This paper documents the performance of the organic soil version of the Can
adian Land Surface Scheme (CLASS) in modelling the hydrology and energy bal
ance of the Beverly Swamp, Southern Ontario. The hydrometeorological datase
t used to assess model performance begins in the autumn of 1983 and spans 3
3 months, presenting the first multi-year characterization of the area. The
Beverly Swamp receives approximately 900 mm of precipitation per year of w
hich one third is lost to net runoff and the remainder to evaporation. Vert
ical drainage at this site is impeded, due to the presence of a marl layer
below the highly decomposed peat soil, at approximately 1-m depth. This mix
ed-forest wetland is unique among surfaces used for CLASS testing to date.
Within CLASS, vertical drainage at the bottom of the soil profile is set to
Zero to represent the marl subsurface boundary. Preliminary runs have show
n that after each melt period this produced ponded water on site which pers
isted from year to year: The inclusion of a simple lateral drainage functio
n in CLASS simulated actual measured lateral surface flow, and effectively
reproduced seasonal differences in water table position.
Comparisons between measured and modelled diurnally averaged energy budget
components taken from two summers indicate that there is a marked tendency
for CLASS to underestimate latent heat flux (Q(E)) by 29% of the observed v
alues, the major cause of this disagreement being due to systematic error C
oncurrent with this error is an overestimation of the magnitude of soil hea
t storage (Q(G)) by a factor of seven, wherein the error is dominantly syst
ematic. Modifications made to the canopy resistance parametrization, based
on site measurements, resulted in improved model estimates of Q(E), reducin
g the underestimation to 12% of observed values, and changing the major cau
se of error from systematic to unsystematic in nature. The improvement in Q
(E) corresponded with a change in the prediction of sensible heat flux (Q(H
)). A tendency to overestimate Q(H) by 20% of the observed values changed t
o an underestimation of Q(H) by 14%, the error being unsystematic in each c
ase. The modifications resulted in no significant change to either the magn
itude or the nature of the error for Q(G).
Modelled daily average temperatures for the third soil layer vel sus temper
ature measured at 1-m depth (the centre of the layer) indicated that modell
ed values had more extreme minima and maxima, although some of this discrep
ancy could be attributable to the heterogeneous nature of the soil column,
and the unavoidable use of point versus layer average temperatures. Discrep
ancies also exist between measured and modelled snow mass duration and the
timing of melt for three consecutive winters. This suggests that further te
sts of CLASS, using winter season data, must be conducted before it can be
determined if the model is able to correctly simulate snow accumulation and
melt. Wintertime total albedo at this site was also poorly modelled during
the fall and winter periods. Further test runs determined that this overes
timation in total albedo was not contributing significantly to the lower mo
delled soil temperatures or to the persistence of the winter snowpack.
The correspondence between modelled and observed data, particularly given t
he complexity of the canopy and surface at this study site, is adequate but
suggests that further code testing and development initiatives should be d
irected towards improving the simulation of latent and soil heat fluxes, sh
ortwave reflectivity winter snowpack dynamics and surface and subsurface mo
isture transfers, by which are especially important in wetland environments
.