54
The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part B7. Beijing 2008
instrument was launched on 18 December 1999 on a morning
platform called Terra, and the second was launched on 4 May
2002 on an afternoon platform called Aqua. The Terra overpass
time is around 10:30AM (local solar time) in its descending
mode and 10:30PM in its ascending mode. The Aqua overpass
time is around 1:30PM in its ascending mode and 1:30AM in its
descending mode.
The validations were carried out for five days during summer,
2003. They are Julian days (day number of a year) 206, 209,
217, 220, 221, covering a period from July to August. These
days are selected because they are clear days for all the
validation sites shown in Table 1. The retrievals of MODIS-
based H and Ta (Air temperature) are averaged over 3*3 pixels
enclosing the ground- based site.
Three MODIS land products are used in this study: the 8-day
leaf area indices product (MODI5), daily atmosphere profile
product (MOD07) and the daily Ts product at 1-km resolution
(MOD/MYD11). To examine the validity of using the different
time pairs, the H retrieved using four combinations was
compared: (1) Terra daytime and nighttime, (2) Aqua daytime
and nighttime, (3) Terra daytime and Aqua daytime, (4) Aqua
nighttime and Terra daytime. First, MODIS products cover
Tongyu site were downloaded from website of EOS Data
Gateway and Atmosphere Archive and Distribution System.
Second, reprojecting and AOI-subset were done using MRT (for
land product) and MRTSwath (for atmosphere product). Third,
average situation of clear-sky cases was analyzed using all four
combinations to find out which is optimal. Fourth, more clear-
sky cases were computed to achieve a preliminary evaluation of
the optimal time pair.
The daytime average sensible heat flux is well correlated with
the temporal change in surface radiometric temperature, which
was illustrated in Fig2. The best choice of time interval will be
that which maximizes both of these curves while large
temperature difference could be observed as much as possible.
The result shows that RMSE of 20-40 W m-2 for daily average
sensible heat. In general, the range of errors reported by other
authors in H flux is very variable. [9] Consider around 50 Wm-2
an acceptable error for H. In the literature, errors in the best
cases are around 22 Wm-2 and similar internal error (20Wm-2)
exists in different measurement instruments (lysimeter, eddy-
covariance and BREB). And converge computing procedure
improves most error to less than 2 °C for Ta estimate at 10:30
am. Although quantity of validation field data is small, the
result is encouraging.
Date
DOY
Site
Period RMSE of H RMSE for Ta
25-Jul
206
Crop
0550-152022.69
0.81
28-Jul
209
Crop
0550-152036.34
1.88
5-Aug
217
Crop
0550-1520 23.10
1.54
8-Aug
220
Crop
0550-152026.52
1.04
9-Aug
221
Crop
0550-1520 21.41
1.76
25-Jul
206*
Grass
0550-152031.34
2.16
28-Jul
209
Grass
0550-1520 26.67
1.25
5-Aug
217
Grass
0550-1520 39.14
0.98
8-Aug
220
Grass
0550-1520 30.45
1.48
9-Aug
221
Grass
0550-152037.18
1.88
*Fied Measurement from 1150 to 1250 is missing
4.2
Table I. Simulation result using CEOP Tongyu field data, 2003
4.3 Comparison Bowen Ratio(BR) with field data
The average energy balance closure ratio (EBCR) is about 0.75
in the whole year of Tongyu, CEOP03, which means H+LE is
smaller than expected, Rn-G, in energy balance theory (TABLE
II). For it makes the direct comparison of estimated LE or ET
and field data problematic, analysis is carried out on Bowen
Ratio(H/LE) between estimation and measurement(Table III).
Figure 2:The correlation between surface temperature A rad and
daytime average sensible heat <H> vs the time interval between
Month
EBCR(Crop)
Available
Measurment
EBCR(Grass)
Available
Measurment
^ rad measurement
1
*missing
*missing
0.66
5.85%
2
0.64
59.90%
0.61
88.69%
In that, combination of Terra daytime and Aqua daytime
3
0.66
89.72%
0.63
95.70%
couldn’t’ work out because too short interval (3hrs) and too
4
0.71
24.72%
0.64
41.94%
small change can be observed (2-3K in summer time) and it is
5
0.70
63.51%
0.65
57.19%
within the bias of surface temperature derivation. To satisfy the
6
0.74
100.00%
0.68
100.00%
aforementioned supposition, we use the nighttime measurement
7
0.74
99.06%
0.69
100.00%
in replacement of a virtual morning measurement. There are
also about 3 hours or 3.5 hours between these two
8
0.74
100.00%
0.70
98.59%
measurements. Because sharp slope of radiometric temperature
9
0.73
98.89%
0.69
98.89%
and PBL development happened at this period, the precision of
10
0.73
97.11%
0.69
100.00%
sensible heat flux is better than Terra daytime - Aqua daytime.
11
0.74
97.22%
0.70
100.00%
Considering the bias in different thermal sensors’ calibration
and cloud cover, we choose Terra nighttime and daytime as
12
0.69
100.00%
0.69
100.00%
optimal time pair, and assume that the temperature at Terra
nighttime (10:30pm) is an approximation of these at 7:00am,
which can be proven reasonable using field measurement. Table
I is the model results using this time interval (7:00am and
10:30am).
4.4 Table II. monthly energy balance closure ratio in TOngyu,
2003