Introduction

Spatially explicit data of soil properties and nutrients are required for better assessment of soil potential for sustainable management of land resources and in precision agriculture. In recent years, availability of high-resolution satellite data and terrain data had provided large opportunities for predictive soil mapping to generate accurate spatially explicit soil maps (McBratney et al. 2003). The variability in soil properties in any landscape is an inherent natural phenomena conditioned by soil-forming factors. Its spatial distribution is also influenced by soil erosion and deposition processes occurring in the landscape. The effects of topography on spatial distribution of soil attributes and hydrologic processes have been intensively studied by several researchers (Moore et al. 1993; Bell et al. 1995; Herbst et al. 2006; Iqbal et al. 2005). Topography controls the flow of water and sediments; hence it mainly affects catenary soil development (Jenny 1941) and the formation of typical patterns of spatially distributed soil attributes and quality. C, N and P are major nutrients in soils governing soil quality and thus vegetation growth. At the regional scale, the distribution and quantity of SOC depend on several factors such as precipitation, temperature, soil texture, soil depth, land management, and vegetation type (Janzen et al. 1998) but at local scale, its spatial pattern is largely governed by topography.

Topography is one of the major factors affecting soil C and N contents at landscape level (Cambardella et al. 1994; Bourennane et al. 2000). Landscape attributes including slope, aspect and elevation, and land use may be the dominant factors impacting nutrients in the soil with homogeneous parent material and a single climate regime (Rezaei and Gilkes 2005). Terrain analysis intends to derive topographic parameters comprising a set of primary and secondary terrain attributes used in digital soil mapping applications (Behrens et al. 2010). Topographic gradients characterize the shape of the land surface thereby dictating the distribution of soil physical and chemical properties (Moore et al. 1991). Umali et al. (2012) investigated the effect of various management practices and topography on the variability of key soil properties. Several researchers (Moore et al. 1993; Zhang et al. 2012) established statistical relationships between topographic indices derived from digital elevation models, and spatially variable soil properties.

Geostatistics is primarily used as a tool to describe spatial patterns by semivariograms and to predict the values of soil attributes at unsampled locations. It recognizes the continuous nature of soils and can account for random variation through modeling the spatial correlation in soil properties often present in the landscape. It also facilitates integration of ancillary information such as terrain parameters available easily at higher resolution for spatial prediction of soil properties (McBratney et al. 2000). It has been extensively used for quantifying the spatial pattern of soil properties and environmental variables (Webster and Oliver 2001; Zhang and McGrath 2004). Geostatistical interpolation techniques such as kriging and co-kriging have been widely used to predict soil properties at unsampled locations and to understand their spatial variation in the landscape (Voltz and Webster 1990; Chien et al. 1997). Spatial dependency of soil properties was improved by incorporating secondary variables, known as co-kriging method (Goovaerts 2000; Triantafilis et al. 2001; Mueller and Pierce 2003; Zhang and McGrath 2004).

The Himalayan landscape witnesses soil erosion by water where topography is a major factor in controlling the erosion and resulting in spatial variation of soil nutrients in the watershed. Thus, the present study aims to: (a) analyze the role of terrain attributes in controlling nutrients such as C, N and P in soil and (b) use terrain attributes and spectral vegetation index as co-variables in improving prediction of spatial distribution of soil nutrients using geostatistical methods.

Study area

The study area is located in the western part of Doon Valley in Dehradun District of Uttarakhand State of India (Fig. 1). The watershed covers an area of 8.05 km2; lies between 30°25′–30°30′N and 77°45′–78°00′E that falls within the Sitlarao watershed, belonging to Asan river system, a tributary of Yamuna River. The CartoDEM was used for terrain analysis. The combination of Cartosat-1 Image and Toposheet 53F/15 was used to delineate the watershed boundary after proper knowledge of study area visit.

Fig. 1
figure 1

Location of the study area

Full size image

The climate of the study area is characterized by humid sub-tropical. The mean annual temperature ranges from 15 °C in winter to 35 °C in summer. The mean annual rainfall received is 2051 mm (1983–2008) and 70 % of rainfall is received during monsoon season (June–September). The soil texture is predominantly sandy loam to loam. The area comprised forest, agriculture and scrub as dominant land use/land cover.

