stract

Soil organic carbon (SOC) and total

soil nitrogen (TSN) are the significant indicators of soil fertility and

biogeochemical cycle. Spatial distribution and variation of SOC and TSN estimation

is central to climate change and sustainable soil management studies. Little

research on spatial prediction of SOC and TSN based on geostatistical

techniques employing secondary variables (sampling location) and auxiliary

information (topographic factors and type of vegetation) has been conducted

globally and under Himalayas in particular. To attempt this ninety-six soils

samples of 0-20 cm depths were taken from small forest area of North Kashmir

Himalayas. The effect of topographic factors-elevation, slope, compound

topographic index (CTI), stream power index (SPI), sediment transport index

(STI), normalized difference vegetation index (NDVI) and vegetation type on SOC

and TSN spatial distribution were studied using regression kriging. Results

indicated regression kriging as better predictor of SOC and TSN spatial

distribution than ordinary kriging with residuals moderately auto-correlated.

Semi-variogram test indicated topographic factors- elevation and slope and

vegetation type as major factors of SOC and TSN spatial variation. The negative

correlation of topographic elevation and slope with spatial distribution of SOC

and TSN reveal better stabilization of SOC and TSN at lower degrees of slope

and lower altitudes. Our study suggest

regression kriging can provide better estimations at larger scale, provided

there is a strong correlation between environmental variables and the SOC and

TSN contents, and residuals are spatially auto-correlated.

Keywords:

Soil organic carbon, total soil nitrogen, spatial distribution, regression

kriging, Kashmir Himalayas

Introduction

The

world climate change studies are centric to carbon-nitrogen cycling. Soil

organic carbon (SOC) and total soil nitrogen (TSN) play an important role in

ecosystem functioning (Gregorich et al., 1994). They act as an important factor

in food and fuel security, reclamation of degraded lands and mitigation of

climate change (Lal, 2004). They act as driving force of agro-ecosystem functions-

controlling soil fertility, water holding capacity and other soil quality factors

(Kosmas et al., 2000; Bangroo et al., 2013).

The

soil biodiversity and soil physical stability is controlled by the spatial

variability of SOC and TSN (Stevenson and Cole, 1999). Therefore, their precise

estimation and spatial distribution is important to comprehend the

carbon-nitrogen dynamics and assist in the decision support system for the

ecosystem recuperation.

The

spatial and temporal SOC and TSN variation with soil and atmosphere is affected

by topographic factors (altitude, aspect and slope), land use/ management,

temperature and soil moisture (Bangroo et al., 2017). Appreciable research is

available on factors affecting SOC and TSN under different physiographic, land

use/management and climatic conditions (Zhang et al., 2012; Peng et al., 2013;

Mondal et al., 2017). A non-uniform SOC and TSN spatial distribution and

correlation with auxiliary information (topography, land use/management,

vegetation and parent material) show a changing continuum on different scales (Tan

and Lal, 2005; Su et al., 2006; Liu et al., 2006).

Attempts

in recent past have been made to assess the SOC

and TSN spatial distribution in relation to these factors by

employing the geostatistical techniques (Kerry and Oliver, 2007; Chai et al.,

2008; Marchetti et al., 2012). Many geospatial prediction models have been developed

to interpolate soil variables into spatially distributed continuum surface from

target sampling points (Harries et al., 2010; Kumar et al., 2012). Not all take

into account the large uncertainty inherent soil spatial heterogeneity such as

ordinary kriging. More recently regression kriging has been extensively used

that combines multiple linear regression using auxiliary information with

kriging and thus incorporates the topography, vegetation and other factors for

higher prediction accuracy.

In

this study, we selected small forest area of Kashmir Himalayan region as a

research site. We used the regression kriging capability to accomplish these objectives

i) to estimate the SOC and TSN spatial

distribution; ii) to evaluate the impact of topographic attributes

and vegetation indices on spatial interpolation accuracy; and iii) to analyze

the spatial prediction accuracy for SOC and TSN using regression and ordinary

kriging methods.

Materials and Methods

Study area

The

Mawer forest range lies between 34° 17? to 34° 22? N and 73° 19? to 74° 59? E

in Kupwara District of Jammu & Kashmir, India (Fig. 1). The area is lacustrine

in origin crossed by the Mawar river. It is spread to an area of 26.1 sq. km

with slope ranging from 15–30% to 30–50%. Being pleistocene and

post-pleistocene in nature the area has good fertility levels. The forests of

the Mawer range is dominated by the coniferous species (Table 1) like Deodar (Cedrus deodara), Himalayan Pine (Pinus wallichiana) and Fir (Abies pindrow). The distribution pattern

of the principal species is influenced mainly by the factors such as altitude,

aspect, and soil. The Deodar and Himalayan Pine on lower belts occur both in

mixtures and in pure stands. The Deodar and Himalayan Pine covers about 44% and

19% of the total area of the commercial forest area of the division. The

broad-leaved species are irregularly distributed throughout the division and

are mainly confined to natural drains, moist depressions, and damp localities.

Soil sampling and soil

analysis

A

total of ninety-six soil samples were selected from Mawer forest range. A 10m x

10m gird sampling design was generated from digital topographic map of forest

range at 1:10,000 scale. The locations of the sampling sites were recorded

using a global positioning system (GPS) receiver (Garmin3790T). Three soil samples

were collected at depth 0–20 cm over a circle of radius 10m surrounding the

specified sampling location and mixed thoroughly. The samples were air-dried

and ground to pass a 2-mm sieve. Nitrogen was determined by Kjeldahl method

(Bremner, 1996) and OC by Walkley and Black method (Nelson and Sommers, 1982).

Acquisition of auxiliary information

The

normalized difference vegetation index (NDVI) was procured from Landsat 8 OLI

and topographic factors like elevation, slope, compound topographic index

(CTI), stream power index (SPI) and sediment power index (SPI) were calculated

from cartosat DEM.

NDVI

is the classical indication of plant health and is used to monitor the changes

in vegetation. The NDVI is closely related to vegetation cover, biomass and the

leaf area index (LAI). The NDVI is given as

where

and

are

the near-infrared and red band spectral reflectance, respectively in the

LANDSAT 8 satellite data. The NDVI ranges from -1 to +1 (NOAA Coastal Service

Centre, 2007). The NDVI derived from the Landsat data along with the terrain

attributes are utilized in the prediction of the soil organic carbon.

Slope and elevation are

derived from the DEM. Both the factors have a strong correlation with the SOC

stabilization (Perruchound et al., 2000; Bangroo et al., 2017).

CTI

is an important aspect of hydrologic system model and provides an indirect

information on land cover and agriculture potential. It is a function of both

slope and upstream contributing area per unit width extraneous to the flow

direction. CTI is defined as

Where

? represents the catchment area per unit width extraneous to the direction of

the flow direction and ? refers to the slope.

SPI is

a measure of the potential flow erosion of water flow based on the assumption

that flow is proportional to the specific topographic surface (Moore et al.,

1991). It takes into account a local slope geometry and site location in the

landscape and measures the erosive power of flowing water at a given point of

the topographic surface. SPI is defined as

Where

?

represents the upstream drainage area (m2/m) and ? refers to the

slope gradient.

STI takes into account the upslope contributing area assuming it to

be directly related to discharge, and slope. STI is defined as (Moore et al.,

1993)

Where

? is the specific catchment area (m2/m), and ? is the slope

gradient.

Regression Kriging Methodology

Sampling

points of SOC and TSN were interpolated in spatial domain by the regression

kriging method (Fig. 2). Regression kriging method can perceive the auxiliary variables

for interpolation of the output at those locations points which is restrained

in the simple kriging method (Hengl et al., 2007). Remote sensing images,

vegetation type, and elevation were considered as common auxiliary predictors

and topographic parameters elevation and slope, normalized difference

vegetation index (NDVI), compound topographic index (CTI), stream power index

(SPI) and sediment power index (SPI) have been used as the predictor variables

here. Regression kriging combines the two approaches of regression and kriging

where regression is applied to fit the explanatory variation and the simple

kriging with an expected value of 0 is applied to fit the residuals, i.e.

unexplained variation (Hengl et al., 2004; Mukherjee et al., 2015):

where,

denotes the interpolated value of the

location,

,

gives

the fitted drift,

denotes the interpolated residual,

stands for the estimated drift model

coefficients (

is

the estimated intercept),

denotes the kriging weights that is determined

by the spatial dependence structure of the residual and

gives

the residual at location

.

