Evaluation of terrestrial carbon cycle models requires different types of data to test processes operating on a range of time-scales from hours to centuries (see Section 3.2.2), including short-term environmental responses of CO2 and water fluxes between vegetation canopies and the atmosphere (e.g., Cienciala et al., 1998), responses of ecosystem carbon balance to interannual climate variability (e.g. Kindermann et al., 1996; Heimann et al., 1997; Gérard et al., 1999; Knorr, 2000; Prentice et al., 2000), and longer-term consequences of historical land-use change (McGuire et al., 2001). Differences and uncertainties in model behaviour have been evaluated through model intercomparison (Cramer et al., 1999, 2001) and sensitivity analyses (Knorr, 2000; Knorr and Heimann, 2001a).
Terrestrial model evaluation has traditionally been carried out as comparisons with in situ field observations of ecosystem variables (e.g., Raich et al., 1991; Foley, 1994; Haxeltine and Prentice, 1996). The largest data sets of relevant field measurements are for NPP and soil carbon content. Other "target" variables include soil moisture, nitrogen mineralisation rate, and the amounts of carbon and nitrogen in different compartments of the ecosystem. Such comparisons have generally shown reasonable agreement between observed and modelled geographic patterns of these variables, but they do not test the time-dependent response of models to environmental variability.
Time-dependent data sets for in situ comparisons are now becoming available, thanks to eddy-covariance measurements of CO2 fluxes (Section 126.96.36.199; Box 3.1). Daily and seasonal cycles of CO2 and water fluxes provide a test of the coupling between the carbon and hydrological cycles as simulated by terrestrial models (Cienciala et al., 1998). Flux measurements are now being carried out on a multi-annual basis at an increasing number of stations, although global coverage remains uneven, with the greatest concentration in Europe and North America and few measurements from the tropics (see Box 3.1). Field campaigns have started to retrieve flux data from more remote regions (e.g., Schulze et al., 1999). The Large-scale Biosphere Atmosphere Experiment in Amazonia, LBA, will yield more comprehensive data on the carbon, water and energy exchanges of tropical terrestrial ecosystems and will allow a more rigorous evaluation of the performance of models in the tropics than has been possible up until now (e.g., Tian et al., 1998). As current models show conflicting responses of global NPP to climate (Cramer et al., 1999), systematic comparisons with seasonal and interannual flux measurements are a priority to reduce uncertainties in terrestrial carbon modelling.
Terrestrial models have also been evaluated at a global scale by comparing simulated ecosystem water balance with river runoff (e.g., Neilson and Marks, 1994; Foley et al., 1996; Kucharik et al., 2000), and simulated seasonal leaf area with satellite observations of "greenness", often based on the normalised difference vegetation index (NDVI) (Field et al., 1995b; de Fries et al., 1995). NDVI data can be translated into estimates of the plant-absorbed fraction of incoming photo-synthetically active radiation (FPAR) (Asrar et al, 1992), which is related to leaf area index (LAI). The first terrestrial model intercomparison showed differences among model simulations of LAI and its seasonality (Bondeau et al., 1999). More recently, it has been shown that constraining a terrestrial model with remotely sensed spatial patterns of FPAR can lead to a reduction of uncertainty in NPP simulations by about one third (Knorr and Heimann, 2001b). Agreement with patterns of remotely sensed FPAR has thus become a standard benchmark for terrestrial models (Haxeltine and Prentice, 1996; Kucharik et al., 2000) and attention has been focused on improving the simulation of LAI and its seasonal variations.
A more direct test of the simulated net exchange of CO2 between the terrestrial biosphere and the atmosphere is provided by comparison with atmospheric CO2 measurements at remote monitoring sites. The comparison requires the use of an atmospheric transport model to simulate CO2 as a passive tracer (Kaminski et al., 1996). The seasonal cycle of atmospheric CO2 shows a strong latitudinal pattern in amplitude and phase, and is dominated by the terrestrial biosphere (Heimann et al., 1998). The ability to simulate this seasonal cycle thus constitutes a benchmark for terrestrial models' response to climate (Denning et al., 1996; Hunt et al., 1996; Heimann et al., 1998; Nemry et al., 1999). Generally, the observed seasonal cycles of CO2 in northern and tropical latitudes can be well simulated, with terrestrial models using NDVI data as input (Knorr and Heimann, 1995), or by fully prognostic models, including DGVMs (Prentice et al., 2000).
Major features of interannual variability of the CO2 increase are also simulated by terrestrial models (Kindermann et al., 1996; Heimann et al., 1997; Gérard et al., 1999; Ito and Oikawa, 2000; Knorr, 2000; Prentice et al., 2000). This finding supports the hypothesis (Section 3.5.2) that terrestrial effects are important in determining the interannual variability of CO2 uptake. During typical El Niño events, terrestrial model results consistently show strongly reduced CO2 uptake or CO2 release by the land. This result has been obtained with a range of models, even when the models differ substantially in the relative sensitivitivities of NPP and heterotrophic respiration to temperature (Heimann et al., 1997; Knorr, 2000). The low CO2 growth rate during the early 1990s has been simulated by some terrestrial models (Prentice et al., 2000; Knorr, 2000).
At the longest time-scales of interest, spanning the industrial period, models of the natural terrestrial carbon cycle show a pronounced response to rising atmospheric CO2 levels as a result of CO2 fertilisation, generally larger than the NPP response to the climate change over this period (Kicklighter et al., 1999). According to CCMLP results, the CO2 increase maintains a lead of NPP over Rh and an increase of the amplitude of the seasonal CO2 cycle (McGuire et al., 2001), consistent with long-term observations (Keeling et al., 1996a), which indicate an increase in amplitude of about 20% since accurate atmospheric measurements began. However, the magnitude of this effect was greatly over- or under estimated by some models, reflecting unresolved differences in the parametrization of the CO2 fertilisation response.
Other reports in this collection