Design Tropical Cyclone Wind Speed when Considering Climate Change
Publication: Journal of Structural Engineering
Volume 146, Issue 5
Abstract
This paper investigated the evolution of tropical cyclone (TC) wind threat from the past into the future and discussed its implications for the building code specifications for two vulnerable coastal cities (Hangzhou and Shanghai) in China. Large samples of synthetic TCs were generated from reanalysis data and climate-model projections over the years 1979–2098. The synthetic data were evaluated with terrain-corrected historical TC wind speed data. Extreme value analysis was applied to the generated data set to estimate the return period of extreme winds. Most climate models considered (five of six) projected that the return period of TC winds for Hangzhou and Shanghai would significantly decrease over the 21st century. Because the traditional return-period-targeted estimation of the design wind speed is no longer appropriate in the context of a nonstationary climate, an alternative method based on the lifetime exceedance probability (LEP) was proposed to estimate the design wind speed. Most climate models considered (five of six) yielded significantly larger design wind speeds than the traditional stationary method, attributable to both projected increase in TC activity in the future climate and the applied nonstationary LEP method. The code-recommended design wind speeds for Shanghai and Hangzhou are smaller than the climatic design wind speeds projected by some (three of six) climate models considered in this study.
Introduction
Tropical cyclones (TCs) are one of the most devastating natural disasters and are responsible for large numbers of fatalities and economic losses. Current annual global damage from hurricanes (i.e., strong TCs that occur in the Atlantic Ocean and northeastern Pacific Ocean) has exceeded US$20 billion, and the cost is predicted to be US$56 billion in the year 2100 (Mendelsohn et al. 2012). Given the booming population and economy in coastal cities, improving wind-resistant performance of buildings and structures in those TC-prone areas is critical to reduce TC-induced losses. Design wind loads acting on a building are dependent mainly on the design wind speed, which can be determined by the extreme wind speed quantile associated with a large mean return period. For example, the US building code ASCE 7-16 (ASCE 2017) defines the wind speed with a 700-year return period as the basic design wind speed for Risk Category II buildings. Many building codes (e.g., AIJ 2004 and Chinese Load Code 2012) define the wind speed with a 50-year return period as the basic design wind speed, which often is applied with an amplifying load factor [e.g., 1.3–1.4 in the Chinese load code GB 50009 2012 (Chinese Load Code 2012)]. It is vital to accurately estimate TC wind speed distribution, especially its tail describing the extremes, to provide reliable design wind speeds.
To obtain a reliable extreme value distribution of TC wind, one needs a large sample of TC wind speed data. Due to the lack of long-term observational data, various statistical models have been developed to simulate synthetic TC events over a long period for extreme value analysis. Batts et al. (1980), Georgiou et al. (1983), and Neumann (1991), among others, first used simulation techniques to estimate extreme hurricane wind speed and assess hurricane risk. The basic idea of these methods is similar: fitting historical data (i.e., best track data) to derive statistical distributions of key TC parameters such as maximum intensity, central pressure, radius of maximum wind, translation speed, and so forth, for a specific site (e.g., a city) of interest and then applying a Monte Carlo approach to sample from these distributions. The reliability of these simulation methods depends on the shape of the tail of the fitted distributions, which is sensitive to extreme events, but unfortunately there is little supporting data for these events, particularly at the site-specific scale. Vickery et al. (2000, 2009b) used historical track and climate data to develop a basin-scale full track model and described TC intensity along each track as a function of previous step values as well as position and environmental variables such as sea surface temperature. Although it partially overcomes the issue of data limitation of the site-specific methods, this simulation method still relies mainly on historical TC data, limiting its ability to account for potential effects of climate change on TC activity in the future.
Increasing evidence shows that the occurrence frequency of strong TCs (e.g., Categories 4 and 5) will increase due to climate change (Emanuel 1987, 2005; Webster et al. 2005; Knutson et al. 2010; Emanuel et al. 2013). Mei and Xie (2016) revealed that typhoons (i.e., strong TCs that occur in the northwestern Pacific Ocean) striking east and southeast Asia had intensified over the past 37 years, and the proportion of storms of Categories 4 and 5 more than doubled. It is critical to investigate how the design wind speed may change in the future. Mudd et al. (2014a, b) applied the method of Vickery et al. (2000, 2009b) to study how future sea surface temperature change will affect the design wind speed and found that the current design wind speed of studied regions might be exceeded under future climate scenarios. Ellingwood and Lee (2016) applied a similar method to account for the effect of sea surface temperature change on hurricane winds; more generally, they suggested that time-variant hurricane occurrence and intensity should be adopted in lifecycle reliability assessment of civil infrastructure facilities. Hurricane frequency and intensity, however, depend on more climate variables (e.g., atmospheric temperature, wind, humidity, and ocean heat content) in more complex ways (e.g., Emanuel et al. 2004; Lin et al. 2017; Lee et al. 2018).
To further investigate how design wind speeds may change in the future climate, we applied the statistical-deterministic model of Emanuel et al. (2008), which can generate basinwide synthetic TCs driven by comprehensive climate conditions involving the environmental wind and humidity, thermodynamic state of the atmosphere, and thermal stratification of the ocean. This TC model was integrated with hydrodynamic surge models into a climatological-hydrodynamic method (Lin et al. 2012; Lin and Emanuel 2016) to project future storm surge hazard probabilities. Here, we applied this TC model to investigate how design wind speeds may change in the future climate. We generated large numbers of synthetic TCs under past and current climate conditions based on reanalysis data as well as synthetic TCs under the projected future climates based on six climate models, and we applied a parametric model to estimate the wind field associated with each synthetic TC. Although the methodology is applicable everywhere, we focused on two study sites, Hangzhou and Shanghai in China, given their susceptibility to typhoons and rapid urban development. Historical observations at the two sites were used to evaluate the synthetic TC wind generation methodology.