Model validation

For

the model validation 29 soil samples out of total 96 samples were randomly

extracted from the data. The mean error (ME) and root mean square error (RMSE)

were used for the model efficiency estimated by comparing SOC and TSN observed

and predicted values from the validation point location. Prediction accuracy

improvement was computed by comparing RK with OK.

where

ME is the mean error; RMSE is the root mean square error; n represents sampling

validation points;

and

are observed and predicted values of the

sampling points, respectively; and R’ is the prediction accuracy improvement from

comparing RK with OK. For positive value of R’, RK has higher prediction

accuracy than that of OK and vice-versa for the negative value of R’.

is

the root mean square error of OK, and

is

that of RK.

Descriptive

and regression analysis of SOC and TSN data was performed in SPSS 20.0 software.

Study area was delineated and topographic factors were extracted from watershed

DEM using ArcGIS 10.2. Spatial and semi-variogram analysis between regression

prediction and residual values of OK were computed and lastly spatial

prediction SOC and TSN distribution maps were produced.

Results

SOC and TSN Descriptive Statistics

The

coefficient of variation (CV), standard deviation, and basic statistical

parameters of mean, range, minimum and maximum are shown in Table 2. The

average SOC and TSN concentration in the study area were 17.74 g kg-1 and

2.31g kg-1 respectively. Both the moderate CV 26.21% and 23.32 % could

be linked to uniform land use pattern, and/or soil erosion.

Correlation between SOC and TSN with

the environmental variables

The

SOC and TSN showed a negative correlation with the elevation (Table 3). This

indicates that the concentration of both SOC and TSN deceases with the

elevation. Similar, correlation was observed with the slope which is an

important soil erosion factor. This reveals that greater the slope more intense

is the soil erosion which results in decrease in SOC and TSN concentrations.

Little

or no correlation of SOC and TSN was observed with CTI, SPI or STI. Correlation

of average SOC and TSN content along the elevation with NDVI was also analyzed

and found to be significant (r2 = 0.673, p<0.001). This indicates
that SOC and TSN increases with an increase in vegetation NDVI.
Spatial variability and distribution
of SOC and TSN
Topographic
factors (elevation, slope, SPI, STI and CTI) and NDVI were used to predict the
spatial variability of SOC and TSN through multiple linear regression method. Among
these, elevation and slope proved to be the optimal factors for the prediction
of SOC and TSN with determination coefficients (r2) of 0.426 (at
P<0.05) and 0.406 (at P<0.001) respectively.
The
regression kriging provided better results for spatial autocorrelation of SOC
and TSN than that of ordinary kriging (Fig. 3). The Nugget/Sill ratio for
regression kriging and ordinary kriging for SOC were 0.28 and 9.81 and for TSN
were 0.24 and 4.59 respectively (Table 4). The semi-variogram analysis showed
that environmental factors such as topography and vegetation were the primary
causes of SOC and TSN spatial variance.
SOC
and TSN were also found to be strongly correlated, with a correlation
coefficient of 0.7121 (P<0.05) and have highly significant linear relationship.
Prediction accuracy of OK and RK
Location
points of SOC and TSN samples were interpolated in spatial domain by the
regression kriging method and using topographic factors (elevation, slope, SPI,
STI and CTI) and NDVI as predictor variables. Regression was applied to fit the
explanatory variation and simple kriging with an expected value of 0 was
applied to fit the residuals, i.e., unexplained variation in regression kriging
method.
Sixty-seven
samples were randomly selected to conduct ordinary kriging interpolation for
regression residual error of SOC and TSN in the study area. In the meantime,
ordinary kriging interpolation was also conducted on these samples as a
control. From the results of prediction errors, regression-kriging was found better
than that of ordinary kriging (Fig. 4). The 29 training samples were used for
model validation and comparison of the two prediction methods (Table 5). Satisfactory results were obtained with
regression kriging with predicted values close to observed ones and much more
detailed concerning the partly variation and topographical relationships than
that of ordinary kriging. The improvements of prediction accuracy (R') of SOC
and TSN were 17.82% and 19.44%, respectively (Table 5).
Discussion
Effect of vegetation type on SOC and
TSN
The
type of vegetation has a significant effect on corresponding changes in
micro-climate in an ecologically fragile environment like of Kashmir Himalayas
which subsequently alter soil nutrient dynamics (Bangroo et al., 2017). The SOC
and TSN concentration in existing dominant vegetation types of the study area
ranked as Pinus wallichina > Cedrus

