de Jong, Steven 
  
5 THE SURFACE TEMPERATURE MODEL 
S.1 Model description 
In this study, a regional model for surface temperature prediction on a sub-catchment scale is built. Most surface 
temperature models however, can only be applied on a point- or small plot scale (Berge, 1986). To apply a surface 
temperature model at a regional scale, the spatial distribution of the input variables must be known i.e. the input 
variables used are maps. The input data used here are field measurements and data derived from the DAIS imagery. For 
additional information and data such as constants and resistance values, the following sources are used: Campbell 
(1985), Feddes et al. (1978), Météo France (1998) and Ward & Robinson (1990). The physical basis of the model used 
here is the *coupled soil moisture and surface temperature prediction model described by Ács et al. (1991). The model 
describes the energy balance of the surface as: 
Gr. FT.) 
ES uui (1) 
ot C 
T, = surface temperature [°K] 
t = time [600 sec. = 10 min.] 
F(T,) = function of surface energy balance components [W m?] 
C, — bulk heat capacity per unit area [J m? K'] 
Where F(T,) is the function of the surface energy balance components: 
F(T,) 2 S- R(T,) - LET, ) - H(T,) - G(T,) Q) 
S — shortwave radiation balance [W m?] 
R(T,) = longwave radiation balance [W m?] 
LE(T,) - latent heat flux [W m?] 
H(T,) = sensible heat flux [W m?] 
G(T,) = heat flux into the soil [W m?] 
In both equations 1 and 2, several components depend on the surface temperature. Consequently, to initialise the model, 
a regional surface temperature map must be available. In this study, the temperature map derived from the DAIS images 
is used as input map. As a result, the model starts at the time and date of DAIS image acquisition. 
5.2 Model results 
A sensitivity analysis of the model revealed that the vegetation type is the most important variable in determining the 
surface temperature. This is confirmed by the high correlation between surface temperature and NDVI. The Root-mean- 
square factor of this correlation is 0.76. An other important factor controlling the modelled surface temperatures is the 
variation of the stomatal- and surface resistance values. These values are necessary for the calculation of the latent heat 
fluxes. The resistance values were taken from literature (Ward & Robinson, 1990) but the original values seemed to 
underestimate the resistance. Therefore, the stomatal- and surface resistance values for these mediterranean 
sclerophyllous vegetation types were slightly enlarged. 
The model is calibrated with the two sets of radiant temperature measurements. The first set consists of radiant 
temperature measurements of all landcover units, collected in the field. These measurements were taken at the moment 
the DAIS flight took place. This set is used to calibrate the average of the surface temperature curve, computed by the 
model. A second model calibration is performed with a temperature timeseries of 12 hours. When comparing the two 
surface temperature curves, it appears that the temperature extremes computed by the model are larger than the maxima 
measured in the field. There are some possible causes for this difference. The first one is the time gap between the 
model and measurement data. An other possible cause for differences, is that the measurements represent only one spot 
of bare soil, while the model results are the average of all the bare soil temperatures in the study area. After calibration, 
the model gives a fairly good average estimate of a 48 hour surface temperature cycle of the study area. The resulting 
temperature curves for the different landcover types are shown in figure 4. 
  
352 International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B7. Amsterdam 2000.