Working Group I: The Scientific Basis

Other reports in this collection Does uncertainty in land surface models contribute to uncertainties in climate prediction?

Uncertainty in climate simulations resulting from the land surface has traditionally been deduced from offline experiments (see Chapter 7 and Section due to difficulties associated with comparing land-surface schemes when forced by different climate models (e.g., Polcher et al., 1998b). Some work since the SAR has focused on the sensitivity of land-surface schemes to uncertainties in parameters (e.g., Milly, 1997) and on whether different land-surface schemes, coupled to climate models, lead to different climate simulations or different sensitivity to increasing CO2 (Henderson-Sellers et al., 1995).

Figure 8.12: An uncertainty ratio for 10-degree latitude bands; (a) control simulations; (b) difference between the control and doubled greenhouse gas simulations. E is evaporation, P is precipitation, Tscr is screen temperature, cld is the percentage cloud cover and Sn is the net short-wave radiation at the surface. The units on the Y-axis are dimensionless. An asterisk means the value is statistically significant at 95% and a diamond at 90% (see Crossley et al., 2000).

Polcher et al. (1998a) used four different climate models, each coupled to two different land-surface schemes to explore the role of the land surface under 1xCO2 and 2xCO2. The modification to the land-surface scheme tended to focus on aspects of the soil-hydrology and vegetation/soil-moisture interactions (Gedney et al., 2000). To measure the uncertainty associated with surface processes, the variance of anomalies caused by the changes to the surface scheme in the four climate models was computed. The uncertainty in climate models was computed by using the variance of annual anomalies relative to a consensus (the average of all models). This measure takes into account the differences between climate models, as well as the internal variance of the atmosphere. With these two variances, a ratio was constructed to evaluate the relative importance of the uncertainty linked to surface processes in comparison to the uncertainty linked to other aspects of the climate model (Crossley et al., 2000). In Figure 8.12a this diagnostic is applied to zonal mean values over land for the 1xCO2 experiments. The highest values, indicating a large contribution of surface processes to the uncertainty, were obtained for evaporation, the variable most affected by surface processes (the asterisk indicates significance at the 95% of this measure). Surface air temperature was strongly dependent on surface processes in the tropics but at high latitudes its uncertainty was dominated by atmospheric processes. In the high latitudes, the hydrological cycle (characterised by precipitation and cloud cover) was partly controlled by the surface as indicated by high values of the ratio. Overall, Figure 8.12 shows that the contribution to total uncertainty in the simulation of climate resulting from the land surface may be large and varies geographically.

Figure 8.12b displays the same diagnostic but for the anomalies resulting from a doubling of CO2. The maximum uncertainty is concentrated in the tropics and the variables most affected are cloud cover and temperature. The uncertainty in evaporation changes is large but does not dominate as in the control climate. In the Northern Hemisphere a secondary peak was found for cloud cover and to some extent for precipitation, indicating that a significant part of the uncertainties in the impact of climate change on the hydrological cycle originates from land-surface processes. The shapes of these curves are very different in the two figures, indicating that different processes are responsible for the uncertainties in the control simulation and the climate change anomalies. This implies that the sensitivity of land-surface schemes to climate change needs to be evaluated and that it can not be deduced from results obtained for present day conditions. Gedney et al. (2000) analysed these results regionally and found that the simulations differ markedly in terms of their predicted changes in evapotranspiration and soil moisture. They conclude that uncertainty in the predicted changes in surface hydrology is more dependent on gross features of the runoff versus soil moisture relationship than on the detailed treatment of evapotranspiration. The importance of hydrology was also demonstrated by Ducharne et al. (1998) and Milly (1997).

Other work has involved using global climate fields provided by GCM analyses to force land-surface models offline (see Dirmeyer et al., 1999). In this study, different land-surface models use the same atmospheric forcings, and the same soil and vegetation data sets. Model outputs are compared with regional runoff and soil moisture data sets and, where available, to observations from large-scale field experiments. To date, results highlight the differences between land-surface model treatments of large-scale hydrology and snow processes; it is anticipated that these and other trials will lead to significant improvements in these problem areas in the near future.

Uncertainty in land-surface processes, coupled with uncertainty in parameter data combines, at this time, to limit the confidence we have in the simulated regional impacts of increasing CO2. In general, the evidence suggests that the uncertainty is largely restricted to surface quantities (i.e., the large-scale climate changes simulated by coupled climate models are probably relatively insensitive to land-surface processes). Our uncertainty derives from difficulties in the modelling of snow, evapotranspiration and below-ground processes. Overall, at regional scales, and if land-surface quantities are considered (soil moisture, evaporation, runoff, etc.), uncertainties in our understanding and simulation of land-surface processes limit the reliability of predicted changes in surface quantities.

Other reports in this collection

IPCC Homepage