Downscaling from Global Climate Change to Local Impacts

To assess climate change impacts, one usually starts from one or several socioeconomic scenarios and uses one or several GCMs to create ''climate scenarios'', i.e. low-resolution global simulations of climate change. Typically, GCMs will be able to provide time series for meteorological variables with a very short time sampling (up to hourly) and a medium to low resolution at the horizontal scale (so far, up to 100 km). Such a spatial resolution is sometimes good enough to carry out impact analyses, e.g. to assess the impact on agriculture or forestry.

Quite often, however, impacts are precisely localized, and a higher resolution is needed. It is the case when local characteristics play an important role and when weather conditions have a high variability. Such a situation occurs in mountain areas, where orography is essential, or in cities, where the urban heat island can play a significant role. In such situations, it is necessary to downscale GCM output, to get climate scenario at a resolution that is high enough to be used in impact models. Some sort of downscaling is also needed when the phenomena that one wants to consider are too small to be adequately reproduced by GCMs. Examples of such phenomena are heavy precipitations that have a spatial scale of the order of a few kilometres or tropical cyclones. Tropical cyclones cannot be reproduced by GCM because their intensity depends on small-scale processes around the eye that a 100km-resolution model cannot resolve.

There are two ways of downscaling: using statistical methods or physical models. The first method uses statistical relationships, calibrated on historical data, to relate large-scale drivers - which GCMs can reproduce - to local phenomena - which GCMs cannot reproduce. Even though our knowledge of the laws of physics helps selecting potential predictors, this method is not directly based on physical laws. Example of this method applied to hurricanes is provided by Elsner and Jagger (2006), who estimate the return level of extreme hurricane wind on the U.S. coastline, as a function of global climate indices like ENSO and NAO. Statistical methods have often a good prediction skill. Statistical models, however, have two main drawbacks: first, they need long series of reliable data; second, it is difficult to know the validity domain of statistical relationships. While physical laws will not change in the future (even though we may find out that they are not what we used to think they are), a statistical relationship can be different in a different climate. For instance, the correlation between sea surface temperature and hurricane intensity is very strong in the present climate (see, e.g., Emanuel, 2005), but it does not mean that if climate gets 2°C warmer, hurricane intensity will automatically increase: the effect of a local or temporary perturbation may be different from the effect of a global or permanent change.

To avoid this validity problem, one may use physical models, which often have a lower skill than statistical models (see an example on hurricanes in Emanuel et al., 2006), but which are based on physical laws that will not change in the future. These physical models, used to downscale GCM output, can be Regional Climate Models (RCM) that take as input a large-scale forcing produced by GCM (see examples of this approach in Christensen and Christensen, 2007), or specific models like hurricane models. An example of application of this method on hurricanes is provided by Emanuel (2006), who uses a hurricane model that takes as input large-scale conditions (vertical wind shear, potential intensity, etc.), and provides hurricane tracks and intensity. Of course, physical models often require calibration, so that the distinction between physical models and statistical models is sometimes fuzzy.

As an illustration of the methodology, the present paper summarizes the main findings of Hallegatte (2007a), which investigates the changes in hurricane risk due to a 10-percent increase in potential intensity. To do so, this work uses the synthetic hurricane tracks, which were produced using a physical model described in Emanuel (2006). Two sets of tracks are used. First, a set of 3000 tracks (with 1862 landfalls), which has been produced using large-scale drivers that correspond to the present climate. This set is referred to as Present Climate (PC). Second, another set of synthetic tracks, which has been produced assuming a 10 percent increase in potential intensity, compared with the present climate conditions. Such an increase in potential intensity, together with other environmental changes and with a large uncertainty, is expected at the end of this century (Emanuel, 2005).

This set also contains 3000 tracks, with 1912 landfalls, and is hereafter referred to as Modified Climate (MC). Comprehensive description and validation of these synthetic tracks are provided in Emanuel (2006).