Data used and methodology

Data used

IRS Cartosat-1 stereo digital data (24 December 2008) were used to generate digital elevation model (DEM) and Resourcesat-1 LISS IV digital data of 12 December, 2004 were used to prepare a land use/land cover map. Cartosat Stereo satellite data were processed using ERDAS to generate digital elevation model. 23 Nos. of GCPs were collected in the area using DGPS for geo-rectification. The DEM generated had the spatial resolution of 30 m and the vertical accuracy of RMSE 4.48 m. SOI Toposheet 53F/15 was used for georeferencing of satellite data and in the field survey.

Methodology

ArcGIS (Ver. 9.3) was used to carry out terrain analysis and integration of maps by spatial analysis. ERDAS Imagine (Ver. 9.0) was used to process the digital satellite data to prepare a land use/land cover map. The methodology in detail has been discussed as: (a) soil sampling to analyze total carbon and total nitrogen (TC and TN); (b) deriving primary and secondary terrain attributes from DEM; (c) statistical relationship of terrain attributes and soil nutrients; (d) variogram analysis to study spatial dependency of soil C and N using (1) kriging and (2) co-kriging methods of geostatistics; and (e) validating spatial prediction of soil nutrients in the landscape. Variogram analysis was done to understand the spatial pattern and trend of the data.

Soil sampling and analysis of total soil carbon (TC) and total nitrogen (TN)

Geostatistical analysis requires a closer sampling density to determine spatial dependency parameters for interpolation of data (Yemefack et al. 2005). Therefore, a regular grid sampling (Fig. 2) was carried out in the watershed. A total of 113 samples at spatial distance of 200 ± 25 m were collected and out of it 100 samples were used for prediction and remaining 13 samples for validation. The soil sampling was done in the month of April–May, 2010. Grid sampling positions were determined by marking grids on large-scale (1:10,000) satellite imagery of IRS Cartosat-1 data. Grids in the fields were located with the help of toposheet and ground features identified on the imagery.

Fig. 2
figure 2

Spatial distribution of soil sampling locations in the watershed

Full size image

Geospatial locations of 113 soil sampling points were recorded using Garmin GPS receiver (GPS 12 CX) and soil samples of the surface layer (0–15 cm) were collected. Soil samples distribution in different land use/land cover are described in Table 1. The soils in the area are skeletal, having >35 % coarse fragments in the top layer. All soil samples were air dried to ensure the removal of soil moisture and stored in air tight containers for total carbon (C) and nitrogen (N) analysis using CHNS analyzer (Nelson and Sommers 1982) and available phosphorus (P) by the Olsen and Sommers (1982) method.

Table 1 Distribution of soil samples collected in various land use/land cover
Full size table

Deriving primary and secondary terrain attributes from DTM

Cartosat DEM generated from stereo-cartosat-1 data was used to characterize primary and secondary terrain attributes using ArcGIS ver. 9.3. Primary terrain attributes such as elevation, slope, plan curvature, profile curvature and flow accumulation and secondary attributes such as terrain wetness index (TWI), stream power index (SPI) and sediment transport index (STI) were generated. Terrain indices are used to characterize the water flow over the surface. These indices were used to analyze statistical relationship with soil nutrients. The terrain indices were calculated using Spatial Analyst in ArcGIS (Environmental Science Research Institute, Redlands, CA, USA, v 9.3). A brief description of various terrain indices derived using primary terrain attributes is discussed as below.

Terrain wetness index (TWI)

It provides a relative index of whether a point in a landscape position is likely to receive runoff water by taking the natural log of the specific catchment area divided by the local slope (Beven and Kirkby 1993). It is a well-studied indicator of soil property and soil moisture distribution at different landscape positions. It is defined by:

$${\text{TWI}} = \ln \left( {A_{\text{s}} /\tan \beta } \right)$$
(1)

where A s is specific catchment area and β is slope in radians.

Higher TWI values indicate a point with high contributing area and/or lower slopes.

Stream power index

It is indicative of erosive power of overland flow and could be used to identify suitable locations for soil conservation measures, to reduce the effect of concentrated surface runoff. High value areas show more prone to erosive power of runoff. It is expressed as:

$${\text{SPI}} = A_{\text{s}} {\kern 1pt} \times {\kern 1pt} {\kern 1pt} { \tan }\,\beta$$
(2)
Sediment transport index

It is widely used to characterize the water flows in landscapes. It is suitable for identifying erosion processes. It accounts for the effect of topography on erosion.

$${\text{STI}} = \left( {A_{\text{s}} /22.13} \right)^{0.6} \,\times\,\left( {\sin \beta /0.0896} \right)^{1.3}$$
(3)

