International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XXXIX-B7, 2012
XXII ISPRS Congress, 25 August — 01 September 2012, Melbourne, Australia
GEOSTATISTICAL ANALYSIS OF SURFACE TEMPERATURE AND IN-SITU SOIL
MOISTURE USING LST TIME-SERIES FROM MODIS
M. Sohrabinia*!, W. Rack? and P Zawar-Reza!
! Department of Geography, Atmospheric Research Centre
? Gateway Antarctica, Centre for Antarctic Studies & Research
University of Canterbury
Christchurch, New Zealand
sohrabinia.m Q gmail.com, wolfgang.rack @canterbury.ac.nz, peyman.zawar-reza 9 canterbury.ac.nz
http://www.canterbury.ac.nz/
Commission VII/1
KEY WORDS: MODIS, MODIS LST, skin temperature, near-surface soil moisture, land cover, geostatistics, time-series
ABSTRACT:
The objective of this analysis is to provide a quantitative estimate of the fluctuations of land surface temperature (LST) with varying
near surface soil moisture (SM) on different land-cover (LC) types. The study area is located in the Canterbury Plains in the South
Island of New Zealand. Time series of LST from the MODerate resolution Imaging Spectro-radiometer (MODIS) have been analysed
statistically to study the relationship between the surface skin temperature and near-surface SM. In-situ measurements of the skin
temperature and surface SM with a quasi-experimental design over multiple LC types are used for validation. Correlations between
MODIS LST and in-situ SM, as well as in-situ surface temperature and SM are calculated. The in-situ measurements and MODIS data
are collected from various LC types. Pearson's r correlation coefficient and linear regression are used to fit the MODIS LST and surface
skin temperature with near-surface SM. There was no significant correlation between time-series of MODIS LST and near-surface SM
from the initial analysis, however, careful analysis of the data showed significant correlation between the two parameters. Night-time
series of the in-situ surface temperature and SM from a 12 hour period over Irrigated-Crop, Mixed-Grass, Forest, Barren and Open-
Grass showed inverse correlations of -0.47, -0.68, -0.74, -0.88 and -0.93, respectively. These results indicated that the relationship
between near-surface SM and LST in short-terms (12 to 24 hours) is strong, however, remotely sensed LST with higher temporal
resolution is required to establish this relationship in such time-scales. This method can be used to study near-surface SM using more
frequent LST observations from a geostationary satellite over the study area.
1 INTRODUCTION
Near surface soil moisture (SM), defined as the water content of
the upper 10 cm of the soil (Wang and Qu, 2009), is measured by
remote sensing satellites using the electromagnetic radiation in
three distinct ranges: the visible and near-infrared region, thermal
region and the microwave region. Image analysis and interpreta-
tion techniques such as soil wetness indexes, directly (Bhagat,
2009) or indirectly (Sgrensen et al., 2005, Grabs et al.. 2009) us-
ing remotely sensed data in the visible and near-infrared region
are applied to estimate wetness of the near-surface soil layer. The
algorithms used in the thermal region are based on the surface
energy balance. These algorithms are based on the partition-
ing of the net surface energy to the sensible, latent and ground
heat fluxes. With the knowledge on the sensible and ground heat
fluxes (Qr and Qa), which is based on the land surface temper-
ature (LST) and the ancillary data about the surface types under
consideration, the latent heat flux (Qz) is estimated (see Eq. 1
and 2). Qzg is used as an indicator of the amount of water con-
tent in the near surface soil layer. Microwave remote sensing
techniques in SM analysis rely on known dielectric properties of
the soil and water (Jackson et al., 1996). The advantage of mi-
crowave sensors is the availability of the observations in almost
all-weather conditions, which enables more frequent data acquisi-
tion; however, compared to the moderate resolution thermal sen-
sors, the spatial resolution of these sensors is coarse (Hain et al.,
2011). Other works have used a combination of optical, thermal
and microwave remote sensing data (Wang et al., 2004, Hassan
et al, 2007, Gruhier et al., 2010, Hain et al, 2011). More com-
plex methods such as the Soil-Vegetation-Atmosphere-Transfer
17
(SVAT) model (Carlson et al, 1994) exploit combination of the
remotely sensed data to establish a relationship between surface
SM, surface temperature and vegetation cover. Considering the
objective of the current research, thermal remote sensing algo-
rithms are of interest in this paper.
LST product is one of many datasets derived from day and night
observations of the Moderate Resolution Imaging Spectroradiome-
ter (MODIS) twin sensors on-board Terra and Aqua satellites,
which is a wealth of data covering most of the land masses of
the globe over the last 10 years. With a more frequent overpass
than Landsat (near-daily) and higher spatial resolution (250, 500
and 1000 m) than some of its predecessors such as the Advanced
Very High Resolution Radiometer (AVHRR), MODIS provides
a comprehensive series of land, ocean, and atmosphere obser-
vations (LPDAAC, 2010). The LST product is suitable for use
in a variety of research including soil and water resources, agri-
culture, climate and atmospheric modelling and research. Infor-
mation on LST is necessary for parameterization of land surface
processes in numerical models (Sun and Pinker, 2004). LST is
dependent on the incoming shortwave and longwave radiation,
but also landcover (LC) type and the amount of near surface SM.
Surface SM affects diurnal change of surface temperature, and it
is a key variable in computing several parameters of the land en-
ergy and water budget (Zhang et al., 2007). LST is used for SM
assessment using rigorous physical models, such as Surface En-
ergy Balance Algorithm for Land (SEBAL), which estimates SM
based on parameterization of surface heat fluxes (Bastiaanssen,
2000). MODIS LST product archived for more than 10 years is a
valuable data source which can be used in these algorithms. The
