id on
| sen-
ever,
max-
lated
ng of
f the
xima
ncide
naly-
d the
(1)
heat
nvec-
riow,
(2)
, un-
clas-
mage
inant
high-
vater,
)osed
e Le
f the
Terra
sepa-
1 SM
were
n the
g the
codes
from
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
Sitename LC type Description
Birdlings Flat Open grass non-irrigated seasonal
grassland
Bottlelake Forest dense forest with spo-
radic logged patches
Rangiora Irrigated crop crop under continuous
irrigation
Rolleston Mixed grass grass mixed with tree,
periodic irrigation
Waimak Gorge Barren bare. soil "in the
Waimakariri river
basin
Table 1: In-situ measurement sites and their LC types
each of the sites. These codes are meant to read data from the
entire list of LST HDF files for the analysis period based on the
coordinates of each test-site and arrange data-fields alongside the
related fields (i.e., LST for day and night fields, quality control
field, view-angle and overpass-time fields). Afterwards, dates
based on HDF filenames and overpass-time field from the LST
SDSs were used in the code to produce sequentially ordered val-
ues as time-series from each test-site. It should be mentioned
that these time-series are restricted to the available MODIS ob-
servations (four times daily in ideal conditions), which is further
restricted to those times when cloudless data were available. An-
other code was written to match dates and times from LST time-
series with the in-situ SM data; this code appends matching data,
alongside with the in-situ SM date-time, as extra columns in the
time-series of each test-site.
3.4 Statistical methods
Pearson's r coefficient of correlation (Eq. 3) is used to calculate
correlations between LST and SM in this paper. Besides Pear-
son's r, squared form of r indicated as R? is usually used in
regression analysis, which is known as the regression coefficient
of determination. In this paper R? is used alongside r in order to
provide an absolute measure of agreement between the two vari-
ables under consideration. It must be clarified that R? as used in
regression analysis is more common when prediction of one vari-
able based on regressor(s) or explanatory variable(s) and accord-
ing to the regression model is the objective of the analysis, while
in correlation analysis R? is only used to express the absolute de-
gree of agreement between only two variables. If the direction of
the correlation is not of interest, R? is easier to use as it provides
a dimensionless scale (ranged between 0 to 1). Besides, R? can
be used to express the magnitude of dependence between the two
variable, and for this end often a percent form of the coefficient
is used. As an example, an R? value of 0.35 from correlation of
SM with LST implies that 35 percent of the variations in SM is
dependent on LST. However, if the direction or sign (i.e., nega-
tive or positive) of the correlation is of interest, such as the case
of SM and LST where an inverse (or negative) correlation is as-
sumed, Pearson's r would contain more information, and easily
can be squared to get the R? value if necessary.
Pe ST el X E) (5 - Y)
E (X; 2 yv (Y. E ya
where n is the number of observations, X and Y are the mean
values of X and Y variables.
(3)
19
4 RESULTS AND DISCUSSION
4.1 Near-surface SM variations based on LC type
Comparison of the in-situ measurements revealed significant dif-
ferences in the volumetric soil moisture over various LC types.
Greatest anisotropy from the dominant trend is seen on the irri-
gated site, while the other LC types have relatively similar trends
over the field measurement period (Fig. 2(a)). Spikes in the SM
visible in most of the sites are due to rainfall events. This can be
interpreted from the rainfall data. Although the rainfall data were
only available in Birdlings Flat site (with LC type ‘Open grass’),
the spikes in SM correspond to the rainfall events in most of the
sites (e.g., 9th. 20% and 21** of November, see Fig. 3). During
the few hours after the rainfall events moisture levels drop signif-
icantly. However, temperature trends dominantly follow day and
night maximum and minimums, respectively (Fig. 2(b)). Apart
from the visual comparison, statistical analysis of the correlations
between the two parameters was necessary to ensure if any long-
term relationship exists between LST and SM in the area, which
is discussed in the next section.
80; E
*- Open Grass
Forest
x
e. Fed Donon ML Irrigated crop
dues air Mixed Grass
i zt Bamen. -. |
_
e
=
9 4
3
3
o
E
2
=
o
E
=
o
> 4
©
o
#
A A,
20 — 3 EI
31 Oct 04 Nov 08 Nov 12 Nov 16 Nov 20 Nov 24 Nov 28 Nov 02 Dec
Time (days)
(a) In-situ SM (30 min rate)
407 TT
~~ Open Grass
Forest
35- * "|rrigated crop -
Mixed Grass
| [+ Barren?
30-| ;
: i
À
In-situ surface skin temperature (€)
0 = i i l = =
31 Oct 04 Nov 08 Nov 12 Nov 16 Nov 20 Nov 24 Nov 28 Nov 02 Dec
Time (days)
(b) In-situ surface skin temperature (30 min rate)
Figure 2: a. Variations of the near-surface (<5 cm depth) volu-
metric soil moisture, and b. variations of surface skin temperature
over different LC types during the field measurements (Nov-Dec.
2011)
4.2 Correlations between LST and SM time-series
Regression analysis was used to discover any long-term relation-
ship between MODIS LST and the in-situ measured SM in the
study area. However, time-series of LST for the month of Novem-
ber showed no significant agreement when correlated against the