Satellite data-derived spectral index

IRS Resourcesat-1 LISS IV (5.8 m) satellite data were used to generate normalized difference vegetation index (NDVI). The spectral index is most widely used for vegetation studies. It is often directly related to other ground parameters such as percent of ground cover, photosynthetic activity of the plant, amount of biomass and surface soil properties etc. NDVI can be used to better capture the spatial variability associated with soil (Rivero et al. 2009). NDVI value of natural vegetation cover (Forest/scrubland) was used as spectral index parameter in spatial prediction of soil properties in the study. NDVI using remote sensing IRS LISS IV was calculated as:

$${\text{NDVI}} = \left( {{\text{NIR}} - {\text{Red}}} \right) /\left( {{\text{NIR}} + {\text{Red}}} \right)$$
(4)

where near infrared (NIR) and Red represent Band 3 and Band 2, respectively, in LISS IV data.

The value for NDVI lies between −1 and +1. Negative values correspond to water bodies, near to zero bare earth or barren lands and near to +1 correspond to highly vegetative lands.

To study correlation of terrain attributes and soil nutrients

Pearson coefficient of correlation (r) and corresponding coefficient of determination (r 2) were computed for all pairs of terrain attributes (predictor variables) and topsoil attributes (predicted variables) to evaluate its predictive capacity. Linear relationships of soil nutrients and terrain attributes such as slope, elevation, As, TWI, STI and SPI were studied using statistical software. Terrain attributes with highest correlation coefficient were selected as co-variable for co-kriging interpolation.

Variogram analysis to study spatial dependency of soil nutrients using geostatistics

The geostatistical analyses were carried out using ArcGIS software (version 9.3) and its extension module of Spatial Analyst (version 2.0). Before applying inferential statistics to study spatial dependency of soil properties, some statistical parameters such as minimum, maximum, mean and SD of the soil properties were calculated to get an impression of how the sample data are distributed (Snedecor and Cocharn 1980). Data were visualized by histogram and box plot using geostatistical software R and Geostatistical wizard of ArcGIS.

Variogram analysis

Computation of variogram also known as semivariogram is a common way of visualizing spatial dependence. An experimental variogram estimates the average semi-variance of all the point pairs in a bin separated by certain vector by the equation:

$$\gamma \left( h \right) = \frac{1}{2N\left( h \right)}\mathop \sum \limits_{i = 1}^{N\left( h \right)} \left( {Z\left( i \right) - Z\left( {i + h} \right)} \right)^{2}$$
(5)

Variogram parameters such as sill, nugget and range provide quantitative expression of spatial structure for the measured property. Sill is the maximum semi-variance that reflects the variability in the absence of spatial dependence. Nugget is the semi-variance as the separation approaches to zero and reflects the unexplained variability of the spatial structure. Range defined as separation distance at which the sill is reached and reflects the distance at which there is no evidence of spatial dependence (Rossiter 2007).

Experimental variograms were computed with geostatistical wizard of ArcGIS for the given dataset and fitted with some theoretical model to explain the local spatial dependence of the soil properties. The computed variogram was then compared on the basis of sum of squared differences to obtain best fitted model.

$${\text{SSD}} = \mathop \sum \limits_{i = 0}^{n} \left( {Z_{i} - Z_{{i^{*} }} } \right)$$
(6)

Corresponding partial sill, nugget and range of the fitted variogram were computed.

The ratio of the nugget to the total sill was assumed to be criterion to classify spatial dependence of soil properties. Ratio lesser than 25 % and higher than 75 % corresponds to strong and weak spatial dependence, respectively (Chang et al. 1998), while ratio between 25 and 50 % indicates moderately high and from 50 to 75 % falls in moderate spatial dependence.

$${\text{NSR}} = \frac{Co}{Co + Cs} \times 100$$
(7)

where Co is the nugget and Cs is the partial sill.

Kriging and co-kriging are typical geostatistical prediction methods used in the study. Ordinary kriging and co-kriging do not require stationarity of the mean over the entire domain. Ordinary kriging uses the local mean in the estimation, whereas simple kriging uses the global mean; thus, simple kriging would be biased if the local mean is different from the global mean. Therefore, the ordinary kriging and co-kriging were selected in this study.