Based on the generated TCs, we applied extreme value analysis to investigate the nonstationarity of the extreme TC winds from the past into the future for Hangzhou and Shanghai. Such a projection of the nonstationary TC winds can be applied to assessing future hurricane damage to buildings (e.g., Li et al. 2016). Because the traditional return period–targeted estimation may no longer be appropriate in the context of a nonstationary climate, we proposed an alternative method based on the lifetime exceedance probability (LEP) to give a reasonable estimation of the design wind speed. In addition to the TC wind, we also accounted for the (relatively small) contribution of the non-TC wind to the design wind speed. The estimated design wind speeds were compared with current building code specifications. This study provides a basis for considering climate change effect in building codes and standards to ensure the target safety and performance levels of buildings and infrastructure systems.
Methodology
Synthetic TC Generation
The statistical/deterministic hurricane model (Emanuel et al. 2006) used in this study generates synthetic TCs under given large-scale atmospheric and ocean environments, which may be estimated from observations or climate modeling. The main procedure of synthetic TC generation includes three steps. First, weak protostorms are seeded uniformly over the basin within the large-scale environment. Once initialized, the storms move in accordance with the large-scale environmental wind plus a beta drift correction. Along each storm track, the Coupled Hurricane Intensity Prediction System (CHIPS) (Emanuel 2004), a dynamic model, is used to estimate the storm intensity and radius of maximum wind, according to environmental conditions such as potential intensity, wind shear, humidity, and the thermal stratification of the ocean. These wind and environmental conditions are modeled statistically based on the reanalysis or climate model data set. Sufficient synthetic TCs in an ocean basin under a given climate are generated to obtain a desired number of TCs that affect a particular coastal area of interest. The annual occurrence frequency of the TCs for the coastal area is estimated as the product of the annual frequency for the basin and the portion of the TCs affecting the area.
In this study, we applied this simulation method to two TC-prone cities on the east coastline of China—Hangzhou and Shanghai (Fig. 1)—to investigate the potential influence of climate change on the design wind speed. We assumed the TC-threatened area for Hangzhou and Shanghai to be within a 350-km-radius circle centered at latitude 29.86° and longitude 121.56° (Fig. 1, simulation point and circle), and all storms passing within this circle with the maximum wind intensity (1-min wind speed at 10 m over sea) greater than (40 knots) were selected (from basin-scale simulations). This storm-selection criterion was chosen to balance computational accuracy and efficiency and to include all historical TC events that had any significant influence on the two cities. To establish a baseline, 5,180 TC events in the observed climate of 1979–2015 (140 events for each year) were simulated for the study area based on the National Centers for Environmental Prediction/National Center for Atmospheric Research (NCEP/NCAR) reanalysis data set (Kalnay et al. 1996; updated). As described in the next section, the estimated wind speeds from this synthetic data set were compared with observations to evaluate the simulation method. To study how TC threats will change from the past to the future for the study area, each of six climate models was applied to generate 2,700 TCs under the (control) climate condition of 1979–2005 (100 events for each year) and 9,300 TCs for the projected climate of 2006–2098 (100 events for each year) under the Intergovernmental Panel on Climate Change 5 (IPCC5) emission scenario RCP8.5. The six climate models, selected from the Coupled Model Intercomparison Project Phase 5 (CMIP5), were CCSM4 (NCAR), GFDL5 (NOAA), HADGEM5 (UK Met Office Hadley Centre), MPI5 (Max Planck Institute), MIROC5 [Center for Climate System Research/National Institute for Environmental Studies/Frontier Research Center for Global Change (CCSR/NIES/FRCGC), Japan], and MRI5 (Meteorological Research Institute, Japan).
ST.1943-541X.0002585/asset/aeb3c218-5cfe-484c-8394-7e8d3369ba6f/assets/images/large/figure1.jpg)
For each synthetic TC, we used the Holland (1980) wind field model to estimate the wind field. This model can capture the structure of a storm core relatively well and it is computationally more efficient (for large numbers of simulations, as in this study) than other boundary-layer wind models, such as the complete wind profile of Chavas et al. (2015). The model calculates the 1-min axisymmetric wind (associated with the storm) at the gradient level based on the storm characteristics including maximum wind speed and radius of maximum wind. Then we converted the gradient wind to the surface level (10 m above the sea surface) with a velocity-reduction factor of 0.85 (Georgiou et al. 1983) and an empirical expression of inflow angles (Bretschneider 1972). We added to the storm wind the surface environmental wind estimated as a fraction (0.55, rotated counter-clockwise by 20°) of the storm translation velocity, to account for the asymmetry of the wind field (Lin and Chavas 2012). Finally, we adjusted the 1-min wind over sea to a 10-min average over open terrain, to be consistent with the code specification (GB 50009-2012), similar to previous studies (Harper et al. 2010; Vickery et al. 2009a; Yeo et al. 2014).
Because potential biases exist in the climate model projections, we applied bias correction to the synthetic TC wind speed data generated from the climate models. This bias correction was accomplished by a quantile-quantile mapping transformation (Boé et al. 2007). For each climate model, the cumulative density function (CDF) of the wind speed series over 1979–2005 was matched with that from the reanalysis simulation to generate a quantile-dependent correction function. Assuming the model bias did not change over the simulation period (1979–2098) for the climate model, this correction function was used to bias correct the climate-model-projected wind speed data quantile by quantile. In addition, we bias-corrected the climate-model-projected TC annual frequency by multiplying it by a correction factor, which was the ratio of the reanalysis-estimated frequency to the climate-model-estimated frequency for the 1979–2005 climate. The bias correction approach applied here for TC wind projection was similar to that applied for TC storm surge projection by previous studies (e.g., Lin et al. 2016).
Model Evaluation with Historical Data
Observed wind speed data from meteorological stations can provide realistic information about local wind climate; therefore we used these data to evaluate the simulated TC wind speed data. Due to the partial absence of the historical wind speed data at the Shanghai meteorological station, observational data from a nearby meteorological station, i.e., Pinghu (Fig. 1), were used instead. Because historical data from meteorological stations always represent a mixed wind climate, including TC and non-TC winds, we used the China Meteorological Administration (CMA) and Shanghai Typhoon Institute (STI) CMA-STI Best Track Dataset to identify TCs passing through the same 350-km-radius simulation circle with a maximum intensity greater than 40 knots. (A test applying separate selection circles to the two cities result in similar results.) The TC wind speed data measured at a 10-m height in the Hangzhou and Pinghu meteorological stations (Fig. 1) over the period from 1979 to 2015 were extracted to estimate the extreme wind speeds with various MRIs. These data series included maximum 10-min mean wind speeds along with their corresponding wind directions and 3-s gust wind speeds. The maximum 10-min mean wind speed data were available for 1979–2015, whereas 3-s gust wind speed data were absent for most years of the 1990s.
