STATISTICAL ESTIMATION OF WAVE CLIMATES IN A MONSOON REGION Akira Kimura1 and Takao Ota2 This paper deals with a statistics of wave climate. NOWPHAS data from 7 observatories along the Sea of Japan coast are investigated. Fourier spectrum of one year time series of significant wave height shows inherent characteristic to the monsoon region. It can be divided into three frequency regions; (A) , (B) and (C) where . Simple but π»apπpβπropriate spπeπctrum modelinπgπ for the regionπ πis proposed in thπiπs study. It isπ πpossible to πmπake Monte-Carlo simulatioπns< of0 w.0a0v6e πclimates a0p.0p0ly6iπng th<e πm<od0el. 2v πery easily. All 0st.a2tπistica<l πpr<ope2rπties can be cπalcu=late1dβ πfrπoπ¦m the simulated data. No formulation for the frequency distribution of wave properties such as annual maximum wave height is necessary. Some useful applications of the present study are shown in the last. Keywords: wave climate, spectrum of a time series of , simulation of wave statistics, Monte-Carlo method π»πβπ INTRODUCTION Design manual for public structures revised in Japan in 2007. The new manual adopts a concept of performance based design. Slight failures of a structure are allowed as far as the structure keeps the demanded performance. In the former manual, a specification based design was used so that damage must not occur during the lifetime. Design wave conditions were determined applying an extreme value statistics (Goda, 2000). After the revision, it becomes necessary to estimate all waves which may damage the structure during its lifetime. For better design and management of structures after the revision, statistics of wave climate become necessary. However the number of relating researches is very limited. Hirose et al., (1982) showed that a frequency distribution of a daily averaged is very close to the Weibull distribution. A difficult point in wave climate study is that some data in the population have correlations. However statistic theory demands the data independency. On the other hand, it is necessarπ»yπ βtoπ show the wave climates in a simple form for its better application to the life cycle management (LCM) of the structure. The present study aims to overcome the restriction of the theory and to clarify statistical characteristics of the wave climates. A spectrum concept is introduced. W data from 7 observatories (NOWPHAS) along the Sea of Japan coast from 1991 to 2008 (18 years) were investigated and showed one year time series of have a inherent spectral characteristic to the monsoon regiona vreegardless of the observatories and years. The spectrum can be divided into three frequency regions, (A) , (B) and (Cπ») πβπ where . The spectrum in the regionπ (πA) decides theπ sπeasonal changπeπ of with aπ sπmall numbeπrπ of Fourier