Kriging

It provides a theoretical weighted moving average of the input parameter over the distance between sampling sites (lag distance). It is the prediction procedure using known values and semivariogram to determine the unknown values. Ordinary Kriging, a standard version of kriging is computed as:

$$Z_{o} = \, \lambda_{1} Z_{1} + \, \lambda_{2} Z_{2} + \cdots \cdots + \, \lambda_{n} Z_{n}$$
(8)

where λ i are the associated weights (Isaaks and Srivistava 1989).

Ordinary Co-kriging

It is a geostatistical technique developed to improve the estimation of a variable using the information on other spatially correlated variables which are generally more densely sampled. The variables are called co-regionalized and are spatially dependent.

$$Z_{o} = \, \sum a_{i} Z_{i} + \, \sum b_{j} V_{j}$$
(9)

where the a i (i = 1, 2,…, n) are the weights applied to the z samples and the b j (j = 1, 2,…, m) are the weights applied to the V samples.

Validation of spatial prediction of soil nutrients in the landscape

Prediction accuracy is usually tested by comparing predicted values with measured values and applying some measures of goodness of fit. In the study, 13 random sample points excluded from the total data set were used for validation. The predicted values at each validation point were extracted using point extraction tool of ArcGIS. Residuals were calculated and plotted in EXCEL spreadsheet. To test the goodness of fit, root mean square error (RMSE) was calculated.

$${\text{RMSE}} = \sqrt {\frac{1}{n}\mathop \sum \limits_{i = 0}^{n} ( Z_{i} - Z_{{i^{*} }} )}$$
(10)

To estimate the improvement of spatial prediction of one method relative to the other method, a relative improvement (RI) (Pang et al. 2009) was used to measure. The prediction accuracy of the x-geostatistical method over the reference method was calculated as

$$RI_{x} = \left( {{\text{RMSE}}_{x} - {\text{RMSE}}_{\text{ref}} / {\text{RMSE}}_{\text{ref}} } \right) \times 100$$
(11)

where RMSE x and RMSEref are RMSE value of the geostatistical predicted and reference methods, respectively. The reference method taken in the study was ordinary kriging (OK) method.

Results and discussion

The spatial distribution of soil nutrients in the landscape is discussed as below.

Statistical relationship between terrain attributes and soil nutrients

Topography is one of the most important soil-forming factors in the hilly landscape and causes variability of soil properties. It also affects infiltration and runoff potentials of the landscape and plays a significant role in the redistribution of soil nutrients through water across the landscape (Ovalles and Collins 1986; Kravchenko et al. 2002; Creed et al. 2002). Soil C and N had shown a higher degree of spatial correlation caused by intrinsic factors of the topography that controls the movement and storage of water, sediment and solutes (Sahrawat 2004).

The Pearson correlation analysis was performed to identify the relationships between terrain attributes and total soil carbon (TC), nitrogen (N) and phosphorus (P) (Table 2). The elevation and slope terrain parameters were observed with negative correlation whereas positive correlation with UCA, TWI, SPI and STP terrain attributes. Higher elevation and steeper slope area had lowest TC and N and vice versa. Slope had significant negative correlation with TC (r 2 = −0.64) and N (r 2 = −0.61). Similarly, Guo et al. (2009) showed negative correlation of soil organic matter with slope and elevation in a hilly area in South-Western China. The upslope contributing area (UCA) terrain attribute has been observed to have a positive correlation with TC (r 2 = 0.38), N (r 2 = 0.39) and P (r 2 = 0.29).

Table 2 Pearson correlations coefficient between terrain attributes and soil nutrients
Full size table

Secondary terrain attributes showed cumulative effect of slope gradient and shape that affect the spatial distribution of soil nutrients. It showed that nutrient distribution was largely influenced by erosion and water distribution in the hilly landscape. Among the terrain indices, TWI showed highest correlation coefficient for TC (r 2 = 0.71), N (r 2 = 0.67) and P (r 2 = 0.66) followed by WPI and STI. A highly significant positive correlation between TWI and TC, N and available P was observed whereas negative correlation with slope parameter. Empirically, under natural conditions, Manning et al. (2001) showed a negative correlation of soil moisture with elevation and high soil carbon accumulation in low lying areas. Mueller and Pierce (2003) showed that topographical wetness index (TWI) reflects the redistribution of water over landscape and is the most appropriate secondary variable in determining SOC distribution. Several researchers reported a strong relationship between terrain attributes and soil C at a field scale (Moore et al. 1993; Yoo et al. 2006).

