7. Istanbul 2004
inalysis shows
e of significant
nitoring of the
RSDAC), 2001,
okyo, 16 p.
correction of
in agricultural
| monitoring,
(7), pp. 1471.
OQ Akagi. S$,
erformance for
in Diego, USA,
Redes Neurais
s ASTER para
egido Norte do
to Nacional de
s, Brasil. 115p.
ster (MODIS),
por product.
quiring.huml
‘ca de imagens
nética de redes
Nnversidade de
d from CNPq
om FEMA-MT
nment) and to
10p-MT for the
| to the Earth
AC-Japan), for
of the project
and land use
lian Amazon"
y - Agreement
"Science and
ms", from the
logy (MCT),
environmental
USE OF HIGH RESOLUTION REMOTE SENSING DATA FOR GENERATING SITE-
SPECIFIC SOIL MANGEMENT PLAN
S. S. Ray", J. P. Singh’, Gargi Das" and Sushma Panigrahy”
"Agricultural Resources Group, Space Applications Centre, Ahmedabad - 380015, India- ssray@sac.isro.org
"Central Potato Research Station, Jalandhar - 144 003, India
"Commission VII, Working Group VII/2
KEYWORDS: Precision farming, Soil parameters, IKONOS data, Principal Components, Within-field Variability
ABSTRACT:
This present study explores the use high-resolution multi-spectral remote sensing data for generating within-field soil variability map
as an inputs required for site-specific management of agriculture. The study was conducted for an experimental plot in Central Potato
Research Station of Jalandhar, India. Thirty-five soil samples were collected from the field at regular intervals. The samples were
analyzed for soil organic carbon, available nitrogen, available phosphorus, available potassium and soil texture. Various soil-related
indices were calculated from IKONOS multispectral data, which included Brightness Index (BNI), Hue Index (HI), Saturation Index
(SI), Coloration Index (CI), Redness Index (RI) and three principal components (PC1, PC2 and PC3). Variability of soil and spectral
parameters were analyzed by estimating coefficient of variation (CV). The correlation analysis was carried out to study the
relationship between soil and spectral parameters. Multiple regression models were generated, using stepwise regression technique,
to estimate soil properties from RS data. The results showed that, CV of soil parameters was highest for available P (29.994),
followed by silt percentage (20.8%). Among the spectral parameters the CV was highest for PC3 (161.9%), followed PC2 (101.4%)
and PCI (84.0%). The soil organic carbon, available N and silt content were significantly correlated with spectral indices. The
multiple regression equation between OC and spectral indices was significant with R = 0.733 and F = 6.277. Available N, silt and
sand also formed significant multiple regression equations with spectral parameters. These empirical equations were used to generate
soil fertility variability plans.
1. INTRODUCTION
It is well known that large variation occurs in soil parameters,
even within a field. One of the objectives of precision farming
is to fine tune the management practices to match this
variability, and thus improve the productivity or reduce the cost
of production and also diminish the chance of environmental
degradation caused by excess use of inputs (Pierce and Nowak,
1999). Hence, soil fertility variability map is one of the major
inputs required for site-specific management of agriculture.
Conventionally this is generated by interpolation of soil
samples taken at regular intervals. However, this approach has
the limitation of large area applicability as it is time consuming
and costly. As against the traditional method of soil sampling
and laboratory analysis of soils, image based remote sensing is
an efficient, fast and economically sustainable way to detect
spatial difference in crop and soil conditions within field. It
offers the potential for identifying fine-scale spatial patterns in
soil properties across a field and optimizing soil sampling
strategies to quantify these patterns (Mulla et al., 2000). Several
soil properties, namely, surface condition, particle size, organic
matter, soil colour, moisture content, iron and iron oxide
content and mineralogy have been found to affect their spectral
behaviour (Dwivedi, 2001). Organic matter is the dominant
factor in determining soil spectral behaviour when it is present
in quantities more than 2 per cent (Baumgardner et al., 1970).
According to the study by Coleman and Montgomery (1987),
an increase in soil moisture and organic matter content resulted
in a decrease in the reflectance values. They showed that near
infra red bands (0.76-0.90 um) were best related to organic
127
matter. Similarly, soil texture also significantly influences the
reflectance pattern. Fine textures generally show greater
reflectance than coarse textures (Horvath et al., 1984). In order
to explore the relationships between remote sensing based
spectral reflectance and soil parameters a number of indices,
like, brightness index, hue index, reflectance index, principal
component analysis (PCA), etc are used. The study carried out
by Leone et al. (1995) shows that organic matter is significantly
related to brightness index. Ray et al. (2001, 2002) have shown
the usefulness of using brightness index in quantifying within-
field variability. Suk. et al. (2002) in their study has shown
how the principal component 2 and 4 are strongly correlated to
soil chemical properties like, organic matter, magnesium (Mg),
and potassium (K) contents. In the recent past, very high
resolution remote sensing (RS) data, such as IKONOS
Multispectral is emerging as a tool for such purpose. This
present study explores the use high-resolution multi-spectral
remote sensing data for generating within-field soil variability
map.
2. STUDY AREA
The study was carried in the farm of Central Potato Research
Station (31.16?N latitude and 75.32? E longitude) in Jalandhar,
Punjab state of India. The farm follows potato-wheat crop
rotation. The soil type of the farm was ranging from very deep
sandy loam, very deep loam to very deep clay loam. The study
was carried out during April in one of the fields having an area
of 4.43 ha, where the land was fallow after the harvest of potato
