Seventeen global models of terrestrial biogeochemistry were compared with r
espect to annual and seasonal fluxes of net primary productivity (NPP) for
the land biosphere. The comparison, sponsored by IGBP-GAIM/DIS/GCTE, used s
tandardized input variables wherever possible and was carried out through t
wo international workshops and over the Internet. The models differed widel
y in complexity and original purpose, but could be grouped in three major c
ategories: satellite-based models that use data from the NOAA/AVHRR sensor
as their major input stream (CASA, GLO-PEM, SDBM, SIB2 and TURC), models th
at simulate carbon fluxes using a prescribed vegetation structure (BIOME-BG
C, CARAIB 2.1, CENTURY 4.0, FBM 2.2 HRBM 3.0, KGBM, PLAI 0.2, SILVAN 2.2 an
d TEM 4.0), and models that simulate both vegetation structure and carbon f
luxes (BIOME3, DOLY and HYBRID 3.0). The simulations resulted in a range of
total NPP values (44.4-66.3 Pg C year(-1)), after removal of two outliers
(which produced extreme results as artefacts due to the comparison). The br
oad global pattern of NPP and the relationship of annual NPP to the major c
limatic variables coincided in most areas. Differences could not be attribu
ted to the fundamental modelling strategies, with the exception that nutrie
nt constraints generally produced lower NPP. Regional and global NPP were s
ensitive to the simulation method for the water balance. Seasonal variation
among models was high, both globally and locally, providing several indica
tions for specific deficiencies in some models.