Spatial prediction of soil nutrients

Descriptive statistics of total samples (113 Nos.) revealed that the total carbon in surface soil varies in the range of 0.22–2.66 % with a mean value of 1.23 % and SD of 0.61. The total nitrogen (N) was found to vary in the range of 0.03–0.37 % with a mean value of 0.17 % and SD of 0.09. P in the soil ranged from 3.21 to 79.48 ppm with mean value of 32.21 and SD of 17.17. The data showed skewness of 0.31 for TC, 0.14 for TN and 1.12 for P. Skewness indicates departure of data from normality, and a value of <1 denotes normal distribution of the data. A logarithmic transformation was considered where the coefficient of skewness is greater than one (Webster and Oliver 2001). Therefore, a logarithmic transformation was performed for P to make the data Gaussian in nature.

Ordinary kriging: semivariogram of soil nutrients

In the study, isotropic variogram was computed for soil TC, N and P as no significant differences in spatial dependency were observed by anisotropic semivariogram. A spherical theoretical covariance model suitable for spatial fields was fitted to experimentally derived covariance values for minimum RMSE value. The semivariogram model with the smallest residual sum of squares was selected as the best fitting model. The semivariogram model parameters for TC, N and P were computed with the fitted model (Table 3). The range of the semivariogram for TC was larger than N and P. The range describes the distance at which values of one variable become spatially independent of another (Kerry and Oliver 2007). It is used to know the soil sampling interval for future survey. Sill indicates the lag distance between measurements at which one value for a variable does not influence neighboring values.

Table 3 Model parameters of isotropic spherical semivariograms calculated for selected soil properties
Full size table

The variogram for TC, N and P showed lowest nugget variance for TC followed by N and P. The large nugget value for P compared to TN and TC showed more variations of its value over small distances. The small nugget value exhibits adequate sampling density to reveal the spatial structure. The range for all three nutrients varies from 278 to 301 m. It supported appropriate soil sampling distance of 200 m for establishing spatial autocorrelation for these soil nutrients in the watershed. It validates a good spatial structure of interpolated soil nutrients maps in the study (Goovaerts 1997).

The nugget to sill ratio (NSR) was assumed to be the criterion to classify spatial dependence of soil properties. Ratio lesser than 25 % and higher than 75 % corresponds to strong and weak spatial dependence, respectively (Chang et al. 1998), while ratio between 25 and 50 % indicates moderately high and from 50 to 75 % falls in moderate spatial dependence. The NSR (%) values for kriging semivariogram were 65, 69 and 70 % for TC, TN and available P, respectively, revealing its moderate spatial dependence. Zhang et al. (2012) reported moderate spatial correlation for soil organic matter based on C0/sill ratio.

Co-kriging method for spatial prediction of soil nutrients

The incorporation of terrain variables into a regression kriging model is a suitable method for increasing the accuracy of prediction of these soil properties (Sumfleth and Duttmann 2008). Terrain indices i.e. terrain wetness index (TWI) and slope were found to be most correlated with soil nutrients (TC and TN) and TWI with available P. The terrain indices can be used as secondary variable to characterize the spatial trend of primary variables (soil nutrients) to generate interpolated map using ordinary kriging. Co-kriging was applied to soil variables having significant correlations with terrain attributes. The semivariogram model parameters using terrain parameters as co-variables for TC, N and P were computed with the fitted model (Table 4). Use of terrain parameters as co-variables in fitting semivariogram model has lowered the model parameters values significantly for all the nutrients showing significant improvement of spatial predictability of soil TC, N and P in the landscape. The nugget effect (variance) has reduced to very small (0.19–0.21) and the values for all the nutrients were very close. NSR has reduced to 44.28, 49.41 and 30.72 % for TC, N and available P and has been categorized as moderately high spatial dependency. The NDVI value of the forest area was taken as co-variable along with TWI for co-kriging and that had improved the correlation coefficient for TC (r 2 = 0.62) but was not observed in case of N and P. Reduction of the RMSE values to large extent for all three nutrients was also noticed.

Table 4 Model parameters of isotropic spherical semivariograms calculated for selected soil properties using co-variables
Full size table

Remote sensing, in combination with multivariate geostatistical methods, has the potential to improve the prediction of soil properties at landscape scale. The normalized difference vegetation index (NDVI) had been widely used for evaluating the spatial pattern of vegetation type and its productivity (Gamon et al. 1993). Yan-li et al. (2013) examined directly the soil parameters with the NDVI. Fabiyi et al. (2013) investigated the correlation between soil fertility and NDVI using geospatial techniques in the south-western Nigeria. Rivero et al. (2009) predicted soil total phosphorus (TP) using remotely sensed data and ancillary landscape properties as supporting variables. Sahrawat (2004) also observed spatial distribution of total C, N contents in relation to the NDVI. Therefore, NDVI has been used as secondary variable along with terrain parameters in analyzing spatial distribution of soil nutrients in the natural vegetation cover (forest land) area of the watershed.

