represent a pedon in 90 m × 90 m pixel of SRTM3 DEM. Actually a 90 m × 90 m pixel usually contains many pedons with different terrain units, the extracted value of a terrain factor used for a profile actually is the mean value of the different pedons in the pixel, this is not the strict one-to-one correspondence and will weaken the accuracy of the study consequences, so DEM data with higher resolution should be found and used in further study.