πcπomponents. Very simple mπod<el0in.0g 0f6or π the region0s.0 (0B6) πand <(c)π ar<e p0o.2s sπible. Statistica0l. 2pπrope<rtiπes< of2 πwave climatπes a=re 1sβimπuπlπ¦ated by Monte-Carlo method using the models. The present studπ»yπ βiπs applicable to study not only wave climates but also extreme wave statistics. Some applications which may be useful in the LCM are shown in the last. STATISTICAL PROPERTIES OF THE OBSERVED DATA The present method bases on the observed data from 7 observatories. Figure 1 shows the locations of the observatories. Figures 2(a) and 2(b) show examples of one year data (observed every 2 hours). These figures show the data from the coast in the north and center of the Sea of Japan (Fig.1). The numbers in the horizontal axis are the days counted from Jan. 1st. Here after the same horizontaπ»l πaβxπis is used. Data in the both ends of the figures (winter) are large, and small in the center part (summer). The term βmonsoonβ in the title of the present study means that data (wave climate) show seasonal characteristics like Fig.2. Figures 3(a) and 3(b) show spectra (absolute value) of one year time series of . The spectrum is given by π»πβπ exp πΉβ²(ππ) ππππ in which = , πΉβ²( πiπs) a= dπiscreπ»teπβ πf(reπ‘qπ)uenc(yβ πw2iπthπ πaπ‘πn) , i n t e r v(aπl =of0 ,1,2,β―,2190 )w h e r e (1) πππ , is aπ tπime sequence with 2 hours interval, data number of one yπeπar total is π43π80 ( π β). β1 ππ = iπnβ F3i6gs5. β3π(a) and 3(b) show the shape that looks like very we1llβ. 3F6ig5uβreπ 4 shows a pπatter=n d1iβaπgπraπ¦m π‘oπ f( πth=e 1s,h2a,pβ―e., 4T3h8e0 s)pectrum can be divided into 3 frequency regions. Their frequency region and c=ha1ra2ctΓer3is6ti5c are|πΉ lβ²i(sπteπd) |in Table 1. The upper bound of the region in (C) is set to avoid aliasing effect. In this study, totally 128 years (7 observatories are inveπsπtigated. Due to space limitation, only limited number of results is shown here after. However there are no πsig<ni2fiπcant differences between the shown and other cases. π»πβπ Γ18 years) 1 NEWJEC, Koyama Kita 6-448-69, Tottori, 680-0941, Japan 2 Tottori Univ., Faculty of Eng., Koyama Minami 4-101, Tottori, 680-8552, Japan 1 2 COASTAL ENGINEERING 2012 Rumoi Setana Sea of Japan ta Akita Niigata Wajima Tottori Hamadaa Figure 1. Locations of observatories. 7 6 5 m) 4 Rumoi 1994 ( 1/3 3 H 2 1 0 0 50 100 150 200 250 300 350 days (a) Rumoi (1994) 7 6 5 4 Niigata 1998 1/3 H 3 2 1 0 0 50 100 150 200 250 300 350 days (b) Niigata (1998) Figure 2. Observed one year time series of , (a) Rumoi (1994) and (b) Niigata (1998) π»πβπ COMPONENTS FROM 3 FREQUENCY REGIONS Seasonal characteristics can be written as sum of components from 3 regions. π»πβπ(π‘) COASTAL ENGINEERING 2012 3 in which , and are wave forms frπ»oπmβπ e(aπ‘)ch= reπ»gπi(oπ‘n) A+, π»Bπ β²a(nπ‘)d +C π»rπesβ²(pπ‘e)c t i v e l y . T h e y a r e g i v e n b y (th2e) inverse Fourier transformations. π»π π»πβ² π»πβ² exp π π»π(π‘π)=π πΉβ²(ππ) (π2ππππ‘π)β« πππ βͺ ππ exp βͺ βͺ π»πβ²(π‘π)=π πΉβ²(ππ) (π2ππππ‘π) (π=1,2,β―,4380 ) (3) πππ exp β¬ πππ βͺ βͺ π»πβ²(π‘π)= π πΉβ²(ππ) (π2ππππ‘π)βͺ in which is the time (NOWPHAS dataπ πeπvπery 2 hours), , and β (Table 1). Figures show the examples : , ππ , 5(c) ππ and 5(d) ππ (Tottori, 20π‘0π0). ππβ0.006π πππ=0.2π ππππβ2π 5(a)~5(d) 5(a) π»πβπ(π‘π) 5(b) π»π(π‘π) π»πβ²(π‘π) π»πβ²(π‘π) 1 6 Setana 1991 4 2 ) 0.1 (m 6 )| 4 ππ fn ππ ( 2 ' F 0.01 | 6 4 2 0.001 0.001 0.01 0.1 1 fn (d-1) (a) Setana (1991) 1 6 4 Hamada 2007 2 ) 0.1 m ( 6 )| 4 n ππ (f 2 ππ ' F 0.01 | 6 4 2 0.001 0.001 0.01 0.1 1 fn (d-1) (b) Hamada (2007) Figure 3. Spectrum of the observed one year time series of |πΉβ²(π)| π»πβπ 4 COASTAL ENGINEERING 2012 A B C 1 6 4 2 m) 0.1 ( 6 | ) 4 ππ (fn 2 ππ ' F 0.01 | 6 4 Rumoi 2001 2 0.001 0.001 0.01 0.1 1 fn (d-1) Figure 4. Pattern diagram of the spectrum Table 1 Frequency regions and characteristics Region A 0.0π0π6 quick decrease with B 0π.π0 0 (6π 0).2 almost constant πΉ(ππ) C ~0. 2 decrease in proportioπnπ to ~ ππ 7 ~ 2 ππ 6 5 (m) 4 Tottori 2000 H 1/33 2 1 0 0 50 100 150 200 250 300 350 days Figure 5(a) (Tottori, 2000) 2.5 π»πβπ(π‘π) 2.0 1.5 m) Tottotttoorrii 22000000 ( HA1.0 0.5 0.0 0 50 100 150 200 250 300 350 days Figure 5(b) (Tottori, 2000) 3 π»π(π‘π) 2 Tottori 2000 ) 1 (m H'B 0 -1 -2 0 50 100 150 200 250 300 350 days Figure 5 (c) β² (Tottori, 2000) π»π(π‘π) COASTAL ENGINEERING 2012 5 3 2 Tottori 2000 H' (m)C 10 -1 -2 0 50 100 150 200 250 300 350 days Figure 5 (d) β² (Tottori, 2000) π»π(π‘π) 2.0 Tottori 2000 0.6 1.5 HA HA (m) 1.0 H'Bsm 0.4' (m)sBm H 0.5 0.2 0.0 0.0 0 50 100 150 200 250 300 350 days (a) and 2.0 π π»π(π‘π) π»πππ(π‘π) Setana 2003 1.5 0.6 (m) 1.0 HH'ACsm 0.4 (m)sm HA 'C H 0.5 0.2 0.0 0.0 0 50 100 150 200 250 300 350 days (b) and π Figure 6. Comparison between (solid line ; π»leπft( π‘aπ)xis) andπ» mππoπnt(hπ‘lπy) standard deviation (black circle ; right axis) of (a) β² (Tottori, 2000) and (b) β² (Setana,2003) π»π(π‘π) Figures 6(a) and 6(b) show coπ»mπp(aπ‘r)isons between (lπ»eπft( π‘a)xis) and the standard deviation of (a) and (b) (right axis), respectively. and are the monthly standard deviation of and . is a dayπ at ceπ»ntπe(rπ‘ eπ)veryπ month. In the both figures, good agreemeπ»nπtsβ² (aπ‘rπe) observed.π» Tπβ²h(iπ‘sπ )means that seasonal characteπ»rπisπtiπcs( π‘πo)f π»π πaπnd(π‘ π) are almost proportional to . π»Thπeβ²(reπ‘πf)ore the π»seπaβ²(sπ‘oπn)al π‘cπha(rπa=cte1ri,sβ―tic,1s 2ca)n be removed by dividing and by . π»πβ²(π‘) π»πβ²(π‘) π»π(π‘π) π»πβ²(π‘) π»πβ²(π‘) π»π(π‘π) Figures 7(a) andπ» 7π((bπ‘)π )sh=owπ» πβ²(π‘π)βπ» πan(π‘dπ ) π»,π r(eπ‘sπp)e=ctiπ»veπlβ²y(.