Other reports in this collection
Changes in variability have not been regularly incorporated in climate scenarios because: (1) less faith has been placed in climate model simulations of changes in variability than of changes in mean climate; (2) techniques for changing variability are more complex than those for incorporating mean changes; and (3) there may have been a perception that changes in means are more important for impacts than changes in variability (Mearns, 1995). Techniques for incorporating changes in variability emerged in the early 1990s (Mearns et al., 1992; Wilks, 1992; Woo, 1992; Barrow and Semenov, 1995; Mearns, 1995).
Some relatively simple techniques have been used to incorporate changes in interannual variability alone into scenarios. Such techniques are adequate in cases where the impact models use monthly climate data for input. One approach is to calculate present day and future year-by-year anomalies relative to the modelled baseline period, and to apply these anomalies (at an annual, seasonal or monthly resolution) to the long-term mean observed baseline climate. This produces climate time-series having an interannual variability equivalent to that modelled for the present day and future, both superimposed on the observed baseline climate. The approach was followed in evaluating impacts of variability change on crop yields in Finland (Carter et al., 2000a), and in the formation of climate scenarios for the United States National Assessment, though in the latter case the observed variability was retained for the historical period.
Another approach is to calculate the change in modelled interannual variability between the baseline and future periods, and then to apply it as an inflator or deflator to the observed baseline interannual variability. In this way, modelled changes in interannual variability are carried forward into the climate scenario, but the observed baseline climate still provides the initial definition of variability. This approach was initially developed in Mearns et al. (1992) and has recently been experimented with by Arnell (1999). However, this approach can produce unrealistic features, such as negative precipitation or inaccurate autocorrelation structure of temperature, when applied to climate data on a daily time-scale (Mearns et al., 1996).
The major, most complete technique for producing scenarios with changes in interannual and daily variability involves manipulation of the parameters of stochastic weather generators (defined in Section 13.3.2). These are commonly based either on a Markov chain approach (e.g., Richardson, 1981) or a spell length approach (e.g., Racksko et al., 1991), and simulate changes in variability on daily to interannual time-scales (Wilks, 1992). More detailed information on weather generators is provided in Chapter 10, Section 10.6.2.
To bring about changes in variability, the parameters of the weather generator are manipulated in ways that alter the daily variance of the variable of concern (usually temperature or precipitation) (Katz, 1996). For precipitation, this usually involves changes in both the frequency and intensity of daily precipitation. By manipulating the parameters on a daily time-scale, changes in variability are also induced on the interannual time-scale (Wilks, 1992). Some weather generators operating at sub-daily time-scales have also been applied to climate scenario generation (e.g., Kilsby et al., 1998).
A number of crop model simulations have been performed to determine the sensitivity of crop yields to incremental changes in daily and interannual variability (Barrow and Semenov, 1995; Mearns, 1995; Mearns et al., 1996; Riha et al., 1996; Wang and Erda, 1996; Vinocur et al., 2000). In most of these studies, changes in variability resulted in significant changes in crop yield. For example, Wang and Erda (1996) combined systematic incremental changes in daily variance of temperature and precipitation with mean climate scenarios in their study of climate change and corn yields in China. They found that increases in the variance of temperature and precipitation combined, further decreased crop yields compared to the effect of the mean change scenarios alone taken from several GCMs.
Studies using the variance changes in addition to mean changes from climate models to form climate scenarios also emerged in the past decade (Kaiser et al., 1993; Bates et al. 1994). For example, Bates et al. (1994, 1996) adapted Wilks’ (1992) method and applied it to changes in daily variability from doubled CO2 runs of the Commonwealth Scientific and Industrial Research Organisation (CSIRO) climate model (CSIRO9). They then applied the changed time-series to a hydrological model. Combined changes in mean and variability are also evident in a broad suite of statistical downscaling methods (e.g., Katz and Parlange, 1996; Wilby et al., 1998). See also Chapter 10, Section 10.6.3, for further discussion of statistical downscaling and changes in variability.
In recent years, more robust and physically meaningful changes in climatic variability on daily to interannual time scales have been found in runs of GCMs and RCMs for some regions (e.g., Gregory and Mitchell, 1995; Mearns et al., 1995a,b; Whetton et al., 1998a; Mearns, 1999; Boer et al., 2000). For example, on both daily and interannual time-scales many models simulate temperature variability decreases in winter and increases in summer in northern mid-latitude land areas (see Chapter 9, Section 9.3). This result is likely to encourage the further application of model-derived variability changes in climate scenario construction.
The most useful studies, from the point of view of elucidating uncertainty in climate scenarios and impacts, are those that compare applying scenarios with only mean changes to those with mean and variability change. Semenov and Barrow (1997) and Mearns et al. (1997) used mean and variance changes from climate models, formed scenarios of climate change using weather generators and applied them to crop models. In both studies important differences in the impacts of climatic change on crop yields were calculated when including the effect of variance change, compared to only considering mean changes. They identified three key aspects of changed climate relevant to the role played by change in daily to interannual variability of climate: the marginality of the current climate for crop growth, the relative size of the mean and variance changes, and the timing of these changes.
It is difficult to generalise the importance of changes in variability to climate change impacts since significance of changes in variability is region, variable, and resource system specific. For example, based on results of equilibrium control and 2xCO2 experiments of DARLAM (a regional model developed in Australia) nested within the CSIRO climate model over New South Wales, Whetton et al. (1998a) emphasised that most of the change in temperature extremes they calculated resulted from changes in the mean, not through change in the daily variance. In contrast, Mearns (1999) found large changes (e.g., decreases in winter) in daily variance of temperature in control and 2xCO2 experiments with a regional climate model (RegCM2) over the Great Plains of the U.S. (Giorgi et al., 1998). These changes were sufficient to make a significant difference in the frequency of daily temperature extremes. Note, however, that these results are not contradictory since they concern two very different regions. More generalised statements may be made regarding the importance of change in the variability of precipitation from climate change experiments for determining changes in the frequency of droughts and floods (e.g., Gregory et al., 1997; Kothavala, 1999). As noted in Chapters 9 and 10, high intensity rainfall events are expected to increase in general, and precipitation variability would be expected to increase where mean precipitation increases.
Other types of variance changes, on an interannual time-scale, based on changes in major atmospheric circulation oscillations, such as ENSO and North Atlantic Oscillation (NAO), are difficult to incorporate into impact assessments. The importance of the variability of climate associated with ENSO phases for resources systems such as agriculture and water resources have been well demonstrated (e.g., Cane et al., 1994; Chiew et al., 1998; Hansen et al., 1998).
Where ENSO signals are strong, weather generators can be successfully conditioned on ENSO phases; and therein lies the potential for creating scenarios with changes in the frequency of ENSO events. By conditioning on the phases, either discretely (Wang and Connor, 1996) or continuously (Woolhiser et al., 1993), a model can be formed for incorporating changes in the frequency and persistence of such events, which would then induce changes in the daily (and interannual) variability of the local climate sites. Weather generators can also be successfully conditioned using NAO signals (e.g., Wilby, 1998). However, it must be noted that there remains much uncertainty in how events such as ENSO might change with climate change (Knutson, et al., 1997; Timmerman et al., 1999; Walsh et al., 1999; see also Chapter 9, Section 9.3.5, for further discussion on possible changes in ENSO events). While there is great potential for the use of conditioned stochastic models in creating scenarios of changed variability, to date, no such scenario has actually been applied to an impact model.
|
Other reports in this collection |
|