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
issue with these models, however, is the complexity and the risk
of compromise in accuracy if good quality data for all the impor-
tant parameters in the model are not available. To overcome this
issue, an assessment of the relationship between the near surface
SM and LST can help to understand the complexity of these two
parameters over various LC types in a particular region. Such a
relationship can be helpful for approximation of one parameter
(SM or LST) with the availability of the other.
This paper is aimed to analyse the correlation between LST using
MODIS product with the measured near-surface SM based on
geostatistical methods. Daily rainfall data is used to detect the
interfering effects of a sudden rainfall on the surface SM, and to
discover the time-scale for the best correlation with LST.
2 DATA AND STUDY AREA
2.1 Data
Two distinct datasets, one from the remotely sensed MODIS LST
and the second from the in-situ SM measurements provide the
inputs for this analysis. The LST product is a scientific dataset
derived from MODIS moming and afternoon thermal observa-
tions on-board Terra and Aqua satellites. Since the observations
in thermal bands of the sensor are also available at night, the prod-
uct contain night-time values for evening and midnight, which are
collected upon the overpasses of Terra and Aqua satellites on the
night side of the planet. The product is derived from bands 31 &
32 (spectral range 10.78-11.28 uum & 11.77-12.27 um range, re-
spectively). Theoretical background and technical details of the
algorithms and procedure for the extraction of LST from MODIS
thermal bands is available in the literature (Wan and Dozier, 1996,
Wan, 1999, Wan et al.. 2004, Wan, 2008). The dataset is available
in hierarchical data format (HDF) and can be accessed online via
Reverb tool.
The in-situ near-surface (<5 cm depth) SM data used in this
paper have been collected in five sites with various land-cover
types. These data have been recorded using MadgeTech® digital
soil volumetric moisture data loggers known as SMR1 108, Fre-
quency rate of the logged data had been set to every 30 minute.
These measurements have been collected from 1st October 2011
till 7th January 2012, however, for consistency with the other
datasets only Nov. and Dec. data are used in the analysis.
2.2 Study Area
The study area is located in Canterbury Plains in South Island
of New Zealand (Fig. 1), at approximate geographic coordinates
43.54 S and 172.31 E in the central point. The in-situ sites were
selected in the area for the measurement of the near-surface SM.
Criteria for selection of the measurement sites included the area
percentage of the dominant LC types, accessibility of the site and
finally, a minimum of 1x1 km homogeneous extent of the dom-
inant LC type in the area so that at least one pixel from that LC
type to be distinguished in the LST satellite dataset.
3 METHODS
3.1 Empirical relationship between near-surface SM and LST
The principle assumption in derivation of SM from LST data in
physical models used in thermal remote sensing of SM, such as
SEBAL, is partitioning of surface energy to latent and sensible
heat fluxes (Eq. 1 and 2), and to relate the latent heat to the
amount of moisture content in the near-surface layer. LST from
18
0 5 10 15
m— — M
Kilom eters
(OQ SMmeasurement point
Study Area
Soll moisture
measurement sites
Canterbury Plains
New Zealand
Figure 1: In-situ soil moisture measurement points overlaid on
Landsat image (TM5, 28 March 2011)
the thermal remote sensing observations is accounted for the sen-
sible heat part of the energy balance in these models. However,
time-lag between the maximum LST on the surface and the max-
imum solar insolation (Wang and Qu. 2009) can also be related
to the amount of surface SM which contribute to the escaping of
heat from the surface via latent heat. Without allocation of the
heat energy to the Qg by the near-surface SM, the two maxima
(maximum solar insolation and maximum LST) would coincide
or be closer in time. To assess this hypothesis, statistical analy-
sis of the relationship between the measured surface SM and the
remotely sensed LST is implemented in this paper.
Q'-—Qc-cr-Qrg-- Qn (1)
where Q* is the net radiation, Qc is the ground or storage heat
flux, QE is the turbulent latent heat flux, and Q g is the convec-
tive sensible heat flux (Rigo and Parlow, 2005, Rigo and Parlow,
2007). Based on Eq. 1, QE can be calculated as:
Qe=Q" — Qc — Qu (2)
3.2 Land-cover analysis
To identify dominant land-cover classes in the study area, un-
supervised classification method has been used. Iterative clas-
sification process was carried out using a Landsat TM-5 image
acquired on 28 March 2011 over the study area. Eight dominant
LC classes were extracted in the region. With the order of high-
est to the lowest percentage these classes included grass, water,
irrigated crop/grass, bush, baresoil/fallow, water, fallow/exposed
soil, and forest. The in-situ sites were chosen only on five LC
classes due to the issues in finding suitable sites for some of the
LC types (Table !).
3.3 Construction of LST Time-series
Inside LST scientific datasets (SDSs) derived from MODIS-Terra
and MODIS-Aqua, data for each day or night are stored in sepa-
rate data fields. To extract correlations between the in-situ SM
data and LST from MODIS, time-series of LST product were
constructed. Regarding the number of datasets involved in the
analysis, reading day and night values for every day during the
analysis period was a cumbersome task, therefore, Matlab® codes
were written for reading and construction of time-series from