These tracks are used to assess how the increase in potential intensity could modify the annual probability of hurricane landfall on the U.S. Atlantic and Gulf coastline. Figure 2 shows these probabilities as derived from the HURDAT database, from the Emanuel's model in the present climate, and from the Emanuel's model in a climate where potential intensity has increased by 10 percent. It can be seen that the model is able to reproduce fairly well the historical probability of landfall, except for the weakest hurricane of category one. This discrepancy arises probably from the track model that assumes that the storm is well structured, which might not be the case of weak storms. For the strongest hurricanes, of category 2-5, the model fits well with present-day data. For these intense hurricanes, the model predicts an increase in the annual probability of landfall, and the higher the category, the larger is the probability increase. For category 5 hurricanes, the annual probability of landfall even soars from 7 percent to 21 percent, suggesting that the risk of large-scale disaster could be significantly increased by climate change.

More useful for risk analysis, the synthetic tracks can provide, see Tables 1 and 2, the annual probability of landfall for the five categories of the Saffir-Simpson scale, in 11 regions of the U.S. Atlantic and Gulf coastline (see the region definition in Hallegatte, 2007a, or in the United States Landfalling Hurricane Project website, http://www.e-transit.org/hurricane/map.asp"). These tables highlight where hurricane risk could be enhanced. In particular, one can notice (i) the 10-fold increase in the category-5 landfall probability in region 3, which includes New Orleans; (ii) the possibility of category 4 landfalls in the regions 9, 10, and 11, that lies from Virginia State to the Canadian border and include Washington D.C., New York City, Boston, and several other major American cities.

Fig. 2 Annual probability of landfall of a hurricane of a given category, according to historical data (HURDAT), and synthetic tracks in the present (PC) and modified (MC) climate

Fig. 2 Annual probability of landfall of a hurricane of a given category, according to historical data (HURDAT), and synthetic tracks in the present (PC) and modified (MC) climate

Table 1 Annual probability of landfall of a hurricane of each category, in the 11 region, for the present climate

Region Category

Table 1 Annual probability of landfall of a hurricane of each category, in the 11 region, for the present climate

Region Category

1

2

3

4

5

1

8.9%

2.6%

1.0%

2.9%

0.0%

2

17.0%

3.2%

2.6%

3.2%

0.6%

3

11.6%

4.8%

3.6%

2.3%

0.3%

4

14.2%

5.5%

3.6%

1.0%

1.3%

5

22.1%

6.4%

7.3%

5.1%

3.6%

6

4.2%

1.9%

1.6%

0.3%

0.0%

7

6.1%

4.8%

3.9%

1.3%

0.6%

8

5.8%

1.3%

2.9%

2.3%

0.3%

9

7.0%

2.6%

0.6%

0.0%

0.0%

10

2.3%

0.6%

0.0%

0.0%

0.0%

11

1.3%

1.0%

0.3%

0.0%

0.0%

Table 2 Annual probability of landfall of a hurricane of each category, in the 11 regions, for the Modified Climate (MC)

Region Category

Table 2 Annual probability of landfall of a hurricane of each category, in the 11 regions, for the Modified Climate (MC)

Region Category

1

2

3

4

5

1

13.9%

2.9%

2.9%

3.9%

1.9%

2

22.8%

7.3%

5.8%

3.2%

3.6%

3

15.6%

5.8%

6.7%

2.9%

2.9%

4

20.5%

7.0%

6.7%

3.2%

2.3%

5

26.9%

9.8%

10.4%

6.4%

9.5%

6

6.4%

3.6%

1.6%

1.0%

0.0%

7

7.0%

3.9%

2.9%

5.5%

2.3%

8

6.7%

3.2%

3.6%

4.2%

0.3%

9

9.8%

5.1%

1.3%

0.3%

0.0%

10

2.9%

0.3%

0.0%

0.3%

0.0%

11

2.9%

0.6%

0.3%

0.3%

0.0%

These results are useful because they describe how hurricane risks could change, which is the most important information for risk managers. They do not tell anything, however, about the economic impact of such a change. To investigate this question, it is necessary to go from hurricane risks - expressed in terms of probability - to the direct losses - expressed in terms (i) of mean annual direct losses and (ii) of probability of exceeding a given level of losses. In particular, these two variables are of the foremost importance for the insurance industry that needs to assess insurance premiums and capital requirements.

Was this article helpful?

0 0
Survival Treasure

Survival Treasure

This is a collection of 3 guides all about survival. Within this collection you find the following titles: Outdoor Survival Skills, Survival Basics and The Wilderness Survival Guide.

Get My Free Ebook


Post a comment