deodara > Abies pindrow. This suggests that vegetation

type had a significant impact on spatial SOC and TSN patterns. Similar, trend

was observed by Peng et al., 2013 and Garcia et al 2016.

Significant

differences in SOC and TSN in varying vegetation types were observed

(P<0.05), this may be attributed to species composition, stand structure,
and management history (Dar and Sundarapandian, 2015). The thicker forest
litter and well flourished soil plant root system of P. wallichina and C. deodara
fix and more SOC and TSN which cause high accumulation. The shrub biomass was also
found highest under C. deodara in Western
Himalayas (Wani et al., 2016). The study area being a protected forest had less
human intervention and less soil erosion in P.
wallichina and C. deodara belt
which favored SOC and TSN accumulation.
Effect of topographic parameters on
SOC and TSN
Topographic
parameters have a significant effect on the spatial distribution of SOC and TSN
(Mondal et al., 2017). Research indicate that SOC is primarily controlled by
the variation in temperature, and soil moisture which vary with elevation
gradients (Griffiths et al., 2009), aspect (Måren et al., 2015; Garcia et al.,
2016) and slope (Perruchoud et al., 2000). While, the N stock variation with
altitude are partly influenced by vegetation type and partly by altitude
(Bangroo et al., 2017). The correlation analysis revealed negative correlation
of SOC and TSN in our study area with the elevation (Table 3). This may be
attributed to the 1) lower mineralization rate and net nitrification rate at
the higher altitude, 2) decline in total tree density, and species richness
with increasing altitude, and 3) better stabilization of SOC at lower
altitudes. A characteristic decline in vegetation was observed across
altitudinal strata. The decrease in species richness in high elevation strata significant
in Himalayan forests could be due to eco-physiological constraints, low
temperature, and productivity (Gairola et al., 2008; Hardy et al., 2001). The
characteristic decline in vegetation with increasing altitude results in less
accumulation of litter and low input of organic carbon in soils.
We
observed negative correlation of SOC and TSN spatial distribution with the
slope (Table 3). This is attributed to 1) higher rates of erosion with the
slope which increases with increasing rainfall, 2) poor soil development which
results in poor retention of SOC and 3) soil temperature gradients along the
slope under different aspects which affect the rate of SOC decomposition. These
results concur with other findings, Bookhagen et al., 2005, observed lowest
rates of soil erosion at less than 2% of slope and highest at more than 20% of
slope resulting in highest SOC loss. High erosion rate in steep slopes along
with low carbon stock causes further depletion of SOC whereas lower areas have
better retention of the SOC stock.
Conclusion
The
spatial distribution of SOC and TSN across the complex topography of small
forest area of Kashmir Himalaya is better predicted by regression kriging as
compared to ordinary kriging with a prediction accuracy of 17.82 % and 19.44 %
respectively. The spatial autocorrelation of SOC and TSN is better explained by
regression kriging with Nugget/Sill ratio of 0.28 and 0.24 respectively. It is
important to select appropriate environmental variables for interpolation
techniques and semi-variogram analyses showed topographic parameters of
elevation, slope and vegetation type/ land use as major factors influencing the
spatial distribution of SOC and TSN.
Both
elevation and slope has significant influence on spatial distribution of SOC
and TSN concentrations. Negative correlation of SOC and TSN with elevation
indicate better stabilization at lower altitudes. High degree of slope has low
vegetation and high soil erosion rate which leads to low SOC and TSN
concentrations.
In
conclusion, regression kriging can provide better estimations at larger scale,
provided there is a strong correlation between environmental variables and the
SOC and TSN concentrations, and residuals are spatially autocorrelated.