A GCM land surface scheme was used, in off-line mode, to simulate the
runoff, latent and sensible heat fluxes for two distinct Australian ca
tchments using observed atmospheric forcing. The tropical Jardine Rive
r catchment is 2500 km(2) and has an annual rainfall of 1700 mm y(-1)
while the Canning River catchment is 540 km(2), has a Mediterranean cl
imate (annual rainfall of 800 mm y(-1)) and is ephemeral for half the
year. It was found that the standard version of a land surface scheme
developed for a GCM, and initialised as for incorporation into a GCM,
simulated similar latent and sensible heat fluxes compared to a basin-
scale hydrological model (MODHYDROLOG) which was calibrated for each c
atchment. However, the standard version of the land surface scheme gro
ssly overestimated the observed peak runoff in the wet Jardine River c
atchment at the expense of runoff later in the season. Increasing the
soil water storage permitted the land surface scheme to simulate obser
ved runoff quite well, but led to a different simulation of latent and
sensible heat compared to MODHYDROLOG. It is concluded that this 2-la
yer land surface scheme was unable to simulate both catchments realist
ically. The land surface scheme was then extended to a three-layer mod
el. In terms of runoff, the resulting control simulations with soil de
pths chosen as for the GCM were better than the best simulations obtai
ned with the two-layer model, The three-layer model simulated similar
latent and sensible heat for both catchments compared to MODHYDROLOG.
Unfortunately, for the ephemeral Canning River catchment, the land sur
face scheme was unable to time the observed runoff peak correctly. A t
entative conclusion would be that this GCM land surface scheme may be
able to simulate the present day state of some larger and wetter catch
ments but not catchments with peaky hydrographs and zero flows for par
t of the year. This conclusion requires examination with a range of GC
M land surface schemes against a range of catchments.