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
10
Rainfall (mm)
1. 2;,3 4.5 6.7.8 9 41; 43 :15- 17
Time (days)
19
21
Figure 3: daily rainfall in Birdlings Flat site during Nov. 2011
time-series of SM for the same period. To reveal the suspected
non-linearity in the relationsship between the two parameter, non-
linear regression analysis also used in the calculation, which did
not provide any improvement in the results. This indicated that
the relationship between LST and SM is more complicated and
needs more careful data analysis. The same complication is vis-
ible when the single time-series of SM are considered (Fig. 2).
Therefore, time-series of SM were divided into two series namely
day and night series to capture the unique variations of SM in
warmer and often drier daylight hours compared to cooler con-
ditions of the night. Nevertheless, no significant improvement
achieved from the correlation of day or night series save few cases
(e.g., Barren LC type both day and night). As a consequence,
time-series were broken into smaller periods.
4.3 Correlations between subsets of LST and SM time-series
following rainfall events
Considering time-series of SM, break-points were chosen based
on the higher points visible in the trends to overcome the sud-
den effects of rainfall events. These higher points, as discussed
earlier, correspond to rainfall events (see figures 2(a) and 2(b)),
therefore, the expected normal SM and surface temperature trend
is significantly interfered by these events. The challenge in this
method was the limited available observations from MODIS LST,
which is limited to four observations in clear and cloudless days.
No significant improvement was observed in the correlations be-
tween combined daily time-series, however, separate time-series
of day and night showed considerable improvements (see Ta-
ble 2). In this case the inverse correlations can be interpreted
from the negative Pearson's r values. Except for the daily time-
series of ‘Irrigated Crop’, all the cases have shown an inverse cor-
relation between MODIS LST and the in-situ SM data. The dif-
ference between day and night series correlation with that of the
combined daily time-series is significant. As a result, it seems the
relationship between LST and near surface SM varies from day
to night, and a change in the direction of the correlation cannot be
ruled out. This lead to an assumption of a nonlinear correlation
between the two parameters. Hence, a non-linear (or curvilinear)
correlation (NLC) fit with a quadratic model was used to calcu-
late correlations between the two parameters. However, except
for few cases, there was no considerable change in the correla-
tion results (see Table 2, second column). Even though corre-
lations for the separate day and night series have increased, the
results still are not substantial. This means the variations in the
two variable happen even in smaller time-frames. Considering
time-series of LST (Fig. 2(b)), largest variations in surface tem-
perature happen about every 12 hours. This can be a clue to look
for a higher inverse correlation between LST and SM from the
20
LC Type r-daily NLC r-daily r-day r-night
Open Grass 0.12 0.12 .—0.51 -025
Forest =0;17 —-0.18 -054 -—0.11
Irrigated Crop 0.01 0.01 -022 —037
Mixed Grass —0.05 —0.06 —0.33 - —0.29
Barren —0.05 —0.04 -0.10 -0.37
Table 2: Pearson’s r values from correlation of LST vs. in-situ
SM after rainfall event on 21° Nov. to 1?* Dec. 2011 over vari-
ous LC types
LC Type r-daily 7-day r-night
Open Grass —0.70 0.31 -0.93
Forest —0.27 0.68 —0.74
Irrigated Crop — —0.58 043 -048
Mixed Grass —0.67 —0.54 —0.68
Barren =0.59 —0.52. —0.88
Table 3: Pearson's r values from correlation of in-situ surface
temperature vs. in-situ SM based on data from 21?* of Nov. 2011
over various LC types
time-series of a single day. Therefore, breaking down the time-
series to even smaller intervals could be one option to further
increase the correlations, however, as mentioned before, there are
not enough observations from the MODIS LST for a single day
(or even a few days). As a consequence, correlations calculated
using the in-situ measured surface temperature and SM data for a
single day is discussed in the next section.
4.4 Correlations between in-situ surface temperature and
SM for a single day
In this section correlations between surface temperature and near-
surface SM in-situ data, both with 30 minute rate, from a sin-
gle day are presented. As the previous sections, correlations are
based on the combined daily time-series as well as the separated
day and night series. Combined daily time-series from all the LC
types showed significant improvement in the correlations (see Ta-
ble 3, first column). Day series correlations are ambiguous, some
of the LC types showed positive while the other LC types showed
relatively strong inverse correlations. Finally, the night series
showed strong inverse correlations for all the LC types. Some of
the LC types, such as forest and the irrigated site, showed lower
correlations which seems is due to the unusual distribution of heat
or moisture on the soil under the effects of the canopy.
5 CONCLUSIONS AND FUTURE WORK
In this paper time-series of MODIS LST product, which is a ther-
mal remote sensing dataset, was compared with the in-situ SM
data over various LC types in Canterbury Plains in New Zealand.
Correlations between time-series of the two parameter for the
month of November (2011) showed insignificant agreement. There-
fore, time-series were broken down to day and night series to
capture possible trends from cool and warmer hours separately.
Nonetheless, correlations were not significant even with data from
smaller time-frames. Lack of continuous observations for a day
from the MODIS product restricted a diurnal analysis using this
dataset, therefore, only in-situ data were used for a single day (as
well as two 12 hours periods for day and night). The agreements
between a single day time-series were significant. These results
indicated that patterns of the relationship between SM and LST
vary during a 12 to 24 hours period, and cannot be captured using
longer time-series of the two variable. Similar results have been