Validation and improvement of different geostatistical prediction methods

Coefficient of determination (r 2) and root mean square error (RMSE) for geostatistical prediction methods (Table 5) showed that the ordinary kriging had the lowest coefficient of determination and highest RMSE values as compared to co-kriging with slope, TWI and TWI + NDVI. As ordinary kriging considers only primary soil variable whereas co-kriging takes into account the secondary variables that helped in improving spatial prediction (López-Granados et al. 2005; McBratney et al. 2000; Odeh and McBratney 2000). Among the primary and secondary terrain attributes, TWI showed highest correlation with soil nutrients (Table 2). Thus, TWI along with NDVI was used as co-variables in co-kriging that has improved the correlation coefficient for TC (r 2 = 0.62), N (r 2 = 0.58) and available P (r 2 = 0.53). RMSE computed was lowest with co-variables of TWI + NDVI for TC (0.17), N (0.067) and available P (7.15) followed by TWI and slope parameter. Higher RMSE for available P was observed in comparison to TC and N using all the methods due to large variation of available P in the soil. The minimum RMSE indicated to the high correlation of soil nutrients with TWI + NDVI parameter. Ordinary kriging had high RMSE for P showed low correlation, whereas the correlation was improved by integrating TWI and TWI + NDVI as co-variables. Spatial map for total carbon, nitrogen and available phosphorus generated using co-kriging with TWI and TWI + NDVI is shown in Fig. 3a–c. However, ordinary kriging had higher RMSE than co-kriging which showed improved prediction highlighting the benefit of residual kriging. The co-kriging with TWI + NDVI, TWI and slope as co-variables had improved the performance with respect to the ordinary kriging by 60.46, 55.81, 44.18 % for TC and 33.63, 21.78, 17.82 % for N, respectively. The prediction accuracy for P was improved with co-variable of TWI + NDVI and TWI by 30.03 and 4.50 %, respectively. It clearly showed that integrating NDVI as co-variables has significantly improved the accuracy for TC followed by N and P. TWI alone as co-variable has improved the performance by 55.81, 21.78 and 4.50 % for TC, N and P, respectively, relative to the ordinary kriging.

Table 5 Prediction results of selected method using RMSE and coefficient of determination as comparison criterion
Full size table
Fig. 3
figure 3

Spatial prediction of soil nutrients by ordinary and co-kriging with TWI + NDVI, respectively, for total carbon (a1, a2) , total nitrogen (b1, b2) and available phosphorus (c1, c2)

Full size image

Conclusions

Topography significantly influences the distribution of soil nutrients in the watershed characterized by hilly topography in the Himalayan landscape. Terrain attributes (primary and secondary) and geostatistical methods were attempted to improve spatial prediction of soil nutrients. Ordinary kriging with primary variable had shown poor correlation resulting in high RMSE. Terrain parameters such as terrain wetness index (TWI) and slope were found most correlated with soil nutrients (TC and TN) and TWI with available P. Among the terrain indices, TWI showed highest correlation coefficient for TC (r 2 = 0.71), N (r 2 = 0.67) and P (r 2 = 0.66) followed by WPI and STI. Secondary terrain parameters had shown high correlation with soil nutrients. Therefore, they were used as co-variable in kriging which had markedly improved the spatial prediction of soil nutrients in the watershed. Co-kriging with the use of TWI and spectral Indices (NDVI) as secondary variables showed the lowest RMSE and highest coefficient of determination. The Co-kriging with TWI + NDVI, TWI and slope as co-variables has improved the performance with respect to the ordinary kriging by 60.46, 55.81, 44.18 % for TC and 33.63, 21.78, 17.82 % for N, respectively. Hence, it can be used as the best prediction method for mapping TC, TN and available P in the watershed of hilly topography. The results showed that the incorporation of ancillary terrain variables can distinctly improve the prediction of soil properties. The spatial prediction of soil nutrients has necessitated the need to identify critical area for sustainable management in the watershed. Spatially explicit data of soil nutrients are required for improved assessment of soil nutrient loss modeling at landscape and watershed scale.