The Chinese load code GB 50009-2012 specifies design wind speed based on the 10-min mean wind speed observed at a reference height of 10 m for open rural exposure. However, due to the rapid urbanization in China since the 1980s, the surface roughness near meteorological stations may have changed dramatically, so that the exposure category may have become very different from the initial open rural exposure. Therefore, it may be necessary to carry out terrain correction on the recorded wind speed data. The terrain correction can be accomplished by the approach proposed by Ashcroft (1994) and applied to Hangzhou by Huang et al. (2018). The method is based on an empirical relationship between the gust factor (ratio of the 3-s gust to the 10-min wind speeds) and the roughness length where = height of wind speed (i.e., 10 m), and empirical parameters and and 2.32, respectively (Ashcroft 1994).
(1)
Fig. 2 shows the time-varying roughness length () in the prevailing TC wind speed directions for the two meteorological stations estimated from the observed gust factor and Eq. (1). For the years missing the gust wind speed, the exposure correction was made by assuming a linear change of the terrain. The roughness length increased significantly from Exposure category B (i.e., open rural terrain) in the 1980s to Category C (i.e., urban terrain) in the 2010s for both sites (Fig. 2), which is attributable to the rapid urban development. Such a time-dependent variation of the roughness confirmed that it was necessary to adjust the original wind speed series to the standard exposure category before using it to estimate the design wind speed. We corrected the recorded wind speed data () to the standard open rural exposure for both meteorological stations, based on an empirical transfer function (Ashcroft 1994)where = terrain factor; and = terrain-corrected wind speed.
(2)
ST.1943-541X.0002585/asset/27c1afc0-f2a5-44eb-a7d8-d5f8b5b7e6f0/assets/images/large/figure2.jpg)
Estimation of Extreme TC Wind Speeds for Various Return Periods
In the classic extreme value studies (Simiu and Heckert 1996; Heckert et al. 1998; Yeo et al. 2014), TC wind speeds were assumed to be independently and identically distributed in a stationary condition. Therefore, stationary extreme value models such as the generalized extreme value (GEV) family were widely adopted by most code provisions to estimate the design wind speed from recorded wind speed data. Whereas the GEV distribution family expresses the limiting behavior of the maximum value of a set of observations, the generalized Pareto distribution (GPD) model describes the limiting behavior of individual extreme events, which allows the analyst to use all the data exceeding a large enough threshold , i.e., a peaks over threshold approach (e.g., Simiu and Heckert 1996). Let be the TC wind speed based on GPDwhere and = scale and shape parameters of GPD, respectively; and = crossing rate over threshold .
(3)
Let be the annual maximum TC wind. Assuming that the storm arrival for the site is a Poisson process (Jagger and Elsner 2006; Lin et al. 2012), the annual exceedance probability of the TC wind iswhere = annual storm frequency for site, obtained from TC generation model.
(4)
Then, based on Eqs. (3) and (4) and the fact that the mean return period is the reciprocal of the annual exceedance probability, the wind speed associated with a given return period () can be determined as
(5)
By using historical data, most current codes estimate design wind speed with different return periods based on a stationary extreme value model such as Eq. (5), with an assumption that the parameters of the extreme value distribution and the storm frequency are constant over time. However, these parameters may no longer be constant because of global climate change. Given the large sample of generated synthetic storms, we applied Eq. (3) to fit the synthetic wind data separately for four periods—the late twentieth century, and the early, middle, and late 21st century—and used Eq. (5) to investigate how extreme wind speeds (with various return periods) evolve over these periods (due to the change in both the distribution parameters and storm frequency). We further investigated yearly varying extreme wind speeds from the late twentieth to the late 21st century, using a 5-year moving window. Thus, the stationary assumption was applied for each analysis period (e.g., 1979–2005 for the late twentieth century and 2034–2038 for a 5-year window), but the obtained extreme wind speeds varied with the climate over time (i.e., the “quasi-stationary” assumption; Lin and Shullman 2017).
Estimation of Design Wind Speed Based on LEP
Traditionally, design criteria for environmental loads on structures are based on the concepts of return periods. In a stationary setting, designing for the load of a return period is equivalent to designing for a quantile of the distribution of the maximum load on the structure over the design lifetime. This equivalence exists because the probability of the lifetime maximum load exceeding a certain level (i.e., the lifetime exceedance probability) is a function only of the return period for the level and the design lifetime. For example, 50-year and 100-year mean return periods correspond to 86.7% and 63.4% exceedance probabilities in 100 years, respectively. The lifetime exceedance probability also is called risk of failure (Fernández and Salas 1999).
In a nonstationary setting, the LEP depends on the return period, which varies over the design lifetime. To account for the dynamic feature of the wind load, therefore, we propose to set the design criteria based on the LEP, rather than the time-varying return period. Given a target LEP () and design lifetime period (), the design TC wind speed () can be determined numerically fromwhere = probability of not being exceeded in th year by TC winds, obtained from yearly varying frequency and GPD of wind speed [Eq. (4)] with the quasi-stationary assumption. Eq. (6) contains contributions of TCs in each year and generally can account for the influence of climate change on TCs over time. Therefore, this LEP-based approach may be regard as a nonstationary method for estimating the design TC wind speed. Rootzén and Katz (2013) also proposed such a design level targeting the LEP, termed a design life level, as a rational criterion for hydrologic engineering design accounting for climate change.
(6)
Recent discussions of accounting for climate change in design pointed out that, when modifying designs based on stationary assumptions to account for future nonstationary environmental changes, the designer may want to conserve the desired risk of failure (e.g., Hunter 2012; Buchanan et al. 2016). We note that design to conserve the LEP is equivalent to design to conserve the mean number of exceedance events over the lifetime. We disagree with Hunter (2012) that the latter is more appropriate than the former. For a design lifetime of years, in a stationary climate, the LEP is and the mean number of exceedances over the lifetime is ; in a nonstationary (or quasi-stationary) climate, the LEP is and the mean number of exceedances over the lifetime is [Eqs. (4) and (6)]. Thus, conserving the LEP is the same as conserving the mean number of exceedances over the lifetime (and also the average of the annual mean number of exceedances), as long as both quantities are calculated strictly in the Poisson framework discussed previously. For example, designing for an exceedance probability of 5% (63%) over a lifetime of 100 years is equivalent to designing for on average about 0.05 (1) exceedances over 100 years, for both stationary and nonstationary cases.