π‘ πC)βorπ»rπes(pπ‘πo)n d i n g (π=1, β―an,d4 380) a r e s h o (w4n) in Figs. 5(c) and 5(d). Seasonal characteristics of Figs. 5(c) and 5(d) are not seen in the figures. Since the spectrum width of and arπ»e πv(eπ‘rπy) narrowπ» πa(nπ‘dπ )their monthly standard deviatioπ»nsπ β²a(rπ‘eπ )almost π»coπnβ²(sπ‘tπa)nt, it can be assumed that these wave forms are stationary. π»π(π‘π) π»π(π‘π) 6 COASTAL ENGINEERING 2012 3 2 Tottori 2000 1 ) m ( HB 0 -1 -2 0 50 100 150 200 250 300 350 days (a) 3 π»π(π‘π) 2 Tottori 2000 1 ) m ( HC 0 -1 -2 0 50 100 150 200 250 300 350 days (b) Figure 7. Seasonal characteristics are removed from data in Figs. 5(c) and 5(d) using eq.(4) (Tottori, 2000). π»π(π‘π) 0.1 | 8 ) 6 n (f 4 F | 2 , |) 0.01 Setana 2006 n 8 (f 6 F' 4 | |F'(fn)|: with seasonal change 2 |F(fn)|: without seasonal change 0.001 2 3 4 5 6 7 8 9 2 0.01 0.1 fn (d-1) (a) Comparison of the spectrum of and 2 π»πβ²(π‘π) π»π(π‘π) 0.1 8 | 6 ) fn 4 ( F | 2 , |) 0.01 fn 8 Setana 2006 ( 6 ' F 4 | |F'(fn)|: with seasonal change 2 |F(fn)|: without seasonal change 0.001 2 3 4 5 6 7 8 9 -1 1 fn(d ) (b) Comparison of the spectrum of and Figure 8. Comparison of the spectrum with and without seasonal characteristics (Setana,2006). π»πβ²(π‘π) π»π(π‘π) New spectrum for and are given as π»π(π‘π) π»π(π‘π) COASTAL ENGINEERING 2012 7 exp ππππ πΉ(ππ)= π π»π(π‘π) (βπ2ππππ‘π), (π=3,β―,73) β« βͺ πππ ππππ exp (5) β¬ Figures 8(a) and 8(b) show comπΉp(aππri)so=nsπ betwπ»eπe(nπ‘ πt)he sp(βecπt2raπ fπoπrπ‘ π()a,) (π= a7n4d, β―,731) βͺand (b) and (Setana, 2006). There are no significπaπnπt changes in the spectra before and after the remβoval of the seasonal characteristics. π»πβ²(π‘π) π»π(π‘π) π»πβ²(π‘π) π» π(π‘π) Phase of From eq.(5), phase spectrum of is given as π(ππ) πΉ(ππ) ππ ImππΉ(ππ)π ππ=tan πβ π (π=3,4,β―,731) (6) in which and mean the real and imagRineaπrπΉy( πpπa)rπts, respectively. Figures 9(a) and 9(b) show frequency distributions of every year for (a) (Hamada) and (b) (Wajima). The distributionRse a(re )almost Iumni(for)ms. πΉ0.π 4(ππ) ππ π»π(π‘π) π»π(π‘π) Hamada 1991-2008 0.3 ) (en0.2 R F 0.1 0.0 1 2 3 4 5 6 en (a) 0.30 ππ (π=3,β―,73) Wajima 1991-2008 0.25 0.20 ) (en0.15 R F 0.10 0.05 0.00 1 2 3 4 5 6 en (b) Figure 9. Frequency distributions of the phase of (a) (Hamada) and (b) (Wajima) ππ (π=74,β―,731) π»π π»π MODELING OF Equation (2) is rewritten as, π―πβπ(π) Spectrum of π»anπβdπ (π‘π)=π» aπr(eπ‘ πa)p{p1r+oxπ»imπa(tπ‘eπd) +sim π»pπly( π‘aπs) }( F i g . 4 ) (, π =1,β―,4380) (7) π»π(π‘π) π»π(π‘π) |πΉ(ππ)|π =πΌπ in which and are the abso|luπΉt(eπ πva)|luπe=s oπΌf πtπheπ πm o d e lπ s p e c t r u m i n t h e r e g i o n s A a n d B (8) respectively, and are constants determined by the least square principle. |πΉ(ππ)|π |πΉ(ππ)|π πΌπ πΌπ 8 COASTAL ENGINEERING 2012 ππ π {|πΉ(ππ)|βπΌπ}π β min β« βͺ πππ πππ (9) ππ π β¬ π {|πΉ(ππ)|βπΌπππ } β min βͺ Calculated and show almππoπsπt constant value with small flβuctuations. Figure 10 shows an example of the values (Akita, 1991 - 2008). πΌπ 0.