A baseline period is needed to define the observed climate with which climate change information is usually combined to create a climate scenario. When using climate model results for scenario construction, the baseline also serves as the reference period from which the modelled future change in climate is calculated.
The choice of baseline period has often been governed by availability of the required climate data. Examples of adopted baseline periods include 1931 to 1960 (Leemans and Solomon, 1993), 1951 to 1980 (Smith and Pitts, 1997), or 1961 to 1990 (Kittel et al., 1995; Hulme et al., 1999b).
There may be climatological reasons to favour earlier baseline periods over later ones (IPCC, 1994). For example, later periods such as 1961 to 1990 are likely to have larger anthropogenic trends embedded in the climate data, especially the effects of sulphate aerosols over regions such as Europe and eastern USA (Karl et al., 1996). In this regard, the "ideal" baseline period would be in the 19th century when anthropogenic effects on global climate were negligible. Most impact assessments, however, seek to determine the effect of climate change with respect to "the present", and therefore recent baseline periods such as 1961 to 1990 are usually favoured. A further attraction of using 1961 to 1990 is that observational climate data coverage and availability are generally better for this period compared to earlier ones.
Whatever baseline period is adopted, it is important to acknowledge that there are differences between climatological averages based on century-long data (e.g., Legates and Wilmott, 1990) and those based on sub-periods. Moreover, different 30-year periods have been shown to exhibit differences in regional annual mean baseline temperature and precipitation of up to ±0.5ºC and ±15% respectively (Hulme and New, 1997; Visser et al., 2000; see also Chapter 2).
The adequacy of observed baseline climate data sets can only be evaluated in the context of particular climate scenario construction methods, since different methods have differing demands for baseline climate data.
There are an increasing number of gridded global (e.g., Leemans and Cramer, 1991; New et al., 1999) and national (e.g., Kittel et al., 1995, 1997; Frei and Schär, 1998) climate data sets describing mean surface climate, although few describe inter-annual climate variability (see Kittel et al., 1997; Xie and Arkin, 1997; New et al., 2000). Differences between alternative gridded regional or global baseline climate data sets may be large, and these may induce non-trivial differences in climate change impacts that use climate scenarios incorporating different baseline climate data (e.g., Arnell, 1999). These differences may be as much a function of different interpolation methods and station densities as they are of errors in observations or the result of sampling different time periods (Hulme and New, 1997; New, 1999). A common problem that some methods endeavour to correct is systematic biases in station locations (e.g., towards low elevation sites). The adequacy of different techniques (e.g., Daly et al., 1994; Hutchinson, 1995; New et al., 1999) to interpolate station records under conditions of varying station density and/or different topography has not been systematically evaluated.
A growing number of climate scenarios require gridded daily baseline climatological data sets at continental or global scales yet, to date, the only observed data products that meet this criterion are experimental (e.g., Piper and Stewart, 1996; Widmann and Bretherton, 2000). For this and other reasons, attempts have been made to combine monthly observed climatologies with stochastic weather generators to allow "synthetic" daily observed baseline data to be generated for national (e.g., Carter et al., 1996a; Semenov and Brooks, 1999), continental (e.g., Voet et al., 1996; Kittel et al., 1997), or even global (e.g., Friend, 1998) scales. Weather generators are statistical models of observed sequences of weather variables, whose outputs resemble weather data at individual or multi-site locations (Wilks and Wilby, 1999). Access to long observed daily weather series for many parts of the world (e.g., oceans, polar regions and some developing countries) is a problem for climate scenario developers who wish to calibrate and use weather generators.
A number of statistical downscaling techniques (see Section 13.4 and Chapter 10, Section 10.6, for definition) used in scenario development employ Numerical Weather Prediction (NWP) reanalysis data products as a source of upper air climate data (Kalnay et al., 1996). These reanalysis data sets extend over periods up to 40 years and provide spatial and temporal resolution sometimes lacking in observed climate data sets. Relatively little detailed work has compared such reanalysis data with independent observed data sets (see Santer et al., 1999, and Widmann and Bretherton, 2000, for two exceptions), but it is known that certain reanalysis variables - such as precipitation and some other hydrological variables - are unreliable.
Other reports in this collection