Uncertainties in estimates of RF due to anthropogenic surface albedo change arise from several factors.
220.127.116.11.1 Uncertainties in the mapping and characterisation of present-day vegetation
The RF estimates reported in the TAR used atlas-based data sets for present-day vegetation (Matthews, 1983; Wilson and Henderson-Sellers, 1985). More recent data sets of land cover have been obtained from satellite remote sensing. Data from the AVHRR in 1992 to 1993 were used to generate two global land cover data sets at 1 km resolution using different methodologies (Hansen and Reed, 2000; Loveland et al., 2000) The International Geosphere-Biosphere Programme Data and Information System (IGBP-DIS) data set is used as the basis for global cropland maps (Ramankutty and Foley, 1999) and historical reconstructions of croplands, pasture and other vegetation types (Ramankutty and Foley, 1999; Klein Goldewijk, 2001) (Table 2.8). The MODIS (Friedl et al., 2002) and Global Land Cover 2000 (Bartholome and Belward, 2005) provide other products. The two interpretations of the AVHRR data agree on the classification of vegetation as either tall (forest and woody savannah) or short (all other land cover) over 84% of the land surface (Hansen and Reed, 2000). However, some of the key disagreements are in regions subject to anthropogenic land cover change so may be important for the estimation of anthropogenic RF. Using the Hadley Centre Atmospheric Model (HadAM3) GCM, Betts et al. (2007) found that the estimate of RF relative to PNV varied from –0.2 W m–2 with the Wilson and Henderson-Sellers (1985) atlas-based land use data set to –0.24 W m–2 with a version of the Wilson and Henderson-Sellers (1985) data set adjusted to agree with the cropland data of Ramankutty and Foley (1999). Myhre and Myhre (2003) found the RF relative to PNV to vary from –0.66 W m–2 to 0.29 W m–2 according to whether the present-day land cover was from Wilson and Henderson-Sellers (1985), Ramankutty and Foley (1999) or other sources.
18.104.22.168.2 Uncertainties in the mapping and characterisation of the reference historical state
Reconstructions of historical land use states require information or assumptions regarding the nature and extent of land under human use and the nature of the PNV. Ramankutty and Foley (1999) reconstructed the fraction of land under crops at 0.5° resolution from 1700 to 1990 (Figure 2.15, Table 2.8) by combining the IGBP Global Land Cover Dataset with historical inventory data, assuming that all areas of past vegetation occur within areas of current vegetation. Klein Goldewijk (2001) reconstructed all land cover types from 1700 to 1990 (Figure 2.15, Table 2.8), combining cropland and pasture inventory data with historical population density maps and PNV. Klein Goldewijk used a Boolean approach, which meant that crops, for example, covered either 100% or 0% of a 0.5° grid box. The total global cropland of Klein Goldewijk is generally 25% less than that reconstructed by Ramankutty and Foley (1999) throughout 1700 to 1990. At local scales, the disagreement is greater due to the high spatial heterogeneity in both data sets. Large-scale PNV (Figure 2.15) is reconstructed either with models or by assuming that small-scale examples of currently undisturbed vegetation are representative of the PNV at the large scale. Matthews et al. (2004) simulated RF relative to 1700 as –0.20 W m–2 and –0.28 W m–2 with the above land use reconstructions.
22.214.171.124.3 Uncertainties in the parametrizations of the surface radiation processes
The albedo for a given land surface or vegetation type may either be prescribed or simulated on the basis of more fundamental characteristics such as vegetation leaf area. But either way, model parameters are set on the basis of observational data that may come from a number of conflicting sources. Both the AVHRR and MODIS (Schaaf et al., 2002; Gao et al., 2005) instruments have been used to quantify surface albedo for the IGBP vegetation classes in different regions and different seasons, and in some cases the albedo for a given vegetation type derived from one source can be twice that derived from the other (e.g., Strugnell et al., 2001; Myhre et al., 2005a). Myhre and Myhre (2003) examined the implications of varying the albedo of different vegetation types either together or separately, and found the RF relative to PNV to vary from –0.65 W m–2 to +0.47 W m–2; however, the positive RFs occurred in only a few cases and resulted from large reductions in surface albedo in semi-arid regions on conversion to pasture, so were considered unrealistic by the study’s authors. The single most important factor for the uncertainty in the study by Myhre and Myhre (2003) was found to be the surface albedo for cropland. In simulations where only the cropland surface albedo was varied between 0.15, 0.18 and 0.20, the resulting RFs relative to PNV were –0.06, –0.20 and –0.29 W m–2, respectively. Similar results were found by Matthews et al. (2003) considering only cropland changes and not pasture; with cropland surface albedos of 0.17 and 0.20, RFs relative to 1700 were –0.15 and –0.28 W m–2, respectively.
126.96.36.199.3 Uncertainties in other parts of the model
When climate models are used to estimate the RF, uncertainties in other parts of the model also affect the estimates. In particular, the simulation of snow cover affects the extent to which land cover changes affect surface albedo. Betts (2000) estimated that the systematic biases in snow cover in HadAM3 introduced errors of up to approximately 10% in the simulation of local RF due to conversion between forest and open land. Such uncertainties could be reduced by the use of an observational snow climatology in a model that just treats the radiative transfer (Myhre and Myhre, 2003). The simulation of cloud cover affects the extent to which the simulated surface albedo changes affect planetary albedo – too much cloud cover could diminish the contribution of surface albedo changes to the planetary albedo change.
On the basis of the studies assessed here, including a number of new estimates since the TAR, the assessment is that the best estimate of RF relative to 1750 due to land-use related surface albedo change should remain at –0.2 ± 0.2 W m–2. In the light of the additional modelling studies, the exclusion of feedbacks, the improved incorporation of large-scale observations and the explicit consideration of land use reconstructions for 1750, the level of scientific understanding is raised to medium-low, compared to low in the TAR (Section 2.9, Table 2.11).