Climate risks and adaptation in Asian coastal megacities: A synthesis

Box 2.1 ■ Strengths and Limitations of Different Downscaling TechniquesSelected for this StudyThe city studies utilized two different downscaling techniques—pattern scaling which is a kind of statistical downscaling, and dynamic downscalingtechniques. The first was undertaken for all the studies. The HCMC study also used dynamic downscaling. Briefly, statistical downscaling techniques are“based on the construction of relationships between large scale and local variables calibrated from historical data” (Jones et al. 2004). These statisticalrelationships are then applied to large-scale climate variables from AOGCM simulations to estimate corresponding local/regional characteristics (liketemperature and precipitation). Pattern scaling, one of the techniques used in this work, is a form of statistical downscaling that requires minimalregional meteorological data and is computationally not very complicated. In pattern scaling the change in the local variable of interest—liketemperature or precipitation—is represented in terms of the GCM’s change for the variable per unit change in global temperature (this is the scalingfactor). This factor is then multiplied by the global change in temperature associated with a specific IPCC scenario (A1FI and B1). In this specific case,the scaling factor was obtained by performing a linear regression on a number of GCM models available (see Sugiyama 2008 for details). The mainadvantage of statistical downscaling is that it is computationally not very expensive and can provide information at point locations. However, amongseveral limitations, one issue is that the statistical relationships may not remain the same in a future climate world and the method does not provideinformation on temporal and location linkages.In contrast, dynamical modeling techniques use physical models of the climate system allowing direct modeling of the dynamics of the physicalsystems that influence the climate of a region. They are rapidly becoming the most widely applied downscaling technique. New systems like PRECIS—usedby the HCMC study—can be utilized from a PC platform, but they can still generate a broad range of daily, site-specific (i.e., 25 kms resolution grids)hydrometeorological data for time periods spanning a century. This allows examination of time slices like 2030 to 2070 to establish frequency relationsfor events in 2050 that can include both floods and droughts. Boundary conditions need to be derived from coupled GCMs. Nevertheless, RCMs also havea number of potential pitfalls that need to be understood. They require large amounts of boundary data linked with the parent GCM, acquire the errors ofthe parent GCM, and (as is the case with PRECIS) may take several months to run. An additional limitation is their potential exclusive reliance on only oneAOGCM. Before applying the RCM, the user needs to understand whether it can model the types of events (e.g., typhoons) that are of particular interestin the study. Typically, for studies that involve downscaling analysis such as disaster risk assessments, when trying to incorporate climate change riskfactors into the analysis, statistical downscaling or regional climate models are likely to be the methods of choice.Source: Authors’ compilation.pact parameters (e.g., precipitation and temperatureincrease) that are generated are applied in estimatingflood impacts at the city level. Trying to quantifythese uncertainties—and given a confidence intervalfor the outcomes—is extremely difficult, and someuncertainties—like future emission scenarios—cannotbe assessed. However, despite these uncertainties,the IPCC AR4 recognizes that the AOGCMclimate change models do allow global forecasts,and that increasingly the downscaling techniquesare providing information on the likely scale of differentclimate impacts at local levels (Giorgi 2008).Estimates for changes in temperature andprecipitation in 2050 based on statisticaldownscaling approachThe statistical downscaling technique used in thestudy (see Sugiyama 2008) 16 provided estimates of(a) the expected temperature increases, (b) extreme24-hour precipitation increase factors, and (c) seasonalmean precipitation increases for 2050 for thetwo climate change scenarios A1FI and B1. Thesefactors were derived from16 AOGCMs models forall four cities and are shown in Table 2.1.The robustness of the relationships was examinedstatistically by the IR3S team by examiningthe scatter among the data points generated by thedownscaling of the 16 AOGCMs. They found robustrelations for global and local temperature andprecipitable water (used as a proxy for extreme precipitation)relationships, and viable but less strongrelationships for the temperature/mean seasonalprecipitation increases. How this was handled isexplained in Box 2.2.16This part of the analysis was carried out by IntegratedResearch System for Sustainability Science (IR3S) at theUniversity of Tokyo. The results are presented in Sugiyama(2008).Methodologies for Downscaling, Hydrological Mapping, and Assessing Damage Costs | 7