πΌ12π 0.10 0.08 Akita a 20.06 a1 a1 a 0.04 2 0.02 0.00 1992 1994 1996 1998 2000 2002 2004 2006 2008 year Figure 10. Yearly change of and Normal distributions for the fluctuation of and aπΌrπe assuπΌmπed. Their means and standard deviations , and , are calculated from the observed data. πΌπ πΌπ πΌπ πΌπ πΌππ πΌππ SIMULATION OF From eq.(7), is approximated as π―πβπ(π) π»πβπ(π‘) cos ππ π»πβπ(π‘π)=π»π(π‘π)π1+ π (πΌπ+πΌπππ ππ) (2πππππ‘π+πππ) ππππ cos πππ ππ + π (πΌπ+πΌπππ ππ)πππ (2πππππ‘π+πππ)π πππππ π‘π=πππ‘ (π=0,1,β―) (10) where and are independent normal raπnπdπo=m 2nπumπ πbπerπs, w πitπhπ 0= m2eπaπ n πaπnπd 1 . 0 s t a n d a r d d e v i a t i o n , (1an1d) are also independent uniform random number . If , any value can be used for in eq.(10).π Hππoweverπ , iπtπ is preferable to use the same value as the observed data interval. is approximaπ tπedππ as π πππ (0 ~1) ππ‘<πβ4 ππ‘ π»π(π‘) π»π(π‘π)βπ»π(π‘π)+π»π(π‘π)π ππ (12π) π 1 π»π(π‘π)= ππ+π πππsin(2ππππ‘π)+ππcos(2ππππ‘π)π (13) 2 πππ ππππ πβπ π 1 π»π(π‘π)=π π ππ»ππ(π‘π)βπ»π(π‘π)ππ ( π=1,2,β―,4380) (14) in which , 18ππ, ππππ and is an average of 18 years . is an average of 18 years . is the -th year (1991 2008) value of . ππ=ReππΉβ²(ππ)π ππ=ImππΉβ²(ππ)π (π=0,1,2) πΉβ²(ππ) πΉβ²(ππ) is a standard deviation of 18 years . is a normal random number with 0 mean and 1.0 s{etaqn.d(a1r)d} deπ»vπia(tπ‘iπo)n. One year uses the value of sπ»aπm(eπ‘ π) π».π π(π‘π) π ~ π»π(π‘π) π»π(Iπ‘fπ )the observation years are not sufficienπ»tlπyπ l(oπ‘nπ)g, π foπlπlowing definition for can be used instead of eq.(12a). π ππ π»π(π‘π) in which is an average of . One year uses also the value of same . Since itself and its fluctuation are small, there are noπ» bπig(π‘ dπ)ifβferπ»enπc(eπ‘sπ) b+etwπ»eπeπ nπ eπq s . ( 1 2 a ) a n d ( 1 2 b ) . (12b) Constaπ»nπts for the simulationπ» aπt (eπ‘aπc)h,( oib=se1rv,β―ato,3ry8 a4r0e) listed in the Table 2. π ππ π»π(π‘π) COASTAL ENGINEERING 2012 9 Table 2 Constants for the simulation Rumoi Setana Akita Niigata Wajima Tottori Hamada 1.119 1.188 1.088 1.000 1.197 1.093 1.119 0.777 0.805 0.680 0.726 0.777 0.650 0.597 ππ ππ 0.093 0.088 0.144 0.106 0.093 0.062 0.081 0.150 0.057 0.120 0.142 0.150 0.101 0.091 ππ ππ 0.067 -0.022 0.020 0.077 0.067 0.078 0.083 0.107 0.120 0.138 0.108 0.129 0.111 0.123 ππ 0.0818 0.0817 0.1000 0.0824 0.0741 0.0777 0.0778 π»π 0.00718 0.00594 0.00860 0.00634 0.00462 0.00557 0.00722 πΌπ 0.0155 0.0150 0.0184 0.0157 0.0130 0.0134 0.0137 πΌππ 0.00093 0.00085 0.00150 0.00107 0.00101 0.00105 0.00128 πΌπ Uinit: : , , : otherwise. πΌππ πβπ πΌπ πΌππ π CORRECTION AND RESULTS Figures 11(a) and 11(b) show observed and simulated one year (Hamada). 