60 ' 
the half-hour corresponding to the satellite overpass, as surface temperature may vary within one hour, due to 
windspeed and solar radiation variations. Precise description of these measurements can be found in the 
references appearing in Table 2, that presents the values of surface parameters used for the considered targets. 
Most of these surface parameters were measured (Maricopa, Walnut Gulch) or estimated (Les Maures, Gharb) 
in situ from independent measurements. The coherence of the used values with other published values was 
checked. Next section will show the sensitivity of the method to these parameters. 
Study Site 
Target 
Spectral charact. 
a e 
Resistances (s/m) 
ra rep 
Reference 
Les Maures, France 
Forest 
0.15 
0.98 
17 
25 
Vidal etal., 1993 
Bare soil 
0.15 
0.95 
60 
0 
Local estimation 
Gharb, Morocco 
Sugar Cane 
0.20 
0.98 
46-52 
0 
Vidal & Perrier, 1990 
Bare Soil 
0.20 
0.95 
44 
0 
Maricopa, Arizona 
Cotton 
0.20 
0.98 
15 
25 
Alfalfa 
0.20 
0.98 
15 
25 
Moran et al., 1992 
Bare Soil 
0.20 
0.89-0.95 
55 
0 
Walnut Gulch, Az. 
Irrigated crops 
0.20 
0.98 
15 
25 
Local estimation 
Rangeland 
0.20 
0.98 
43 
0 
Moran et al., 1993 & 
1994b 
Table 2 : Surface parameters used for targets 
3 - RESULTS AND DISCUSSION 
3.1. Validation of the method on warm and cold targets 
In a first step, the measured temperatures of warm (resp. cold) targets were compared with the 
predicted ones using eq. (9) (resp. eq. (10)), in order to test the validity of the method for surface temperature 
retrieval. The result is shown on Figure 1 for each study site, and shows a very good agreement between 
measured and predicted temperatures. The global RMS error for all sites is equal to 0.32 K, which gives a high 
level of confidence to the proposed method. 
However, it can be seen from equations (9) and (10) that the precision of the method strongly 
depends on the resistances estimation, especially the aerodynamic resistance ra. This is particularly true for 
Ts max, as typical values of Rn-G for bare soils are around 400 W/m 2 , whereas Rn-G-LEp for canopies usually 
ranges from -100 to 100 W/m 2 . In the cases described here (7jmax around 50°C, Ta around 30°C), an error of 
20% on ra would yield an error of 7% on Ts max, i.e. STs max = 1.4°C, which is wider than the overall RMSE, 
but remains quite reasonable. The same computation would show that error on Ts min would not exceed 0.1 K. 
A good means for getting a better estimation of ra is to do several measurements of surface 
temperature of the identified warm target, in conjunction with the meteorological measurements required by the 
method (global radiation, air temperature and moisture). Equation (9) can then be used to derive a local value 
of ra. Furthermore it may be assumed that this resistance for a bare soil is quite constant and independent of 
windspeed, as thermal unstability partially compensates for windspeed decrease. For the cold target in arid 
environments, advection should also be accounted for as it may yield LE>LEp due to additional heat from 
advection, what may cause some slight additional error on Ts min. 
3.2. Validation of the method in real conditions 
In a second step, the method was applied as it would be in operational conditions. Among the 
available targets, the ones corresponding to the highest and the lowest Landsat TM DN were selected, their 
temperatures were estimated using eq. (9) and (10), and were then used to estimate parameters a and b of eq. 
(3) by a least square method. 
Figure 1 : C 
te 
G 
F< 
and Gharb, a and 
temperatures of tht 
Figure 2 , and again 
H 
comprised between 
than the ones of t 
relatively importan 
162