Description |
-
Abstract: Extreme winds cause significant damage to infrastructure in the United States. Climate change effects to extreme winds including increasing trends have been predicted in climate change scenarios; however, little observational evidence exists to support the hypothesis of increasing winds in U.S. coastal communities due to climate change. In this study, we use the historical record of peak... read more3-s gust winds at sites along the Eastern Seaboard to determine if nonstationarity exists in the historic wind record. We evaluate nonstationarity at individual stations and within regional "superstation" clusters. In order to evaluate nonstationarity, both parametric (Student's t-test) and non-parametric (Mann-Kendall) trend tests are used. As Lombardo and Ayyub (2014) separated winds by storm types, we observe evidence of nonstationarity by storm types (commingled, nonthunderstorms, thunderstorms and tropical storms). For commingled data, 23 out of 108 stations exhibit evidence of nonstationarity. Roughly 16% of these stations show a positive trend from Florida to NY. In New England, 6% of stations exhibit a negative trend. In addition to the single station results, we cluster similar wind sites together using the k-means algorithm, to extend observation records and observe nonstationary behavior of extreme winds regionally. Incorporating L-moments in regional frequency analysis for clustering purposes requires regional standardized L-moment parameters (Eslamian et al., 2012; Parida et al., 1998). A combination of three parameters consisting of latitude, longitude and L-CV is used to define the k-means clustering. Regional frequency analysis (Hosking and Wallis, 1997) is carried out using L-moments to confirm homogeneity of the cluster prior to evaluation of nonstationarity. Once the clusters are defined, we employ the "superstation" (Peterka, 1992; Peterka and Shahid., 1993) method for each cluster to extend and regionalize the record. By creating a virtual "superstation", we have a single database with extended records and reduced sample errors. Using trend tests, we find evidence of statistically significant regional nonstationarity of 3-s gust wind speeds in 4 out of 7 clusters, all resulting in a positive trend. Two clusters are in Florida, one is along the mid-coast and the final one is New Englnad. The cluster in New England exhibits heterogeneity according to the L-moment homogeneity tests. The trend tests for three of the regional clusters do not exhibit a statistically significant trend. For the two statistically significant clusters in Florida, we apply a nonstationary homoscedastic trend model to identify return levels with time of interest at years 2010, 2030, 2050.
Thesis (M.S.)--Tufts University, 2018.
Submitted to the Dept. of Civil Engineering.
Advisor: Laurie Baise.
Committee: Richard Vogel, Franklin Lombardo, and Jonathan Lamontagne.
Keyword: Civil engineering.read less
|
This object is in collection