In addition, we note that design to conserve the LEP is equivalent to design to conserve the product and thus also the geometric mean of the annual nonexceedance probabilities (i.e., CDF) over the lifetime [Eq. (6)]. The design level that conserves the LEP also approximately conserves the sum and thus the arithmetic mean (i.e., average) of the annual exceedance probability. This is because, for large [small ], the sum of the annual exceedance probability over the lifetime is approximately for a stationary climate and for a nonstationary climate [Eq. (4)], equivalent to the mean number of exceedances over the lifetime. In practice, because the target failure probability always is small, the design level that conserves the LEP always is similar to the design level that conserves the average annual exceedance probability. Buchanan et al. (2016) proposed designing to conserve the average annual exceedance probability, but approximated the average annual exceedance probability with the average annual mean number of exceedances, resulting in the same level designed to conserve the LEP. Thus, LEP-based design is consistent with the traditional stationary design in conserving the risk of failure measured by various probabilistic metrics.
Finally, the design wind speed often represents a mixed climate including both TC and non-TC wind conditions (Gomes and Vickery 1978), e.g., the design wind speeds for Hangzhou and Pinghu in the Chinese load code (GB 50009-2012). To incorporate the effect of non-TC winds (mainly extratropical cyclones such as monsoons for the study sites), we applied the yearly maximum 10-min non-TC wind data of 1979–2015 collected from the meteorological stations for the two study sites. After terrain correction, we fitted the data using the classic Gumbel’s Type I extreme value distribution. Assuming that this distribution is stationary over time [e.g., Roberts et al. (2017) and Lin et al. (2019) found that extratropical cyclones are not likely to change much with climate change], we combined it with the time-varying extreme value distribution of TC wind speeds to determine the design wind speed of a mixed climate. Specifically, given a target LEP () and design lifetime period (), the design wind speed () is determined numerically fromwhere = the annual nonexceedance probability of non-TC winds, and TC winds and non-TC winds are assumed to be independent of each other.
(7)
Results: Influence of Climate Change on Extreme TC Wind Speed
Comparison of Historical and Reanalysis TC Data
Simulated and observed (and terrain-corrected) TC data for different mean return periods along with their corresponding GPD fittings derived from maximum likelihood estimation are compared in Fig. 3. The reanalysis and observational data matched well for both cities for the short return periods covered by the historical data, indicating that the synthetic TC wind simulation method is accurate. Although observation and simulation fittings had good agreement for Hangzhou even for very long return periods, the simulation-based estimation was significantly higher than the observation-based estimation for Pinghu for long return periods. This difference likely reveals a significant underestimation of the extreme winds by the observation-based extreme value fitting for Pinghu {the simulated and observed data points agreed well [Fig. 3(b)]}. Given limited data, the observation-based extreme analysis may result in large fitting errors in the tail of the GDP model corresponding to high wind speeds. The synthetic method can generate a sufficiently large number of events and thus is more reliable for estimating the extremes. This comparison provides confidence in the use of the synthetic TC generation method to investigate the evolution of the extreme TC winds under future climate projections for the two sites.
ST.1943-541X.0002585/asset/99d4cbbb-efc2-4695-970f-b1995e3ea2dd/assets/images/large/figure3.jpg)
Nonstationary Characteristic of Extreme TC Wind Speed
Because TCs may vary with the changing climate, potential nonstationarity of the TC wind speed deserves investigation. We divided TC wind speed series from each climate model into four periods: 1979–2005 (late twentieth century), 2006–2036 (early 21st century), 2037–2067 (middle 21st century), and 2068–2098 (late 21st century). For each period, the TC wind speed as a function of mean return period was estimated for Hangzhou and Pinghu (Figs. 4 and 5, respectively). Most climate models considered project that the mean return period of TC winds for both cities will decrease over the 21st century, implying an increasing TC threat to these cities. Compared with the late twentieth century (i.e., 1979–2005), the TC wind speeds, especially those for long return periods, were projected to significantly increase over the 21st century in the GFDL5, HADGEM5, and MIROC5 models for both sites. MRI5 projected a significant increase in the extreme winds for Hangzhou and a modest increase for Pinghu. MPI5 projected a relatively small to modest increase for the two sites. CCSM4 was the only model projecting that the extreme winds may not change or even may slightly decrease over the 21st century. Uncertainties exist among the climate model projections, but most model results indicated that using historical data (e.g., 1979–2005) to determine the design TC wind speed may underestimate future TC activity and thus compromise building safety.
ST.1943-541X.0002585/asset/961bfc7c-6dbd-4e95-8d46-616a75b6f828/assets/images/large/figure4.jpg)
ST.1943-541X.0002585/asset/fbfbdda3-0a3a-4e77-9636-1a4a087f4394/assets/images/large/figure5.jpg)
We further investigated the yearly variation of extreme TC wind speed to better understand the climate-change-induced nonstationarity. To increase available yearly data for extreme value analysis and to reduce the effect of the climate variability, we used a 5-year moving window, i.e., TC data for the two years before and after each year were included in the analysis for the year. Abandoning the first and last two years, the whole time series was 1981–2096. Figs. 6 and 7 show the extreme TC wind speeds of the six climate models as a function of time and mean return period for Hangzhou and Pinghu, respectively. The TC wind speed projected by CCSM4 seemed to be stationary (i.e., no trend) over the years 1981–2096 for all return periods. For other models, larger extreme values appeared in the middle and late 21st century, showing a significant nonstationary feature. For Hangzhou, GFDL5, HADGEM5, and MPI5 found that intensive TC events may occur between 2055–2070, whereas the MIROC5 and MRI5 models suggested that powerful TCs are more likely to appear around 2095 and 2030, respectively. Similar nonstationary features appeared for Pinghu, which also may be subject to serious TC impact during the period 2050–2070 (shown by GFDL5, HADGEM5, MPI5, and MRI5) and toward the end of the century (shown by MIROC5).
