462 
Figure 1: Time series of (a) rainfall records, and (b) watershed average soil moistures in the surface 
zone. Dash line represents the streamflow-derived initial condition; solid line represents the remote 
sensing based simulation. PBMR and SAR soil moisture estimates are represented by A and + 
respectively. 
3.3 Simulation Results 
The water balance model described above was used to simulated the hydrologic fluxes for the 
WD38 subwatershed during the period from July 9 to July 20, 1990. Spatially uniform rainfall, 
estimated from two raingages within the watershed, was used to drive the model during the storm 
event. Distributed inputs such as radar rainfall images, however, can be easily accommodated by the 
model. The potential evaporation was calculated using the Priestley-Taylor method, using the 30- 
minute average solar and long wave radiation data measured from Eppley black-and-white and precise 
infrared pyranometers respectively. This data was used to drive the model during the interstorm 
period. The total potential evaporation for the 12-day period was approximately 45 mm. 
The impact of remotely sensed soil moisture data upon simulation is evaluated through compar 
isons between the time series of WD38 watershed average surface zone soil moisture, based on the 
two different initial conditions. The results are shown in Figure 2(b). Rainfall records are plotted in 
Figure 2(a) for references. It can be seen from this figure that, while both simulations correctly reflect 
the weather conditions, the model prediction based on streamflow-derived initial condition appears 
to be too wet. Incorporation of the remotely sensed data significantly improves the simulation results 
in surface soil moisture. We suspect that the observed discrepancy is due to the fact that, during the 
summer seasons, the strong atmospheric demand would cause the surface zone disconnected with the 
lower portion of the soil column which provides the observed base flow in the stream. As a result, 
the standard initialization procedure provides limited information about the state of the surface soil 
moisture. 
Despite the good performance of the new initialization technique, it is noted that, as shown in 
Figure 2, there still exists a large difference between the model simulated and PBMR measured 
surface soil moistures on the second sampling day (i.e. July 15, 1990). This indicates that the model 
prediction error might accumulate in time. This problem is attacked by reformulating the system 
into a Kalman filter framework (Anderson and Moore, 1979) which updates the state of the model 
every time when remotely sensed data becomes available. The preliminary results are shown in 
Figure 3 in which the PBMR measurements are assumed to be error-free. In other words, the surface 
soil moistures are adjusted to the level reflected by the PBMR measurements during the simulation, 
resulting in a number of sharp discontinuities as shown in Figure 3. This framework is particularly 
suitable for the multi-temporal data provided by spaceborne sensors such as the ERS-1 C-band SAR.