the same. References should be made to the frozen grid analyses work done in the Climate Research Unit (CRU) in the mid-1980s to the mid-1990s.
3. In general, the chapter would benefit from a more careful dissection of the global mean and a recognition that radiosondes are not near global. It does not address global mapping or the need for evaluations at co-located sites of sondes (see Hurrell et al., 2000; Agudelo and Curry, 2004; and Free and Seidel, 2005). There are no latitude-time series presented. This chapter does identify differences over high latitude land as being the main reason for surface being larger than troposphere trends in the extratropics (page 30), but this is not carried forward to the Executive Summary or Chapter 5. Weakening or removal of inversions over cold land or ice is a very good reason why the surface should warm more and a good example of why the global mean should be dissected.
4. The chapter also places too much emphasis on linear trends. Only linear trends as a function of latitude are presented, however this presentation can hide many things. The claimed agreement between radiosondes is not shown except for the linear trends (e.g., Table 3.6.1) (see Free and Seidel, 2005). There is nothing on root mean square differences, which are very revealing (Hurrell et al., 2000), or on monthly differences (smoothing the time series can be misleading).
5. In a number of places, the assertion is made that the troposphere has warmed more than the surface. However, the differences in trends are often quite small, particularly for the 1958-2004 period. It is not clear that these differences are statistically significant. Although statistical significance is assessed for the trends themselves, no analysis of the significance of trend differences is presented. When comparing trend differences between two estimates of the same quantity (e.g., tropospheric temperatures from radiosondes and satellites), it is more appropriate to examine the trend of the difference time series, rather than trends for each time series individually (because the data contain similar overall variability). This is an omission that should be corrected, and the text should reflect the results of such an analysis. In particular, any statement about differences in warming should be weakened considerably if the differences are not statistically significant.
6. The trends calculated from reanalyses are downplayed because the input data sets are not homogenized. Although there is potential independent value of reanalysis products, it is not clear what trends from a reanalysis model mean in the context of temporally varying inputs. Therefore, the committee agrees with the decision of the authors to downplay this source of information.
7. The issue of regional land-use and land-cover changes is brought up in a number of places, but the implications are not clearly addressed. For example, in lines 94-96 it is suggested that regional land-use change must be considered in the development of land-based data sets. However, if regional changes are large enough to have a measurable influence on global temperature, then these changes will be sampled and detected by the existing land-based networks. As such, why is this an issue when analyzing the differences among the data sets? There is an issue related to land-use and land-cover changes that could be addressed here or in other chapters. In the modeling discussions in Chapters 1, 5, and 6, land-use and land-cover is considered to be a forcing (with uncertain magnitude in the past) that is incorporated in some models and not in others. The committee believes this is correct and that land-use and land-cover should be considered as a forcing. Any land-use and land-cover effects in observational datasets should