ST.1943-541X.0002585/asset/2559ddd8-a19a-4f33-9fe7-e1725fe9933a/assets/images/large/figure6.jpg)
ST.1943-541X.0002585/asset/88fc6bb7-ab6f-469d-ae60-cfcdc6c8e69d/assets/images/large/figure7.jpg)
To substantiate the preceding observation, we applied the Mann–Kendall (MK) test (Kendall 1975) to statistically assess if there was a monotonic upward or downward trend existing in extreme TC wind speed series. Because extreme wind speeds with 50- and 100-year return periods are widely used as design wind speeds for buildings, and the trend was clearer for the more extreme winds (Figs. 4–7), we applied the MK test to the 50- and 100-year winds (Table 1). The MK test statistic was larger than (95% percentile of the standard normal distribution) for most climate models, indicating that both the 50- and 100-year wind speeds tended to increase from 1981 to 2096. This result again indicates that the design TC wind speed estimated based on historical climate may fail to fulfill the wind-resistant requirements in the future. Therefore, considering climate-change-induced nonstationarity of the TC wind speed is necessary when designing building and infrastructure systems that may be used over a long period.
| Climate model | 50-year return period | 100-year return period | ||||||
|---|---|---|---|---|---|---|---|---|
| Hangzhou | Pinghu | Hangzhou | Pinghu | |||||
| Trend | Trend | Trend | Trend | |||||
| CCSM4 | No | 0.07 | No | No | No | |||
| GFDL5 | 9.90 | Upward | 6.83 | Upward | 9.56 | Upward | 5.81 | Upward |
| HADGEM5 | 2.90 | Upward | 4.49 | Upward | 2.73 | Upward | 4.21 | Upward |
| MIROC5 | 9.87 | Upward | 9.83 | Upward | 9.62 | Upward | 8.92 | Upward |
| MPI5 | 2.13 | Upward | 2.09 | Upward | 1.50 | No | 1.82 | No |
| MRI5 | 2.88 | Upward | 3.17 | Upward | 2.09 | Upward | 2.21 | Upward |
Results: Design Wind Speed
According to the MK test results in Table 1, the GFDL5 and MIROC5 models showed the most significant increase trends in the design wind speed. Thus, the data of the 100-year period from 1997–2096 of these two models were used to study the potential difference between the stationary return-period-based approach [applying the GPD analysis to the entire data set and calculating the design winds for certain return periods by Eq. (5)] and nonstationary LEP method [assuming yearly variation and calculating the design winds for certain LEPs by Eq. (6)]. For comparison, the obtained design wind speeds are plotted as functions of the return period (to which the LEP corresponds in a stationary setting) for Hangzhou and Pinghu in Fig. 8. The design TC wind speeds estimated by the LEP method were significantly larger than the corresponding stationary results, particularly for Hangzhou, and the differences increased with increasing return period (i.e., decreasing LEP). For example, when considering the 1,000-year return period ( for the period of 100 years), the nonstationary design wind speed of GFDL5 and MIROC5 for Hangzhou (Pinghu) was 14% (6%) and 12% (7%) larger than the stationary ones, respectively. Thus, even based on the same physically nonstationary data set, the conventional stationary statistical modeling may underestimate the design wind speed compared with the nonstationary LEP method, especially for long return periods/small LEPs.
ST.1943-541X.0002585/asset/b423a1ac-d46f-42e5-b1d5-94548fd49a83/assets/images/large/figure8.jpg)
To study design TC wind speed for the two cities for the 21st century, we applied the nonstationary LEP method to analyze the generated data from 1997–2096 from each of the six climate models. For comparison, we applied the stationary return-period-based method to the generated data from the reanalysis and the historical data from the meteorological stations from 1979–2015, consistent with the classic code estimation. The nonstationary and stationary results are denoted climatic and normative design wind speeds, respectively. Tables 2 and 3 list the estimated design TC wind speeds for Hangzhou and Pinghu, respectively, for various LEPs (corresponding to large return periods from 50 to 1,000 years in the stationary setting). The normative design wind speeds obtained from the reanalysis were comparable with those obtained from historical analysis for Hangzhou, but they were higher for Pinghu, as expected, given the large underestimation of extreme wind speeds for Pinghu based on the limited historical data (Fig. 3). Climate models, except CCSM4, yielded larger design TC wind speed than the traditional stationary method, which is attributable to projected increases in TC activity in the future climate as well as the applied nonstationary LEP method. According to the magnitude of the design TC wind speed, the climate models considered may be divided into three groups. Taking Hangzhou as an example, the GFDL5 and MIROC5 models gave the largest increase in the design wind speeds compared with the stationary case, and were considered Group 1; CCSM4 gave similar results to those in the stationary case, and was considered Group 3; and the other three models gave relatively modest increases of the design wind speeds, and were considered Group 2. Similar classification could be made for Pinghu except for adding HADGEM5 to Group 1. These comparisons again indicate that it may be unconservative to use a stationary statistical model along with historical data to estimate the design wind speed for buildings and infrastructures, which likely will be subjected to increasing TC effects in the future.
| LEP in 100 years (%) | Normative design wind speed | Climatic design wind speed (1997–2096) () | |||||||
|---|---|---|---|---|---|---|---|---|---|
| (year) | Historical () | Reanalysis () | CCSM4 | GFDL5 | HADGEM5 | MIROC5 | MPI5 | MRI5 | |
| 86.7 | 50 | 22.4 | 22.2 | 21.4 | 28.6 | 23.7 | 28.2 | 23.4 | 24.3 |
| 63.4 | 100 | 24.6 | 24.3 | 23.0 | 32.3 | 26.2 | 31.9 | 25.3 | 27.1 |
| 18.1 | 500 | 29.1 | 28.9 | 27.2 | 42.4 | 33.0 | 42.9 | 31.0 | 35.1 |
| 9.52 | 1,000 | 30.9 | 30.7 | 29.2 | 47.4 | 36.5 | 49.0 | 34.4 | 39.4 |
| LEP in 100 years (%) | Normative design wind speed | Climatic design wind speed (1997–2096) () | |||||||
|---|---|---|---|---|---|---|---|---|---|
| (year) | Historical () | Reanalysis () | CCSM4 | GFDL5 | HADGEM5 | MIROC5 | MPI5 | MRI5 | |
| 86.7 | 50 | 21.2 | 23.6 | 22.2 | 26.8 | 27.0 | 29.9 | 24.3 | 25.4 |
| 63.4 | 100 | 22.1 | 25.8 | 24.1 | 29.0 | 30.5 | 33.4 | 26.6 | 27.7 |
| 18.1 | 500 | 23.6 | 30.5 | 28.7 | 34.4 | 41.1 | 43.4 | 32.8 | 33.9 |
| 9.52 | 1,000 | 24.0 | 32.4 | 30.7 | 37.0 | 46.9 | 49.0 | 36.0 | 36.5 |
The Chinese load code (GB 50009-2012) suggests that design wind speeds associated with 50- and 100-year return periods for Hangzhou and Pinghu are 26.8 and , respectively. These design wind speeds represent a mixed climate including both TC and non-TC wind conditions. To derive the mixed-climate design wind speed [Eq. (7)], we incorporated the effect of non-TC wind based on the observational data of 1979–2015 collected from the meteorological stations. Table 4 lists the mixed-climate design wind speeds with 86.7% and 63.4% LEPs in 100 years (corresponding to 50- and 100-year return periods in a stationary climate) for the six climate models. The contribution of the non-TC winds generally was very small for the 86.7% and 63.4% LEPs (Tables 2–4) and was negligible for smaller LEPs (not shown), implying that TCs are the main factors controlling wind-resistant design of buildings and infrastructures in these two coastal cities. The code-recommended design wind speeds (which may be derived based on different historical data using different stationary distributions) are higher than the normative design wind speeds obtained in this study, but they are smaller than the climatic design wind speeds projected by the Group 1 climate models. Thus, current code specifications for the two cities may not be conservative.
| Location | LEP in 100 years (%) | Normative design wind speed () | Climatic design wind speed (1997–2096) () | ||||||
|---|---|---|---|---|---|---|---|---|---|
| Historical | Reanalysis | CCSM4 | GFDL5 | HADGEM5 | MIROC5 | MPI5 | MRI5 | ||
| Hangzhou | 86.7 | 22.8 | 22.7 | 22.0 | 28.7 | 23.9 | 28.3 | 23.6 | 24.5 |
| 63.4 | 24.8 | 24.6 | 23.5 | 32.3 | 26.3 | 31.9 | 25.5 | 27.2 | |
| Pinghu | 86.7 | 22.1 | 24.4 | 23.2 | 27.0 | 27.4 | 30.1 | 24.9 | 25.8 |
| 63.4 | 23.1 | 26.4 | 25.1 | 29.2 | 30.7 | 33.5 | 27.1 | 28 | |
Conclusions
This paper investigated the nonstationarity of the extreme TC winds under the effects of climate change and discussed its implication in building code specifications. A large number of synthetic TCs were generated using a statistical-deterministic method for the study areas of Shanghai and Hangzhou, under the reanalysis and climate-model-projected climates for the years 1979–2098. Based on the generated data set, extreme value analysis was applied to investigate the nonstationarity of the extreme TC winds from the past to the future for Hangzhou and Shanghai. Because the traditional return-period-targeted estimation is no longer appropriate in the context of a nonstationary climate, an alternative method based on the lifetime exceedance probability was proposed to give a reasonable estimation of the design wind speed. We argue that the LEP-based design is consistent with the traditional stationary design in conserving the risk of failure measured by various probabilistic metrics.
The synthetic TC winds generated from the reanalysis data agreed with (terrain-corrected) historical wind observations for return periods reliably covered by the historical data, indicating that the synthetic TC wind simulation method is accurate. Given limited data, the observation-based analysis may result in large fitting errors in the tail of the distribution corresponding to high wind speeds and thus underestimate the design wind. Most climate models considered (five of six) projected that the mean return periods of TC winds for Hangzhou and Shanghai will significantly decrease over the 21st century, implying an increasing TC threat to these two cities.
Conventional stationary statistical modeling may underestimate the design wind speed compared with the nonstationary LEP method. Most climate models considered (five of six) yielded significantly (e.g., 6%–28% for Hangzhou for the 50-year return period) larger design TC wind speeds than the traditional stationary method attributable to projected increases in TC activity in the future climate as well as the applied nonstationary LEP method. Thus, it may be unconservative to use a stationary statistical model with historical data to estimate the design wind speed for buildings and infrastructures, which are likely to experience increasing TC effects over their service lives.
The contribution of the non-TC winds generally is very small, implying that TCs are the main factors controlling wind-resistant design of buildings and infrastructures in Shanghai and Hangzhou. The code-recommended design wind speeds for these two cities are higher than the normative design wind speeds obtained in this study, but they are smaller than the climatic design wind speeds projected by some (three of six) climate models considered in this study. Both physical and statistical nonstationary features of extreme winds may be accounted for in building code and standards based on methods proposed in this study to ensure the target safety and performance levels of buildings and infrastructure systems.
Acknowledgments
The work described in this paper was supported by the US National Science Foundation (CMMI-1652448, EAR-1520683), the National Natural Science Foundation of China (Project Nos. 51508502, 51978614, and 51838012), and the China Postdoctoral Science Foundation (Project No. 2015M581938). We thank Kerry Emanuel of MIT for generating the synthetic TC data sets and providing helpful comments.
References
AIJ (Architectural Institute of Japan). 2004. AIJ recommendations for loads on buildings. Tokyo: AIJ.
ASCE. 2017. Minimum design loads for buildings and other structures. ASCE 7-16. Reston, VA: ASCE.
Ashcroft, J. 1994. “The relationship between the gust ratio, terrain roughness, gust duration and the hourly mean wind speed.” J. Wind Eng. Ind. Aerodyn. 53 (3): 331–355. https://doi.org/10.1016/0167-6105(94)90090-6.
Batts, M. E., M. R. Cordes, and E. Simiu. 1980. “Sampling errors in estimation of extreme hurricane winds.” J. Struct. Div. 106 (10): 2109–2115.
Boé, J., L. Terray, F. Habets, and E. Martin. 2007. “Statistical and dynamical downscaling of the Seine basin climate for hydro-meteorological studies” Int. J. Climatol. 27 (12): 1643–1655. https://doi.org/10.1002/joc.1602.
Bretschneider, C. 1972. “A non-dimensional stationary hurricane wave model.” In Proc., Offshore Technology Conf. Houston: Offshore Technology Conference.
Buchanan, M. K., R. E. Kopp, M. Oppenheimer, and C. Tebaldi. 2016. “Allowances for evolving coastal flood risk under uncertain local sea-level rise.” Clim. Change 137 (3–4): 347–362. https://doi.org/10.1007/s10584-016-1664-7.
Chavas, D., L. Lin, and K. Emanuel. 2015. “A model for the complete radial structure of the tropical cyclone wind field. Part I: Comparison with observed structure.” J. Atmos. Sci. 72 (9): 3647–3662. https://doi.org/10.1175/JAS-D-15-0014.1.
Chinese Load Code. 2012. Load code for the design of building structures. GB 50009. Beijing: China Academy of Building Research.
Ellingwood, B. R., and J. Y. Lee. 2016. “Life cycle performance goals for civil infrastructure: Intergenerational risk-informed decisions.” Struct. Infrastruct. Eng. 12 (7): 822–829. https://doi.org/10.1080/15732479.2015.1064966.
Emanuel, K. A. 1987. “The dependence of hurricane intensity on climate.” Nature 326 (6112): 483–485. https://doi.org/10.1038/326483a0.
Emanuel, K. A. 2004. Tropical cyclone energetics and structure, 165–192. Cambridge, UK: Cambridge University Press.
Emanuel, K. A. 2005. “Increasing destructiveness of tropical cyclones over the past 30 years.” Nature 436 (7051): 686–688. https://doi.org/10.1038/nature03906.
Emanuel, K. A. 2013. “Downscaling CMIP5 climate models shows increased tropical cyclone activity over the 21st century.” Proc. Natl. Acad. Sci. 110 (30): 12219–12224. https://doi.org/10.1073/pnas.1301293110.
Emanuel, K. A., C. DesAutels, C. Holloway, and R. Korty. 2004. “Environmental control of tropical cyclone intensity.” J. Atmos. Sci. 61 (7): 843–858. https://doi.org/10.1175/1520-0469(2004)061%3C0843:ECOTCI%3E2.0.CO;2.
Emanuel, K. A., S. Ravela, E. Vivant, and C. Risi. 2006. “A statistical deterministic approach to hurricane risk assessment.” Bull. Am. Meteorol. Soc. 87 (3): 299–314. https://doi.org/10.1175/BAMS-87-3-299.
Emanuel, K., R. Sundararajan, and J. Williams. 2008. “Hurricanes and global warming: Results from downscaling IPCC AR4 simulations.” Bull. Am. Meteorol. Soc. 89 (3): 347–367.
Fernández, B., and J. D. Salas. 1999. “Return period and risk of hydrologic events. I: Mathematical formulation.” J. Hydrol. Eng. 4 (4): 297–307. https://doi.org/10.1061/(ASCE)1084-0699(1999)4:4(297).
Georgiou, P. N., A. G. Davenport, and P. J. Vickery. 1983. “Design wind speeds in regions dominated by tropical cyclones.” J. Wind Eng. Ind. Aerodyn. 13 (1–3): 139–152. https://doi.org/10.1016/0167-6105(83)90136-8.
Gomes, L., and B. J. Vickery. 1978. “Extreme wind speeds in mixed wind climates.” J. Wind Eng. Ind. Aerodyn. 2 (4): 331–344. https://doi.org/10.1016/0167-6105(78)90018-1.
Harper, B. A., J. D. Kepert, and J. D. Ginger. 2010. Guidelines for converting between various wind averaging periods in tropical cyclone conditions. Geneva: World Meteorological Organization.
Heckert, N. A., E. Simiu, and T. Whalen. 1998. “Estimates of hurricane wind speeds by peaks over threshold method.” J. Struct. Eng. 124 (4): 445–449. https://doi.org/10.1061/(ASCE)0733-9445(1998)124:4(445).
Holland, G. 1980. “Analytic model of the wind and pressure profiles in hurricanes.” Mon. Weather Rev. 108 (8): 1212–1218. https://doi.org/10.1175/1520-0493(1980)108%3C1212:AAMOTW%3E2.0.CO;2.
Huang, M., Q. Li, H. Xu, W. Lou, and N. Lin. 2018. “Non-stationary statistical modeling of extreme wind speed series with exposure correction.” Wind Struct. 26 (3): 129–146. https://doi.org/10.12989/was.2018.26.3.129.
Hunter, J. 2012. “A simple technique for estimating an allowance for uncertain sea-level rise.” Clim. Change 113 (2): 239–252. https://doi.org/10.1007/s10584-011-0332-1.
Jagger, T. H., and J. B. Elsner. 2006. “Climatology models for extreme hurricane winds near the United States.” J. Clim. 19 (13): 3220–3236. https://doi.org/10.1175/JCLI3913.1.
Kalnay, E., et al. 1996. “The NCEP/NCAR 40-year reanalysis project.” Bull. Am. Meteorol. Soc. 77 (3): 437–471. https://doi.org/10.1175/1520-0477(1996)077%3C0437:TNYRP%3E2.0.CO;2.
Kendall, M. G. 1975. Rank correlation methods. 4th ed. London: Charles Griffin.
Knutson, T. R., J. L. McBride, J. Chan, K. Emanuel, G. Holland, C. Landsea, I. Held, J. P. Kossin, A. K. Srivastava, and M. Sugi. 2010. “Tropical cyclones and climate change.” Nat. Geosci. 3 (3): 157–163. https://doi.org/10.1038/ngeo779.
Lee, C.-Y., M. K. Tippett, A. H. Sobel, and S. J. Camargo. 2018. “An environmentally forced tropical cyclone hazard model.” J. Adv. Model. Earth Syst. 10 (1): 223–241. https://doi.org/10.1002/2017MS001186.
Li, Q. W., C. Wang, and H. Zhang. 2016. “A probabilistic framework for hurricane damage assessment considering non-stationarity and correlation in hurricane actions.” Struct. Saf. 59 (Mar): 108–117. https://doi.org/10.1016/j.strusafe.2016.01.001.
Lin, N., and D. Chavas. 2012. “On hurricane parametric wind and applications in storm surge modeling.” J. Geophys. Res. 117 (D9): D09120. https://doi.org/10.1029/2011JD017126.
Lin, N., and K. A. Emanuel. 2016. “Grey swan tropical cyclones.” Nat. Clim. Change 6: 106–111. https://doi.org/10.1038/nclimate2777.
Lin, N., K. Emanuel, M. Oppenheimer, and E. Vanmarcke. 2012. “Physically based assessment of hurricane surge threat under climate change.” Nat. Clim. Change 2 (6): 462–467. https://doi.org/10.1038/nclimate1389.
Lin, N., R. Jing, Y. Wang, E. Yonekura, J. Fan, and L. Xue. 2017. “A statistical investigation of the dependence of tropical cyclone intensity change on the surrounding environment.” Mon. Weather Rev. 145 (7): 2813–2831. https://doi.org/10.1175/MWR-D-16-0368.1.
Lin, N., R. E. Kopp, B. P. Horton, and J. P. Donnelly. 2016. “Hurricane Sandy’s flood frequency increasing from year 1800 to 2100.” Proc. Natl. Acad. Sci. USA 113 (43): 12071–12075. https://doi.org/10.1073/pnas.1604386113.
Lin, N., R. Marsooli, and B. Colle. 2019. “Storm surge return levels induced by the mid-to-late-twenty-first-century extratropical cyclones in the Northeastern United States.” Clim. Change. https://doi.org/10.1007/s10584-019-02431-8.
Lin, N., and E. Shullman. 2017. “Dealing with hurricane surge flooding in a changing environment: Part I. Risk assessment considering storm climatology change, sea level rise, and coastal development.” Stochastic Environ. Res. Risk Assess. 31 (9): 2379–2400. https://doi.org/10.1007/s00477-016-1377-5.
Mei, W., and S. P. Xie. 2016. “Intensification of landfalling typhoons over the northwest Pacific since the late 1970s.” Nat. Geosci. 9 (10): 753–756. https://doi.org/10.1038/ngeo2792.
Mendelsohn, R., K. A. Emanuel, S. Chonabayashi, and L. Bakkensen. 2012. “The impact of climate change on global tropical cyclone damage.” Nat. Clim. Change 2 (3): 205–209. https://doi.org/10.1038/nclimate1357.
Mudd, L., Y. Wang, C. Letchford, and D. Rosowsky. 2014a. “Assessing climate change impact on the U.S. East Coast Hurricane hazard: Temperature, frequency, and track.” Nat. Hazards Rev. 15 (3): 1–13. https://doi.org/10.1061/(ASCE)NH.1527-6996.0000128.
Mudd, L., Y. Wang, C. Letchford, and D. Rosowsky. 2014b. “Hurricane wind hazard assessment for a rapidly warming climate scenario.” J. Wind Eng. 133 (Oct): 242–249. https://doi.org/10.1016/j.jweia.2014.07.005.
Neumann, C. J. 1991. The national hurricane center risk analysis program (HURISK). Washington, DC: US Dept. of Commerce.
Roberts, K. J., B. A. Colle, and N. Korfe. 2017. “Impact of simulated 21st century changes in extratropical cyclones on coastal flooding at the battery, New York City.” J. Appl. Meteorol. Climatol. 56 (2): 415–432. https://doi.org/10.1175/JAMC-D-16-0088.1.
Rootzén, H., and R. W. Katz. 2013. “Design life level: Quantifying risk in a changing climate.” Water Resour. Res. 49 (9): 5964–5972. https://doi.org/10.1002/wrcr.20425.
Simiu, E., and N. A. Heckert. 1996. “Extreme wind distribution tails: A peak over threshold approach.” J. Struct. Eng. 122 (5): 539–547. https://doi.org/10.1061/(ASCE)0733-9445(1996)122:5(539).
Vickery, P. J., P. F. Skerjl, and L. A. Twisdale. 2000. “Simulation of hurricane risk in the U.S. using empirical track model.” J. Struct. Eng. 126 (10): 1222–1237. https://doi.org/10.1061/(ASCE)0733-9445(2000)126:10(1222).
Vickery, P. J., D. Wadhera, M. D. Powell, and Y. Chen. 2009a. “A hurricane boundary layer and wind field model for use in engineering applications.” J. Appl. Meteorol. Climatol. 48 (2): 381–405. https://doi.org/10.1175/2008JAMC1841.1.
Vickery, P. J., D. Wadhera, L. A. Twisdale Jr., and F. M. Lavelle. 2009b. “U.S. hurricane wind speed risk and uncertainty.” J. Struct. Eng. 135 (3): 301–320. https://doi.org/10.1061/(ASCE)0733-9445(2009)135:3(301).
Webster, P. J., G. J. Holland, J. A. Curry, and H. R. Chang. 2005. “Changes in tropical cyclone number, duration, and intensity in a warming environment.” Science 309 (5742): 1844–1846. https://doi.org/10.1126/science.1116448.
Yeo, D. H., N. Lin, and E. Simiu. 2014. “Estimation of hurricane wind speed probabilities: Application to New York City and other coastal locations.” J. Struct. Eng. 140 (6): 1–5. https://doi.org/10.1061/(ASCE)ST.1943-541X.0000892.
Information & Authors
Information
Published In
Journal of Structural Engineering
Volume 146 • Issue 5 • May 2020
Copyright
This work is made available under the terms of the Creative Commons Attribution 4.0 International license, http://creativecommons.org/licenses/by/4.0/.
History
Received: Jul 10, 2018
Accepted: Sep 24, 2019
Published online: Feb 29, 2020
Published in print: May 1, 2020
Discussion open until: Jul 29, 2020
Authors
Metrics & Citations
Metrics
Citations
Download citation
If you have the appropriate software installed, you can download article citation data to the citation manager of your choice. Simply select your manager software from the list below and click Download.