The terrestrial biosphere is believed to play a prominent role in the global carbon cycle at time scales from one year to several decades. Consequently, models are required that simulate vegetation activity on a global scale and are able to predict biogeochemical fluxes of plant and soil carbon, CO₂ and various isotopes.

However, recent biosphere model intercomparisons, some initiated by the International Biosphere Geosphere Programme (IGBP), have shown large discrepancies in model results, such as net primary production (NPP) or the net vegetation-atmosphere CO₂ flux. Those discrepancies probably arise from considerable conceptual uncertainties regarding how vegetation activity should be represented on large spatial scales. As a consequence, reliable validation strategies are needed in order to increase confidence in global vegetation models and eventually allow predictions into the future.

Long-term and global measurements of visible and near-infrared reflectances of the earth's surface carried out on board a series of NOAA polar orbiting satellites constitute a particularly well suited data set to check and improve global vegetation models. Such data have often been converted to vegetation indices (e.g. NDVI) and then translated into biophysical quantities, such as LAI or biomass. This approach, however, leads to serious problems of accuracy, because viewing conditions, soil background colour and atmospheric conditions have a large impact on the signal.

Such difficulties have so far seriously impeded quantitative exploitation of satellite data. Therefore, a different strategy is adopted in the present work that is able to avoid such problems to a large extent. Instead of using satellite measurements directly as input data, the study aims at a synthesis of vegetation modelling and remote sensing technology. It is decisive that the vegetation model developed in this context can be run on its own, without reference to satellite data.

The strategy consists of predicting the satellite signal by a combination of vegetation and radiative transfer models. It is thus possible to consider both errors in the process of measurement and uncertainties of vegetation modelling. As a next step, the model is constrained such that measurements and simulations agree within the estimated range of uncertainties. Only with this approach, quantifiable indications of the usefulness of satellite data for vegetation carbon-cycle modelling can be delivered.

The model developed within the context of this work simulates the photosynthetic rate of land plants embedded within a full energy and water budget of the earth's surface. The link between CO₂ uptake and water loss by transpiration through stomatal control is represented explicitly. Plant and soil respiration are also calculated. Various results with two photosynthesis schemes and different vegetation maps are compared, and the sensitivity of the model against uncertainties implied in several parametrisations is assessed. Mean global NPP is thus calculated to be 76 GtC (billion tons of carbon) with an error estimate of +/- 50 GtC. It appears that the large scatter range between models mentioned above is a true result of cumulated uncertainties involved in a mechanistic description of vegetation activity.

A comparison of simulated satellite data with measurements shows good agreement for most vegetation zones. Deviations exist for boreal coniferous forests (too "green" in the model), tundra (too barren) and for the contrast between wet and dry season in the tropics (too large). Human impact can also be detected in some instances. A constraint of model calculations to fit the satellite data reduces the global error estimate in NPP to +/- 36 GtC. The impact is largest for neadle-leaved forests and tropical savannas. This version of the vegetation model agrees with conclusions by other authors that large parts of the tropical rainforests depend on large soil water storage during the dry season.

Results are checked against the seasonal cycle of atmospheric CO₂ concentration through an atmospheric tracer transport model. Transport is prescribed from routine weather forecasts of high accuracy. It appears that within the modelling uncertainties, the simulations that have been constrained by satellite data all agree with CO₂ measurements - within the error implied by this test. By comparison, there is less agreement for the mean unconstrained simulation, and some simulations within the uncertainty range of the unconstrained model versions clearly contradict the measurements.

The conclusion is that on a global scale, satellite data have at least the same value for determining vegetation activity as have CO₂ measurements in the free atmosphere. In highly productive but water limited areas, most threatened by human impact or a possible climate change, their usefulness is particularly large. Because a multitude of micrometeorological and optical factors can influence the signal, a quantitative interpretation of those data is made possible only by a synthesis of observations and vegetation model simulations. At present, these two data sets probably constitute the main constraint we have on vegetation models.

Download complete text in English by chapter (PDF files):