7 π»πβπ(π‘) 6 5 m) 4 Hamada 2000 ( 3 1/ 3 H 2 1 0 0 50 100 150 200 250 300 350 days (a) (observed) π»πβπ(π‘) 6 4 ) m ( 3 / H1 2 0 0 50 100 150 200 250 300 350 days (b) (simulated) Figure 11. Observed (Hamada, 2000) and simulated one year (Hamada) π»πβπ(π‘) Correction π»πβπ(π‘) Simulated in Fig.11(b) shows a similar seasonal characteristic to Fig.11(a). However some data show negative value. Figure 12 shows frequency distributions of the observed { ; solid line, Wajima π»πβπ(π‘) 1991-2008} and simulated { ; dotted line WajimaοΌ10000 years simulation} . ππππππ»πβππ ππππππ»πβππ π»πβπ(π‘) 10 COASTAL ENGINEERING 2012 0.8 Wajima 0.6 Observed (1991-2008) Psim Simulated , 0.4 Pobs 0.2 0.0 -1 0 1 2 3 4 5 H1/3 (m) Figure 12. Frequency distributions of observed (solid line) and simulated (dotted line) π»πβπ(π‘) There is a difference between and in the area of . About 2% in all the simulated values are negative. However and show good agreement in the area of ππππππ»πβππ ππππππ»πβππ π»πβπ(π‘)<1.0 π . In this study, data are corrected applying frequency distribution conversion method. The conversion is made using the cumulative distribπuπtπiπoπnπ»s πwβπhπich are πcπaπlπcuπlπ»aπteβdπ πas, π»πβπ(π‘)>1.0 π ππβπ πΉπππππ»πβππ= π ππππ(π»)ππ» β« βͺ πππ ππβπ (15) β¬ πΉπππππ»πβππ= π ππππ(π»)ππ»βͺ in which and are the cumulative distributionsπ πoπfπ and β respectively. Data are converted using the following relation. πΉπππ πΉπππ ππππ ππππ β in which is a target cumulative distribuπΉtπiπoπn πtπoπ» cπoβπnπvπeπrπte=d πΉtheπ πdπ»atπaβ πfπrπoπmππ t h e s i m u l a t e d d a t a . F o r t h e t a(r1g6e)t distribution,β must be 0 and very close to . However if the observation period is not sufficieπΉntlyβ long, uneven properties appear on the . In that case, modified cumulative distribution of can be used. πΉFigure 13 showπsπ» aπ βmπoβ€di0fiπcation of the frequenπΉcπyππ diπsπ»trπibβπut>ion0.π Dotted line shows . Solid line shows the function to modify . πΉisπ πgπiven as πΉπππ ππππβππππ πΈπππ ππππ πΈπππ βππππππ»πβππ A: π»πβπβ€πΌπ β§ βͺ π»πβπβπΌπ πΈπππππ»πβππ= π2 β1πππππ(πΌπ) πΌπβ€π»πβπβ€πΌπ (17) β¨ πΌπβπΌπ 30x10-3 β©βͺ ππππππΌπ+πΌπβπ»πβππ BβΆ π»πβπβ₯πΌπ B Approximation 20 Pobs-Psim 10 Emod 0 C -10 -20 A -1.0 -0.5 0.0 0.5 1.0 1.5 2.0 H1/3 (m) Figure 13. Modification of the frequency distribution is a straight line between the points A and B (Fig.13). is point symmetric for C : . and are determined so that difference between and becomes minimum . In FπΈigπ.1π3π, and . Modified frequency distribπΈuπtioπnπ is given by (πΌπβπΌπ)β2,0 πΌπ πΌπ ππππβππππ βπΈπππ π π»πβπβ₯πΌππ SubstitutingπΌπ =0. 1iπn eq.(15πΌ)π, i=ts 0cu.3mπuβlative distribution is calculatπeπdππ. β ππ ππππ»1β3π=ππ ππππ»β1β3π+πΈπππππ»1β3π (18) ππππ πΉπππ
Description: