Mobile QR Code QR CODE : Journal of the Korean Society of Civil Engineers

  1. ์ „๋ถ๋Œ€ํ•™๊ต ํ† ๋ชฉ๊ณตํ•™๊ณผ ์„์‚ฌ๊ณผ์ • (Chonbuk National University)
  2. ์ „๋ถ๋Œ€ํ•™๊ต ํ† ๋ชฉ๊ณตํ•™๊ณผ ๋ฐ•์‚ฌ์ˆ˜๋ฃŒ (Chonbuk National University)
  3. ์ „๋ถ๋Œ€ํ•™๊ต ํ† ๋ชฉ๊ณตํ•™๊ณผ ๋ฐ•์‚ฌ๊ณผ์ • (Chonbuk National University)
  4. ์ „๋ถ๋Œ€ํ•™๊ต ํ† ๋ชฉ๊ณตํ•™๊ณผ ๋ถ€๊ต์ˆ˜, ๊ณตํ•™๋ฐ•์‚ฌ (Chonbuk National University)


๋ฒ ์ด์ง€์•ˆ ๊ธฐ๋ฒ•, BIC, ๋นˆ๋„ํ•ด์„, ํ˜ผํ•ฉ Gumbel ๋ถ„ํฌํ˜•
Bayesian method, BIC, Frequency analysis, Mixed Gumbel distribution

  • 1. ์„œ ๋ก 

  • 2. ๋Œ€์ƒ์ž๋ฃŒ

  • 3. Bayesian ๊ธฐ๋ฒ• ๊ธฐ๋ฐ˜์˜ ํ˜ผํ•ฉ Gumbel ๋ถ„ํฌ ๋งค๊ฐœ๋ณ€์ˆ˜ ์ถ”์ •

  •   3.1 Bayesian ๊ธฐ๋ฒ•์„ ์ด์šฉํ•œ ํ˜ผํ•ฉ ํ™•๋ฅ ๋ถ„ํฌ ๋งค๊ฐœ๋ณ€์ˆ˜ ์ถ”์ • ๋ฐฉ๋ฒ•

  •   3.2 Bayesian Markov Chain Monte Carlo ๋ชจ์˜

  •   3.3 ๋ชจ์˜์‹คํ—˜

  • 4. ๋ถ„ํฌํ˜•์— ๋”ฐ๋ฅธ ๋งค๊ฐœ๋ณ€์ˆ˜ ์ถ”์ • ๋ฐ ๋นˆ๋„ํ•ด์„ ํ‰๊ฐ€ ๊ฒฐ๊ณผ

  • 5. ๊ฒฐ ๋ก 

1. ์„œ ๋ก 

์šฐ๋ฆฌ๋‚˜๋ผ์˜ ๊ฐ•์ˆ˜ ํ˜„์ƒ์€ ๊ณ„์ ˆ์ ์œผ๋กœ ์—ฌ๋ฆ„์ฒ ์— ์ง‘์ค‘๋˜๋ฉฐ ์ด ๊ธฐ๊ฐ„์˜ ๋Œ€ํ‘œ์ ์ธ ๊ฐ•์ˆ˜๋ฐœ์ƒ ๋ฉ”์ปค๋‹ˆ์ฆ˜์œผ๋กœ๋Š” ์žฅ๋งˆ์™€ ํƒœํ’์ด ์žˆ๋‹ค. ์ตœ๊ทผ ๋“ค์–ด ์—ฌ๋ฆ„์ฒ  ๊ฐ•์ˆ˜ํŠน์„ฑ์ด ๊ณผ๊ฑฐ์™€ ๋งŽ์ด ๋‹ฌ๋ผ์ง€๊ณ  ์žˆ๋‹ค๋Š” ์—ฐ๊ตฌ๋“ค์ด ๋งŽ์•„์ง€๊ณ  ์žˆ๋‹ค(Park et al., 2008). ์žฅ๋งˆ์™€ ํƒœํ’์œผ๋กœ ์ธํ•œ ๊ฐ•์ˆ˜๋Ÿ‰์€ ๊ฐ•์šฐ์ง€์†์‹œ๊ฐ„ ๋ฐ ๊ฐ•์šฐ๊ฐ•๋„ ์ธก๋ฉด์—์„œ ์„œ๋กœ ๋‹ค๋ฅธ ํŠน์ง•์„ ๊ฐ€์ง€๊ณ  ์žˆ์œผ๋ฉฐ(Kwon et al., 2008b), ์ด๋Ÿฌํ•œ ํŠน์ง•์„ ๊ณ ๋ คํ•œ ๊ฐ•์šฐ๋ถ„์„๊ณผ ๊ด€๋ จ๋œ ์—ฐ๊ตฌ๊ฐ€ ๋‹ค์ˆ˜ ์ง„ํ–‰๋˜์—ˆ๋‹ค. ์ฆ‰, ์šฐ๋ฆฌ๋‚˜๋ผ์˜ ์—ฌ๋ฆ„์ฒ  ๊ทน์น˜๊ฐ•์ˆ˜๋Ÿ‰์€ ์ด๋Ÿฌํ•œ ํŠน์ง•์œผ๋กœ ์ธํ•ด ์ง€์†์‹œ๊ฐ„์ด ๊ธธ์–ด์ง€๋Š” ๊ฒฝ์šฐ ์ด์ค‘์ฒจ๋‘(multi mode)๋ฅผ ๊ฐ€์ง€๋Š” ๊ฒฝ์šฐ๊ฐ€ ๋งŽ์œผ๋ฉฐ(Ho et al., 2003), ๋‹จ์ผ ํ™•๋ฅ ๋ถ„ํฌ๋ฅผ ์ด์šฉํ•œ ์ผ๋ฐ˜์ ์ธ ๊ฐ•์šฐ๋นˆ๋„ํ•ด์„ ์‹œ ์ด๋Ÿฌํ•œ ์ด์ค‘์ฒจ๋‘(multi mode)์˜ ํŠน์ง•์„ ๊ณ ๋ คํ•˜๋Š”๋ฐ ํ•œ๊ณ„๊ฐ€ ์žˆ๋‹ค.

์šฐ๋ฆฌ๋‚˜๋ผ๋Š” ์‹ค๋ฌด์ ์ธ ๊ด€์ ์—์„œ ํ™•๋ฅ ๊ฐ•์šฐ๋Ÿ‰๋„ ์ž‘์„ฑ์„ ์œ„ํ•ด Gumbel ๋ถ„ํฌํ˜•์„ ์ถ”์ฒœํ•˜๊ณ  ์žˆ์œผ๋ฉฐ ๋งค๊ฐœ๋ณ€์ˆ˜ ์ถ”์ •๋ฐฉ๋ฒ•์œผ๋กœ๋Š” ํ™•๋ฅ ๊ฐ€์ค‘๋ชจ๋ฉ˜ํŠธ ๋ฐฉ๋ฒ•์ด ์ผ๋ฐ˜์ ์œผ๋กœ ์‚ฌ์šฉ๋˜๊ณ  ์žˆ๋‹ค. ์šฐ๋ฆฌ๋‚˜๋ผ์—์„œ ๊ฐ•์šฐ๋นˆ๋„ํ•ด์„ ์‹œ ํ™•๋ฅ ๋ถ„ํฌํ˜• ์„ ์ •์— ๋Œ€ํ•œ ์—ฐ๊ตฌ๊ฐ€ ๋‹ค์ˆ˜ ์ง„ํ–‰๋˜์—ˆ๋‹ค. Heo et al.(1999)์€ ์ „๊ตญ 22๊ฐœ ๊ด€์ธก์ง€์ ์˜ ์—ฐ ์ตœ๋Œ€์น˜๊ณ„์—ด ์ž๋ฃŒ๋ฅผ ์ด์šฉํ•˜์—ฌ ์šฐ๋ฆฌ๋‚˜๋ผ์˜ ๊ฐ•์ˆ˜ ํ™•๋ฅ ๋ถ„ํฌํ˜•์„ ๋ถ„์„ํ•œ ๊ฒฐ๊ณผ, GEV ๋ถ„ํฌ๋ฅผ ์šฐ๋ฆฌ๋‚˜๋ผ ์—ฐ์ตœ๋Œ€ ๊ฐ•์šฐ์ž๋ฃŒ์— ๊ฐ€์žฅ ์ ํ•ฉํ•œ ํ™•๋ฅ ๋ถ„ํฌํ˜•์œผ๋กœ ์„ ์ •ํ•˜์˜€๋‹ค. Lee et al.(2000)์€ 21๊ฐœ ๊ด€์ธก์ง€์ ์˜ ์—ฐ ์ตœ๋Œ€์น˜๊ณ„์—ด ์ž๋ฃŒ๋ฅผ ์ด์šฉํ•˜์—ฌ ์šฐ๋ฆฌ๋‚˜๋ผ์˜ ๊ฐ•์ˆ˜ ํ™•๋ฅ ๋ถ„ํฌํ˜•์„ ๋ถ„์„ํ•œ ๊ฒฐ๊ณผ, Gumbel ๋ถ„ํฌ๋ฅผ ๋Œ€ํ‘œ ํ™•๋ฅ ๋ถ„ํฌํ˜•์œผ๋กœ ์„ ์ •ํ•˜์˜€๋‹ค. Lee et al.(2009)์€ ๋ชจ๋ฉ˜ํŠธ๋ฒ•, ์ตœ์šฐ๋„๋ฒ•, ํ™•๋ฅ ๊ฐ€์ค‘๋ชจ๋ฉ˜ํŠธ๋ฒ•์„ ์ด์šฉํ•˜์—ฌ ๋Œ€ํ‘œํ™•๋ฅ ๋ถ„ํฌํ˜• ํ›„๋ณด๊ตฐ์„ ์„ ์ •ํ•œ ํ›„, resampling ๋ฐฉ๋ฒ•์ธ Jacknife๊ธฐ๋ฒ•์„ ์ ์šฉํ•˜์—ฌ Gumbel ๋ถ„ํฌ๊ฐ€ ์ง€์†๊ธฐ๊ฐ„ 12์‹œ๊ฐ„, 24์‹œ๊ฐ„์— ๋Œ€ํ•ด ๊ฐ€์žฅ ์ ํ•ฉํ•œ ๊ฒƒ์œผ๋กœ ํ‰๊ฐ€ํ–ˆ๋‹ค.

๋งค๊ฐœ๋ณ€์ˆ˜์  ํ™•๋ฅ ๋ถ„ํฌํ˜•๊ณผ๋Š” ๋‹ค๋ฅด๊ฒŒ ํ•ต๋ฐ€๋„ํ•จ์ˆ˜(kernel density function)๋ฅผ ์ด์šฉํ•œ ๋น„๋งค๊ฐœ๋ณ€์ˆ˜์  ๋ฐฉ๋ฒ•์— ๋Œ€ํ•œ ์—ฐ๊ตฌ๋„ ๋‹ค์ˆ˜ ์ง„ํ–‰๋˜์—ˆ๋‹ค(Kwon et al., 2007). ๋Œ์œ„ํ—˜๋„ ๋ถ„์„์‹œ ๋ถ„ํฌํ˜• ์„ ์ •์ด ์š”๊ตฌ๋˜์ง€ ์•Š๋Š” ๋น„๋งค๊ฐœ๋ณ€์ˆ˜์  ๋ฐฉ๋ฒ•์„ ํ™œ์šฉํ•˜์—ฌ ๋‹ค์ค‘์ฒจ๋‘๋ฅผ ๊ฐ€์ง€๋Š” ๊ฒฝ์šฐ์—๋„ ์œ ์—ฐํ•˜๊ฒŒ ํ™•๋ฅ ๋ถ„ํฌ ์ถ”์ •์ด ๊ฐ€๋Šฅํ•˜๋‹ค๊ณ  ์ œ์‹œํ–ˆ๊ณ , ๊ธฐ์กด ๋งค๊ฐœ๋ณ€์ˆ˜์  ๋ฐฉ๋ฒ•์—์„œ ๋Œ์•ˆ์ „์„ฑ์ด ๋น„ํ˜„์‹ค์ ์œผ๋กœ ์ถ”์ •๋˜๋Š” ๋ฐ˜๋ฉด, ๋น„๋งค๊ฐœ๋ณ€์ˆ˜์  ๋ฐฉ๋ฒ•์€ ์ฃผ์š” ๋งค๊ฐœ๋ณ€์ˆ˜๋“ค์˜ ํ•ฉ๋ฆฌ์ ์ธ ๋Œ์•ˆ์ „์„ฑ ํ‰๊ฐ€๋ฅผ ์ œ๊ณตํ•˜์˜€๋‹ค(Kwon and Moon, 2006). ํŠนํžˆ, ์ž๋ฃŒ์˜ ๊ธฐ์ง€์  ๋‚ด๋ถ€์˜ ๋‚ด์‚ฝ์— ์žˆ์–ด์„œ๋Š” ์šฐ์ˆ˜ํ•œ ์„ฑ๋Šฅ์„ ๋ฐœํœ˜ํ•˜๋Š” ์žฅ์ ์ด ์žˆ๋‹ค. ํ•ต๋ฐ€๋„ํ•จ์ˆ˜ ๋ฐฉ๋ฒ•์€ ์—ฌ๋Ÿฌ ์žฅ์ ์ด ์žˆ๋Š” ๋ฐ˜๋ฉด ๊ผฌ๋ฆฌ(tail)๋ถ€๋ถ„์˜ ์™ธ์‚ฝ์˜ ๊ฒฝ์šฐ ์ƒ๋Œ€์ ์œผ๋กœ ๋นจ๋ฆฌ ์ˆ˜๋ ดํ•˜๋Š” ๋‹จ์ ์ด ์žˆ์–ด ๋นˆ๋„ํ•ด์„ ์‹œ ํŠน์ •๋นˆ๋„ ์ด์ƒ์—์„œ๋Š” ํ™•๋ฅ ๊ฐ•์ˆ˜๋Ÿ‰์˜ ๋ณ€ํ™”๊ฐ€ ๋ฏธ๋ฏธํ•œ ๊ฒฝ์šฐ๊ฐ€ ์žˆ์œผ๋ฉฐ, ์ด๋Ÿฌํ•œ ๋ฌธ์ œ์ ์„ ํ•ด๊ฒฐํ•˜๊ณ ์ž ํ•ต๋ฐ€๋„ํ•จ์ˆ˜๋ฅผ ๊ฐœ๋ฐœํ•˜๋Š” ์—ฐ๊ตฌ๋„ ์ง„ํ–‰๋˜์—ˆ๋‹ค(Kwon and Moon, 2006). ๋น„๋งค๊ฐœ๋ณ€์ˆ˜์  ๋ฐฉ๋ฒ•์˜ ๋˜ ๋‹ค๋ฅธ ๋ฌธ์ œ์ ์€ Quantile ํ•จ์ˆ˜๊ฐ€ ์กด์žฌํ•˜์ง€ ์•Š์œผ๋ฉฐ, ์ˆ˜์น˜์ ์ธ ์ ‘๊ทผ๋งŒ์ด ๊ฐ€๋Šฅํ•˜๋‹ค๋Š” ์ ์ด๋‹ค. ์ด๋Ÿฌํ•œ ์ ์—์„œ 2๊ฐœ ์ด์ƒ์˜ ๋งค๊ฐœ๋ณ€์ˆ˜์  ํ™•๋ฅ ๋ถ„ํฌ๊ฐ€ ํ˜ผํ•ฉ๋œ ํ˜ผํ•ฉ๋ถ„ํฌ์˜ ์ ์šฉ์— ๋Œ€ํ•œ ์—ฐ๊ตฌ๊ฐ€ ํ•„์š”ํ•˜๋‹ค.

ํ†ต๊ณ„์ ์œผ๋กœ ๋‹จ์ผํ•œ ๋ชจ์ง‘๋‹จ์ด๋ผ๋Š” ๊ฐ€์ • ํ•˜์—์„œ ์šฐ๋ฆฌ๋‚˜๋ผ์˜ ๊ฒฝ์šฐ Gumbel ๋ถ„ํฌ๊ฐ€ ์ „๋ฐ˜์ ์œผ๋กœ ์ตœ์  ๋ถ„ํฌํ˜•์œผ๋กœ ์„ ์ •๋˜๊ณ  ์žˆ์ง€๋งŒ, ๋งŽ์€ ๊ฒฝ์šฐ์— ์žˆ์–ด์„œ ํ˜ผํ•ฉ๋ถ„ํฌ์˜ ํ˜•ํƒœ๋ฅผ ๊ฐ–๋Š” ๊ฒฝ์šฐ๊ฐ€ ์žˆ์œผ๋ฉฐ ์ด๋กœ ์ธํ•ด ๊ฐ•์šฐ๋นˆ๋„ํ•ด์„ ์‹œ ๊ณผ๋Œ€/๊ณผ์†Œ ์ถ”์ •๋˜๋Š” ๋ฌธ์ œ๊ฐ€ ๋ฐœ์ƒํ•  ์ˆ˜ ์žˆ๋‹ค. ์ด๋Ÿฌํ•œ ํ˜ผํ•ฉ๋ถ„ํฌ์˜ ๋ฐœ์ƒ์›์ธ์€ ์—ฌ๋Ÿฌ ๊ฐ€์ง€๊ฐ€ ์žˆ์„ ์ˆ˜ ์žˆ๋‹ค. ์ฒซ์งธ, ๊ฐ•์šฐ์˜ ํ‘œ๋ณธ์˜ค์ฐจ(sampling error)๋กœ ๊ธฐ์ธํ•  ์ˆ˜ ์žˆ๋‹ค. ์ฆ‰, ๋นˆ๋„ํ•ด์„์„ ์œ„ํ•œ ์ž๋ฃŒ๊ฐ€ ์ถฉ๋ถ„์น˜ ์•Š์•„์„œ ๋‚˜ํƒ€๋‚˜๋Š” ์ผ์‹œ์ ์ธ ํ˜„์ƒ์ผ ์ˆ˜ ์žˆ๋‹ค. ๋‘˜์งธ, ๊ฐ•์šฐ์˜ ํŠน์„ฑ์ด ๋‹ค๋ฅธ ๋‘ ๊ฐœ ์ด์ƒ์˜ ๋ฐœ์ƒ์›์ด ์„œ๋กœ ํ˜ผ์žฌ๋˜์–ด ์žˆ๋Š” ๊ฒฝ์šฐ๋กœ ๊ฐ€์ •ํ•  ์ˆ˜ ์žˆ๋‹ค. ํ˜ผํ•ฉ๋ถ„ํฌํ˜•๊ณผ ๊ด€๋ จ๋œ ๊ตญ๋‚ด์—ฐ๊ตฌ ์‚ฌ๋ก€์—์„œ๋Š” ๊ฐ•์šฐ์ž๋ฃŒ๋ฅผ ๋ชจ์ง‘๋‹จ๋ณ„๋กœ ๋ถ„๋ฅ˜(ํŠน์ •์‚ฌ๊ฑด ๋˜๋Š” ๊ธฐ๊ฐ„๋ณ„๋กœ ์—ญ์ถ”์ ํ•˜์—ฌ ์žฅ๋งˆ์™€ ํƒœํ’์œผ๋กœ ๋ถ„๋ฅ˜)ํ•ด์„œ ํ•ด์„ํ•˜๋Š” ๋ฐฉ๋ฒ•(Yoon et al., 2012)๊ณผ ํŠน์ •ํ•œ ๋ถ„๋ฅ˜ ์—†์ด ํ˜ผํ•ฉ๋ถ„ํฌ๋ฅผ ์ ์šฉํ•˜์—ฌ ํ•ด์„ํ•˜๋Š” ๋ฐฉ๋ฒ•์œผ๋กœ ๊ตฌ๋ถ„๋œ๋‹ค(Shin and Lee, 2014). ํ˜ผํ•ฉ๋ถ„ํฌ๋ฅผ ์ด์šฉํ•œ ๋นˆ๋„ํ•ด์„ ์—ฐ๊ตฌ๋Š” ๋งค๊ฐœ๋ณ€์ˆ˜ ์ถ”์ •์— ๋Œ€ํ•œ ์–ด๋ ค์›€์œผ๋กœ ์ œํ•œ์ ์œผ๋กœ ์ˆ˜ํ–‰๋˜๊ณ  ์žˆ์œผ๋ฉฐ, ์ตœ์šฐ๋„๋ฒ• ๋˜๋Š” EM (expectation-maximization) ์•Œ๊ณ ๋ฆฌ์ฆ˜์ด ์ฃผ๋กœ ์ œ์•ˆ๋˜๊ณ  ์žˆ์œผ๋ฉฐ, ํŠนํžˆ EM ๊ธฐ๋ฒ•์ด ๊ฐ€์žฅ ๋Œ€ํ‘œ์ ์ธ ํ˜ผํ•ฉ๋ถ„ํฌ ๋งค๊ฐœ๋ณ€์ˆ˜ ์ถ”์ •๋ฐฉ๋ฒ•์ด๋ผ ํ•  ์ˆ˜ ์žˆ๋‹ค. ๊ทธ๋Ÿฌ๋‚˜ EM ๊ธฐ๋ฒ•์˜ ๊ฒฝ์šฐ ์ดˆ๊ธฐ์น˜์— ๋งค์šฐ ๋ฏผ๊ฐํ•˜๋ฉฐ ์ˆ˜์น˜ํ•ด์„์ƒ์˜ ๋ถˆ์•ˆ์ •์„ฑ์œผ๋กœ ์ธํ•ด ์ž๋ฃŒ๊ฐ€ ๋ถ€์กฑํ•œ ๊ฒฝ์šฐ ๋งค๊ฐœ๋ณ€์ˆ˜ ์ถ”์ •์— ์–ด๋ ค์›€์ด ํฌ๋‹ค(Redner and Walker, 1984). ์ด๋Ÿฌํ•œ ์ ์—์„œ ๋ณธ ์—ฐ๊ตฌ์˜ ์ฃผ์š” ๋ชฉ์ ์€ Bayesian ๊ธฐ๋ฒ•์„ ์ด์šฉํ•œ ํ˜ผํ•ฉ๋ถ„ํฌ์˜ ๋งค๊ฐœ๋ณ€์ˆ˜ ์ถ”์ •๋ฐฉ๋ฒ•์„ ๊ฐœ๋ฐœํ•˜๊ณ ์ž ํ•˜๋ฉฐ, ์ด๋ฅผ ํ†ตํ•ด ํ•ด์„์ƒ์˜ ์•ˆ์ •์„ฑ์„ ๋„๋ชจํ•˜๊ณ  ํ‘œ๋ณธ์˜ค์ฐจ์— ๋Œ€ํ•œ ์˜ํ–ฅ์„ ๋ถˆํ™•์‹ค์„ฑ์œผ๋กœ ํ‘œํ˜„ํ•˜๊ณ ์ž ํ•œ๋‹ค. ๋ณธ ์—ฐ๊ตฌ์—์„œ๋Š” ๊ฐ•์šฐ์ž๋ฃŒ๋ฅผ ํŠน์ •์‚ฌ๊ฑด ๋˜๋Š” ๊ธฐ๊ฐ„์„ ์กฐ์‚ฌํ•˜์—ฌ ๋ชจ์ง‘๋‹จ์„ ๋ถ„๋ฆฌํ•˜์ง€ ์•Š์•˜์œผ๋ฉฐ, Bayesian MCMC (Markov Chain Monte Carlo)๊ธฐ๋ฒ•์„ ์ ์šฉํ•˜์—ฌ ๋งค๊ฐœ๋ณ€์ˆ˜ ์ถ”์ • ๋ฐ ๊ฐ•์šฐ๋นˆ๋„ํ•ด์„์„ ์ˆ˜ํ–‰ํ•˜์˜€๋‹ค. ๋ชจํ˜•์˜ ์ ํ•ฉ์„ฑ์„ ํ‰๊ฐ€ํ•˜๊ธฐ ์œ„ํ•˜์—ฌ ๋ชจ์˜์‹คํ—˜์„ ํ†ตํ•˜์—ฌ ๋งค๊ฐœ๋ณ€์ˆ˜ ์ถ”์ • ๊ฒฐ๊ณผ๋ฅผ ๋น„๊ต ๊ฒ€ํ† ํ•˜์˜€๋‹ค. ์ตœ์ข…์ ์œผ๋กœ ์ž๋ฃŒ๊ธฐ๊ฐ„์ด ๋น„๊ต์  ๊ธด ์ „์ฃผ์ง€์ ์„ ๋Œ€์ƒ์œผ๋กœ ์•ž์„œ ๋‹จ์ผ Gumbel ๋ถ„ํฌ๋ชจํ˜•๊ณผ ํ˜ผํ•ฉ Gumbel ๋ถ„ํฌ๋ชจํ˜•์„ ๊ตฌํ˜„ํ•˜์—ฌ ์ฃผ์š” ์ง€์†์‹œ๊ฐ„์— ๋Œ€ํ•˜์—ฌ ์ƒํ˜ธ๋น„๊ต๋ฅผ ์‹ค์‹œํ•˜์˜€๋‹ค.

2. ๋Œ€์ƒ์ž๋ฃŒ

๋ณธ ์—ฐ๊ตฌ์—์„œ๋Š” ์ž๋ฃŒ๊ธฐ๊ฐ„์ด ๋น„๊ต์  ๊ธด ์ „์ฃผ์ง€์ ์˜ 1961๋…„๋ถ€ํ„ฐ 2016๋…„๊นŒ์ง€ 56๋…„์˜ ๊ธฐ์ƒ์ฒญ ๊ฐ•์šฐ์ž๋ฃŒ๋ฅผ ํ™œ์šฉํ•˜์˜€์œผ๋ฉฐ, 8๊ฐœ์˜ ์ง€์†์‹œ๊ฐ„๋ณ„ ์—ฐ์ตœ๋Œ€๊ฐ•์šฐ๋Ÿ‰์„ ์ถ”์ถœํ•˜์—ฌ ์‚ฌ์šฉํ•˜์˜€๋‹ค. ์ง€์†์‹œ๊ฐ„๋ณ„ ๊ฐ•์ˆ˜์ž๋ฃŒ์˜ ํ•ต๋ฐ€๋„ํ•จ์ˆ˜์™€ Gumbel ํ™•๋ฅ ์ง€(probability plot)๋ฅผ ์ด์šฉํ•ด Fig. 1์— ์ œ์‹œํ•˜์˜€๋‹ค. ๋Œ€๋ถ€๋ถ„์˜ ์ง€์†์‹œ๊ฐ„์—์„œ ๋‹ค์ค‘์ฒจ๋‘์˜ ํ˜•ํƒœ๋ฅผ ๋„๋ฉฐ, ๋‹จ์ผ๋ถ„ํฌ๋ณด๋‹ค๋Š” ํ˜ผํ•ฉ๋ถ„ํฌ์˜ ์ ์šฉ์ด ์ž๋ฃŒ์˜ ๋ถ„ํฌํŠน์„ฑ์„ ๋ฌ˜์‚ฌํ•˜๋Š”๋ฐ ์œ ๋ฆฌํ•  ๊ฒƒ์œผ๋กœ ํŒ๋‹จ๋œ๋‹ค. ํ™•๋ฅ ์ง€๋ฅผ ํ†ตํ•ด ๋ณด๋”๋ผ๋„ 1์‹œ๊ฐ„ ์ง€์†์‹œ๊ฐ„์˜ ๊ฒฝ์šฐ ์ด๋ก ์ ์ธ Gumbel ๋ถ„ํฌ๋ฅผ ๋‚˜ํƒ€๋‚ด๋Š” ๋นจ๊ฐ„ ์‹ค์„ ์„ ๋ฒ—์–ด๋‚˜๋Š” ์ž๋ฃŒ๊ฐ€ 2๋ฒˆ์งธ ๋ชจ๋“œ์—์„œ ๋‚˜ํƒ€๋‚˜๊ณ  ์žˆ์œผ๋ฉฐ, 6์‹œ๊ฐ„ ์ง€์†์‹œ๊ฐ„์˜ ๊ฒฝ์šฐ๋„ 2๋ฒˆ์งธ ๋ชจ๋“œ์—์„œ ์ด๋ก ์ ์ธ ๋ถ„ํฌ์—์„œ ๋ฒ—์–ด๋‚˜๊ณ  ์žˆ์Œ์„ ํ™•์ธํ•  ์ˆ˜ ์žˆ๋‹ค. Fig. 1์—์„œ ์ง€์†์‹œ๊ฐ„ 1, 6, 9, 12์‹œ๊ฐ„์˜ ์ž๋ฃŒ์—์„œ ์„œ๋กœ ๋‹ค๋ฅธ ์ฒจ๋‘๊ฐ€ ์œก์•ˆ์œผ๋กœ ์‹๋ณ„๋˜์—ˆ๋‹ค. ๋”ฐ๋ผ์„œ ์ „์ฃผ ์ง€์ ์˜ ๊ฐ•์šฐ์ž๋ฃŒ๋Š” ์„œ๋กœ ๋‹ค๋ฅธ 2๊ฐœ ์ด์ƒ์˜ ๋ชจ์ง‘๋‹จ์œผ๋กœ ์ด๋ฃจ์–ด์กŒ๋‹ค๊ณ  ํŒ๋‹จํ•  ์ˆ˜ ์žˆ๋‹ค.

Fig. 1.

A Graphical Representation of Annual Maximum Rainfall at Jeonju Station. Kernel Density Functions are Displayed in the Left Panel While Gumbel Probability Plots are Illustrated in the Right Panel for Different Durations

Figure_KSCE_38_2_07_F1.jpg
Fig. 1.

A Graphical Representation of Annual Maximum Rainfall at Jeonju Station. Kernel Density Functions are Displayed in the Left Panel While Gumbel Probability Plots are Illustrated in the Right Panel for Different Durations (Continued)

Figure_KSCE_38_2_07_F1_Continued.jpg

3. Bayesian ๊ธฐ๋ฒ• ๊ธฐ๋ฐ˜์˜ ํ˜ผํ•ฉ Gumbel ๋ถ„ํฌ ๋งค๊ฐœ๋ณ€์ˆ˜ ์ถ”์ •

๋ณธ ์ ˆ์—์„œ๋Š” Bayesian ๊ธฐ๋ฒ• ๊ธฐ๋ฐ˜์˜ ๋งค๊ฐœ๋ณ€์ˆ˜ ์ถ”์ •๊ณผ ๋ถˆํ™•์‹ค์„ฑ์„ ํ‰๊ฐ€ํ•˜๋Š” ๋ฐฉ์•ˆ์— ๋Œ€ํ•ด์„œ ์„œ์ˆ ํ•˜์˜€๋‹ค. ์ด์™€ ๋”๋ถˆ์–ด, ๋ณธ ์—ฐ๊ตฌ์—์„œ ์ œ์•ˆํ•œ Bayesian MCMC ๊ธฐ๋ฒ•์„ ๋ชจ์˜์‹คํ—˜์„ ํ†ตํ•˜์—ฌ ๊ฒ€์ฆํ•˜์˜€๋‹ค.

3.1 Bayesian ๊ธฐ๋ฒ•์„ ์ด์šฉํ•œ ํ˜ผํ•ฉ ํ™•๋ฅ ๋ถ„ํฌ ๋งค๊ฐœ๋ณ€์ˆ˜ ์ถ”์ •๋ฐฉ๋ฒ•

Bayesian๊ธฐ๋ฒ•์„ ํ†ตํ•œ ๋งค๊ฐœ๋ณ€์ˆ˜ ์ถ”์ •๊ธฐ๋ฒ•์€ ๊ธฐ์กด ๋ฐฉ๋ฒ•๋“ค(์ตœ์šฐ๋„๋ฒ•, ๋ชจ๋ฉ˜ํŠธ๋ฒ•, ํ™•๋ฅ ๊ฐ€์ค‘๋ชจ๋ฉ˜ํŠธ๋ฒ•)๊ณผ๋Š” ๋‹ค๋ฅด๊ฒŒ ๋งค๊ฐœ๋ณ€์ˆ˜๋ฅผ ํ•˜๋‚˜์˜ ํ™•๋ฅ ๋ณ€์ˆ˜๋กœ ์ทจ๊ธ‰ํ•œ๋‹ค. ์ฆ‰, ๋งค๊ฐœ๋ณ€์ˆ˜๊ฐ€ ๋‹จ์ผ ๊ฐ’์ด ์•„๋‹Œ ํ™•๋ฅ ๋ถ„ํฌ์˜ ํ˜•ํƒœ๋กœ ๋ถ€์—ฌ๋˜๋ฉฐ ์ตœ์ข…์ ์œผ๋กœ ๋งค๊ฐœ๋ณ€์ˆ˜์˜ ์‚ฌํ›„ ๋ถ„ํฌ(posterior distribution)๋ฅผ ์ถ”์ •ํ•˜๋Š”๋ฐ ๋ชฉ์ ์„ ๋‘๋ฉฐ, Bayesโ€™ ์ •๋ฆฌ(Eq. (1))๋ฅผ ๊ธฐ๋ฐ˜์œผ๋กœ ํ•œ๋‹ค.

PICA39A.gif (1)

์—ฌ๊ธฐ์„œ, PICA3BA.gif์€ ์‚ฌํ›„ ๋ถ„ํฌ๋กœ, 2์žฅ์—์„œ ์ œ์‹œํ•œ ๊ฐ•์šฐ์ž๋ฃŒ์˜ ๊ฐ ๋ชจ์ง‘๋‹จ๋“ค์ด Gumbel ๋ถ„ํฌ(PICA3BB.gif), (PICA3CC.gif)๋ฅผ ๋”ฐ๋ฅธ๋‹ค๊ณ  ๊ฐ€์ •ํ•˜๋ฉฐ, PICA3EC.gif๋Š” Gumbel ๋ถ„ํฌ์— ๋Œ€ํ•œ ์ „์ฒด๋งค๊ฐœ๋ณ€์ˆ˜๋“ค์˜ ์ง‘ํ•ฉ์„ ๋‚˜ํƒ€๋‚ธ๋‹ค. PICA3FD.gif๋Š” ๊ด€์ธก์ž๋ฃŒ x์˜ ์ฃผ๋ณ€๋ถ„ํฌ(marginal distribution), PICA40E.gif๋Š” ๋งค๊ฐœ๋ณ€์ˆ˜๋“ค์˜ ์‚ฌ์ „๋ถ„ํฌ๋ฅผ, PICA40F.gif๋Š” ๊ทน์น˜์ž๋ฃŒ x์˜ ์šฐ๋„ํ•จ์ˆ˜(likelihood function)๋ฅผ ์˜๋ฏธํ•˜๋ฏ€๋กœ Eq. (2)์™€ ๊ฐ™์ด ๋‚˜ํƒ€๋‚ผ ์ˆ˜ ์žˆ๋‹ค.

PICA410.gif (2)

         PICA420.gif 

์—ฌ๊ธฐ์„œ, PICA431.gif์€ ๊ฐ•์ˆ˜์‹œ๊ณ„์—ด์˜ ์ž๋ฃŒ์—ฐ์ˆ˜๋ฅผ ๋‚˜ํƒ€๋‚ธ๋‹ค. ๋งค๊ฐœ๋ณ€์ˆ˜๋“ค์˜ ์‚ฌ์ „๋ถ„ํฌ(PICA432.gif)๋Š” ์‚ฌ์ „์ •๋ณด๋ฅผ ๊ธฐ๋ฐ˜์œผ๋กœ ํ•˜๋Š” Informative ์‚ฌ์ „๋ถ„ํฌ์™€ ์‚ฌ์ „์ •๋ณด์— ํฌ๊ฒŒ ์˜์กดํ•˜์ง€ ์•Š๋Š” Noninformative ์‚ฌ์ „๋ถ„ํฌ๊ฐ€ ์žˆ์œผ๋ฉฐ, Noninformative ์‚ฌ์ „๋ถ„ํฌ๋กœ๋Š” ์ •๊ทœ๋ถ„ํฌ, ๊ท ์ผ๋ถ„ํฌ, ์ง€์ˆ˜๋ถ„ํฌ๊ฐ€ ๋Œ€ํ‘œ์ ์œผ๋กœ ํ™œ์šฉ๋œ๋‹ค(Gelman et al., 2004). ์‚ฌ์ „๋ถ„ํฌ๋ฅผ ๊ฒฐ์ •ํ•˜๋Š”๋ฐ ์žˆ์–ด์„œ ๊ณต์•ก๋ถ„ํฌ(conjugate distribution), ์ฆ‰ ์‚ฌ์ „๋ถ„ํฌ์™€ ์‚ฌํ›„๋ถ„ํฌ๊ฐ€ ๋™์ผํ•˜๊ฒŒ ๊ฒฐ์ •๋˜๋Š” ์‚ฌ์ „๋ถ„ํฌ๋ฅผ ํ™œ์šฉํ•˜๋Š” ๊ฒƒ์ด ๋งค๊ฐœ๋ณ€์ˆ˜ ์ถ”์ •์— ์•ˆ์ •์„ฑ์„ ๋„๋ชจํ•  ์ˆ˜ ์žˆ์œผ๋‚˜, ๋ณธ ์—ฐ๊ตฌ์™€ ๊ฐ™์ด ๊ณต์•ก๋ถ„ํฌ ์ถ”์ •์ด ์–ด๋ ค์šด ๊ฒฝ์šฐ, ๋น„๊ณต์•ก๋ถ„ํฌ๋ฅผ ์‚ฌ์šฉํ•˜๋Š” ๊ฒƒ์ด ์ผ๋ฐ˜์ ์ด๋‹ค(Gelman et al., 2004). ์ด๋Ÿฌํ•œ ์ ์—์„œ ๋ณธ ์—ฐ๊ตฌ์—์„œ๋Š” ์‚ฌ์ „๋ถ„ํฌ๋กœ ๊ฐ๊ฐ์˜ ๋งค๊ฐœ๋ณ€์ˆ˜๋ฅผ ์ •๊ทœ๋ถ„ํฌ์™€ Gamma ๋ถ„ํฌ, Dirichlet ๋ถ„ํฌ๋กœ ๊ฐ€์ •ํ•˜์˜€๋‹ค. ๋‹ค์‹œ ๋งํ•ด, ์œ„์น˜๋งค๊ฐœ๋ณ€์ˆ˜์— ๋Œ€ํ•œ ์‚ฌ์ „๋ถ„ํฌ๋กœ์„œ ๋ถ„์‚ฐ์ด ํฐ ์ •๊ทœ๋ถ„ํฌ๋ฅผ ํ™œ์šฉํ•˜์˜€์œผ๋ฉฐ, ๊ทœ๋ชจ๋งค๊ฐœ๋ณ€์ˆ˜์— ๋Œ€ํ•ด์„œ๋Š” ์Œ์˜ ๊ฐ’์„ ๋ฐฉ์ง€ํ•˜๊ธฐ ์œ„ํ•˜์—ฌ Gamma ๋ถ„ํฌ๋ฅผ ํ™œ์šฉํ•˜์˜€๋‹ค. ๋˜ํ•œ, ๊ฐ ๋ชจ์ง‘๋‹จ๋ณ„ ๊ฐ€์ค‘์น˜(weight)๋งค๊ฐœ๋ณ€์ˆ˜๋Š” ์ดํ•ฉ์ด 1์ธ ๊ฒƒ์„ ์ด์šฉํ•˜์—ฌ Dirichlet ๋ถ„ํฌ๋ฅผ ํ™œ์šฉํ•˜์˜€๋‹ค. ๋ณธ ์—ฐ๊ตฌ์—์„œ ์ถ”์ •์ด ํ•„์š”ํ•œ 5๊ฐœ์˜ ๋งค๊ฐœ๋ณ€์ˆ˜ ์ˆ˜์— ๋น„ํ•ด ์ „์ฃผ์ง€์  56๋…„์˜ ์ž๋ฃŒ ์—ฐ์ˆ˜๊ฐ€ ์ถฉ๋ถ„ํžˆ ํฐ ์ ์„ ๊ณ ๋ คํ•  ๋•Œ, Noninformative ์‚ฌ์ „๋ถ„ํฌ๋ฅผ ํ†ตํ•œ ์ถ”์ •์—๋„ ํฐ ๋ฌด๋ฆฌ๊ฐ€ ์—†์„ ๊ฒƒ์œผ๋กœ ํŒ๋‹จํ•˜์˜€๋‹ค.

PICA442.gif (3a)

PICA443.gif (3b)

PICA454.gif (3c)

PICA455.gif (3d)

PICA466.gif (3e)

Eqs. (3a)~(3e)์—์„œ ์œ„์น˜๋งค๊ฐœ๋ณ€์ˆ˜ PICA467.gif๋Š” ๊ฐ๊ฐ PICA477.gif๊ณผ PICA478.gif๋ฅผ ๊ฐ–๋Š” ์ •๊ทœ๋ถ„ํฌ๋ฅผ ๋”ฐ๋ฅด๊ณ , ๊ทœ๋ชจ๋งค๊ฐœ๋ณ€์ˆ˜ PICA489.gif๋Š” ๊ฐ๊ฐ PICA48A.gif์™€ PICA49B.gif๋ฅผ ๊ฐ–๋Š” Gamma ๋ถ„ํฌ๋ฅผ ๋”ฐ๋ฅด๋ฉฐ, ๊ฐ€์ค‘์น˜๋งค๊ฐœ๋ณ€์ˆ˜ PICA49C.gif๋Š” Dirichlet ๋ถ„ํฌ๋ฅผ ๋”ฐ๋ฅธ๋‹ค. ๋ณธ ์—ฐ๊ตฌ์—์„œ๋Š” ๊ฐ ๋ชจ์ง‘๋‹จ์—์„œ ๊ฐ€์ค‘์น˜PICA49D.gif์™€ ๊ฐ Gumbel ๋ถ„ํฌ์˜ 2๊ฐœ์˜ ๋งค๊ฐœ๋ณ€์ˆ˜๊ฐ€ ๋ชจ๋‘ ๋ถˆํ™•์‹ค์„ฑ์„ ๊ฐ€์ง€๊ณ  ์žˆ๋‹ค๋Š” ๊ฐ€์ •์•„๋ž˜ ํ™•๋ฅ ์  ์ถ”๋ก (statistical inference)์„ ์ˆ˜ํ–‰ํ–ˆ์œผ๋ฉฐ, 5๊ฐœ์˜ ๋งค๊ฐœ๋ณ€์ˆ˜์— ๋Œ€ํ•œ ๊ฒฐํ•ฉ ํ™•๋ฅ ๋ฐ€๋„ํ•จ์ˆ˜(joint distribution)๋Š” Eq. (4)์™€ ๊ฐ™๋‹ค.

PICA4AD.gif (4)

Eq. (2)~(4)์—์„œ ์ •์˜๋˜๋Š” ๋งค๊ฐœ๋ณ€์ˆ˜๋“ค์˜ ์‚ฌ์ „๋ถ„ํฌ๋“ค์„ Eq. (1)์— ๋Œ€์ž…์‹œํ‚ด์œผ๋กœ์„œ ๋งค๊ฐœ๋ณ€์ˆ˜๋“ค์˜ ์‚ฌํ›„๋ถ„ํฌ๋ฅผ Eq. (5)์™€ ๊ฐ™์ด ์ถ”์ •ํ•  ์ˆ˜ ์žˆ๋‹ค.

PICA4BE.gif

(5)

Eq. (5)์—์„œ ๋ชจ๋“  ๋งค๊ฐœ๋ณ€์ˆ˜์— ๋Œ€ํ•œ ์ ๋ถ„์„ ํ†ตํ•ด ์ง์ ‘์ ์œผ๋กœ ์ถ”์ •ํ•˜๋Š” ๊ฒƒ์€ ๋ถˆ๊ฐ€๋Šฅํ•˜๋‹ค. ๋”ฐ๋ผ์„œ, ๋ณธ ์—ฐ๊ตฌ์—์„œ๋Š” ์•ž์„œ ์–ธ๊ธ‰ํ•œ MCMC๋ฐฉ๋ฒ•์„ ๋„์ž…ํ•˜์—ฌ ๊น์Šคํ‘œ๋ณธ๋ฒ•(gibbs sampling)์„ ์ด์šฉํ•˜์—ฌ ๋งค๊ฐœ๋ณ€์ˆ˜๋“ค์˜ ์‚ฌํ›„๋ถ„ํฌ๋ฅผ ์ถ”์ •ํ•˜์˜€๋‹ค.

3.2 Bayesian Markov Chain Monte Carlo ๋ชจ์˜

Bayesian MCMC๊ธฐ๋ฒ•์€ ์‚ฌํ›„๋ถ„ํฌ๋ฅผ ์ถ”์ •ํ•˜๋Š”๋ฐ ์žˆ์–ด์„œ ๋‹ค๋ณ€๋Ÿ‰์— ๋Œ€ํ•œ ๋ณต์žกํ•œ ์ ๋ถ„์„ ์œ„ํ•ด์„œ ์ ์šฉ๋˜๋Š” ์ˆ˜์น˜ํ•ด์„ ๊ธฐ๋ฒ•์œผ๋กœ ๋งค๊ฐœ๋ณ€์ˆ˜๋ฅผ ์ถ”์ •ํ•˜๋Š” ์ˆ˜๋‹จ์œผ๋กœ ํ™œ์šฉ๋˜๊ณ  ์žˆ๋‹ค(Kwon et al., 2008a). ์ผ๋ฐ˜์ ์ธ Monte Carlo๊ธฐ๋ฒ•์€ ํ™•๋ฅ ๋ณ€์ˆ˜๋“ค ๊ฐ„์˜ ๋…๋ฆฝ์„ฑ์„ ๊ฐ€์ •์œผ๋กœ ์ด๋ฃจ์–ด์ง€๋Š” ์ƒ˜ํ”Œ๋ง ๋ฐฉ๋ฒ•์ด๋ผ๋ฉด, MCMC๊ธฐ๋ฒ•์€ ๋‹ค๋ณ€๋Ÿ‰์— ๋Œ€ํ•ด์„œ ์ข…์†์„ฑ์„ ๊ธฐ์ค€์œผ๋กœ ์กฐ๊ฑด๋ถ€ ์ƒ˜ํ”Œ๋ง์ด ๊ฐ€๋Šฅํ•œ ๋ฐฉ๋ฒ•์ด๋ผ ํ•  ์ˆ˜ ์žˆ๋‹ค. ์ฆ‰, ์ฃผ์–ด์ง„ ๋‹ค๋ณ€๋Ÿ‰ ํ™•๋ฅ ๋ถ„ํฌ๊ฐ€ ๋ณต์žกํ•˜๊ณ  iid (independent identically distributed) ๋‚œ์ˆ˜ ๋Œ€์‹  Markov Chain์— ๊ทผ๊ฑฐํ•œ ๋‚œ์ˆ˜๋ฅผ ์ถ”์ถœํ•˜๋Š”๋ฐ ์ ํ•ฉํ•˜๋‹ค. Markov Chain์„ ํ†ตํ•ด ๋‹ค๋ณ€๋Ÿ‰ ํ™•๋ฅ ๋ถ„ํฌ๋ฅผ ๋”ฐ๋ฅด๋Š” ๋‚œ์ˆ˜๋ฅผ ์žฌํ˜„ํ•˜๊ธฐ ์œ„ํ•ด์„œ๋Š” ๋ฐ˜๋ณต ์‹œํ–‰์„ ํ†ตํ•ด ๋ถ„ํฌ์— ์ˆ˜๋ ด์‹œํ‚ค๋Š” ๊ณผ์ •์ด ํ•„์š”ํ•˜๋‹ค. ๋”ฐ๋ผ์„œ MCMC๊ธฐ๋ฒ•์€ Bayesian ํ†ต๊ณ„ ๊ธฐ๋ฒ•์—์„œ ์‚ฌํ›„๋ถ„ํฌ์˜ ์ถ”๋ก ์„ ์œ„ํ•ด์„œ ํ™œ์šฉ๋  ์ˆ˜ ์žˆ์œผ๋ฉฐ ๋งค๊ฐœ๋ณ€์ˆ˜๋“ค ๊ฐ„์˜ ๊ฒฐํ•ฉ ํ™•๋ฅ  ํŠน์„ฑ์„ ๊ณ ๋ คํ•œ ํ•ด์„์„ ๊ฐ€๋Šฅํ•˜๊ฒŒ ํ•œ๋‹ค.

MCMC๊ธฐ๋ฒ•์˜ ๋Œ€ํ‘œ์ ์ธ ๋ฐฉ๋ฒ•์œผ๋กœ ๋ฉ”ํŠธ๋กœํด๋ฆฌ์Šค-ํ—ค์ŠคํŒ… ์•Œ๊ณ ๋ฆฌ์ฆ˜(Metropolis-Hastings algorithm)๊ณผ ๊น์Šคํ‘œ๋ณธ๋ฒ• ๋“ฑ์ด ์žˆ์œผ๋ฉฐ, ๋ณธ ์—ฐ๊ตฌ์—์„œ๋Š” ๊น์Šคํ‘œ๋ณธ๋ฒ•์„ ์ด์šฉํ•˜์˜€๋‹ค. ๊น์Šคํ‘œ๋ณธ๋ฒ•์€ ์›ํ•˜๋Š” ๋‹ค๋ณ€๋Ÿ‰ ํ™•๋ฅ ๋ถ„ํฌ์—์„œ iid ํ‘œ๋ณธ์„ ์ถ”์ถœํ•˜๋Š” ๊ฒƒ์ด ๋ณต์žกํ•  ๋•Œ ์‚ฌ์šฉํ•˜๋Š” ๋ฐฉ๋ฒ•์œผ๋กœ์จ 2๊ฐœ์˜ ๋ณ€์ˆ˜๋ฅผ ๊ฐ–๋Š” ๋‹ค๋ณ€๋Ÿ‰ ํ™•๋ฅ ๋ถ„ํฌ๋ฅผ ์ด์šฉํ•˜์—ฌ ์„ค๋ช…ํ•˜๋ฉด ๋‹ค์Œ๊ณผ ๊ฐ™๋‹ค. 2๊ฐœ์˜ ๋ณ€์ˆ˜๋ฅผ ๊ฐ–๋Š” ๋‹ค๋ณ€๋Ÿ‰ ํ™•๋ฅ ๋ฐ€๋„ํ•จ์ˆ˜๋ฅผ PICA4BF.gif๋ผ๊ณ  ์ •์˜ํ•œ๋‹ค. ๊น์Šคํ‘œ๋ณธ๋ฒ•์€ ํ™•๋ฅ ๋ฐ€๋„ํ•จ์ˆ˜๋กœ๋ถ€ํ„ฐ ์ง์ ‘ ํ‘œ๋ณธ์„ ์ถ”์ถœํ•  ์ˆ˜๋Š” ์—†์œผ๋‚˜ ๊ฐ๊ฐ์˜ ๋ณ€์ˆ˜๋“ค์˜ ๋Œ€ํ•ด์„œ ๋‹ค๋ฅธ ๋‘ ๋ณ€์ˆ˜๋“ค์ด ์ฃผ์–ด์กŒ์„ ๋•Œ์˜ ์กฐ๊ฑด๋ถ€ ๋ถ„ํฌ๊ฐ€ ์•Œ๋ ค์ ธ ์žˆ๊ณ  ์ด๋กœ๋ถ€ํ„ฐ์˜ ํ‘œ๋ณธ์ถ”์ถœ์ด ๊ฐ€๋Šฅํ•œ ๊ฒฝ์šฐ์— ์‚ฌ์šฉํ•  ์ˆ˜ ์žˆ๋‹ค(Kwon et al., 2012). ์•Œ๊ณ ๋ฆฌ์ฆ˜์„ ๊ฐ„๋‹จํžˆ ์ •๋ฆฌํ•˜๋ฉด ๋‹ค์Œ๊ณผ ๊ฐ™๋‹ค.

[1] Gumbel ๋ถ„ํฌ์˜ ๋‘ ๋งค๊ฐœ๋ณ€์ˆ˜์— ๋Œ€ํ•œ ์ดˆ๊ธฐ ๊ฐ’(PICA4CF.gif)์„ ๋ถ€์—ฌํ•œ๋‹ค.

[2] PICA4D0.gif๋ฒˆ์งธ ๋‚œ์ˆ˜ ๋ฒกํ„ฐ (PICA4E1.gif)๊ฐ€ ์ฃผ์–ด์กŒ์„ ๋•Œ PICA4E2.gif๋ฒˆ์งธ ๋‚œ์ˆ˜ ๋ฒกํ„ฐ๋ฅผ ๋‹ค์Œ๊ณผ ๊ฐ™์€ ์กฐ๊ฑด๋ถ€ ๋ถ„ํฌ์—์„œ ์ถ”์ถœํ•œ๋‹ค.

(1) PICA4E3.gif

(2) PICA4F4.gif

[3] ์œ„์˜ ๊ณผ์ •์„ ์ถฉ๋ถ„ํžˆ ๋ฐ˜๋ณตํ•œ ํ›„ ์ดˆ๊ธฐ์˜ ์ผ์ •๋ถ€๋ถ„ ๋‚œ์ˆ˜๋ฅผ ์ œ๊ฑฐํ•œ ์ดํ›„์˜ ๋‚œ์ˆ˜๋“ค์„ ์ด์šฉํ•œ๋‹ค. ์ด๋Ÿฌํ•œ ์ œ๊ฑฐ๊ณผ์ •์„ Burning์ด๋ผ๊ณ  ํ•˜๋ฉฐ Bayesian ํ•ด์„์—์„œ ์ผ๋ฐ˜์ ์œผ๋กœ ์š”๊ตฌ๋˜๋Š” ๋‹จ๊ณ„์ด๋‹ค(George and McCulloch, 1993).

์œ„์˜ ๋ฐฉ๋ฒ•์—์„œ ๊น์Šคํ‘œ๋ณธ๋ฒ•์€ ํ˜„์žฌ์‹œ์ ์˜ ๋งค๊ฐœ๋ณ€์ˆ˜๋“ค์˜ ๊ฐ’์€ ์ •ํ™•ํ•˜๊ฒŒ ๋ฐ”๋กœ ์ง์ „์˜ ์ถ”์ถœ๋œ ๊ฐ’๋“ค์ด ์‚ฌ์šฉ๋˜๊ฒŒ ๋˜๋ฉฐ, ๋”ฐ๋ผ์„œ ์กฐ๊ฑด๋ถ€ ๋ถ„ํฌ์—์„œ ์ถ”์ถœ๋œ ๋‚œ์ˆ˜๋“ค์ด ์•ˆ์ • ์ƒํƒœ์— ๋„๋‹ฌํ•˜๋Š” ๊ฒƒ์ด ์ฃผ์–ด์ง„ ๋‹ค๋ณ€๋Ÿ‰ ํ™•๋ฅ ๋ถ„ํฌ๋ฅผ ์ •ํ™•ํžˆ ๋”ฐ๋ฅด๋Š” ๋‚œ์ˆ˜๊ฐ€ ๋˜๋Š” ์ฒ™๋„๊ฐ€ ๋˜๋ฉฐ ๊น์Šคํ‘œ๋ณธ๋ฒ•์„ ๊ตฌํ˜„ํ•˜๋Š”๋ฐ ๊ฐ€์žฅ ์ค‘์š”ํ•œ ๋ถ€๋ถ„์ด ๋œ๋‹ค(Kwon et al., 2012)

3.3 ๋ชจ์˜์‹คํ—˜

Bayesian๊ธฐ๋ฒ•๊ณผ ์—ฐ๊ณ„ํ•œ ํ˜ผํ•ฉ Gumbel ๋ถ„ํฌ์˜ ๋งค๊ฐœ๋ณ€์ˆ˜๋ฅผ ์‹ค์ œ ์ž๋ฃŒ์— ์ ์šฉํ•˜๊ธฐ ์ „์— ๋ชจ์˜์‹คํ—˜์œผ๋กœ ๊ฒ€์ฆํ•˜๋Š” ๊ณผ์ •์ด ํ•„์š”ํ•˜๋‹ค. ๋ณธ ์—ฐ๊ตฌ์—์„œ๋Š” ๊ฐ•์šฐ์ž๋ฃŒ์˜ ๋ฐœ์ƒ ์›์ธ์„ ๋‘ ๊ฐ€์ง€๋กœ ํŒ๋‹จํ•˜์—ฌ ๋‘ ๊ฐœ์˜ ๋ชจ์ง‘๋‹จ์œผ๋กœ ์„ค์ •ํ•˜์˜€๋‹ค. ๋‘ ๋ชจ์ง‘๋‹จ์œผ๋กœ ํ˜ผํ•ฉ๋ถ„ํฌํ˜•์„ ํ‘œํ˜„ํ•˜๊ธฐ ์œ„ํ•ด์„œ๋Š” ๋‘ ๊ฐœ์˜ ์œ„์น˜๋งค๊ฐœ๋ณ€์ˆ˜, ๋‘ ๊ฐœ์˜ ๊ทœ๋ชจ๋งค๊ฐœ๋ณ€์ˆ˜, ํ•œ ๊ฐœ์˜ ํ˜ผํ•ฉ๋น„(mixing ratio) ๋“ฑ ์ด ๋‹ค์„ฏ ๊ฐœ์˜ ๋งค๊ฐœ๋ณ€์ˆ˜๊ฐ€ ํ•„์š”ํ•˜๋‹ค. ๋ชจ์˜์‹คํ—˜ ์ˆœ์„œ๋Š” ๋‹ค์Œ๊ณผ ๊ฐ™๋‹ค.

[1] ๊ธฐ์ง€์˜ ๋งค๊ฐœ๋ณ€์ˆ˜(PICA4F5.gif)๋ฅผ ์„ค์ •ํ•˜๊ณ  ๋ฐœ์ƒ๋น„์œจ์— ๋งž๊ฒŒ ๊ฐ ๋ชจ์ง‘๋‹จ๋ณ„๋กœ ์ž๋ฃŒ๋ฅผ ๋ชจ์˜๋ฐœ์ƒ์‹œ์ผœ ํ•˜๋‚˜์˜ ๊ฐ•์šฐ์ž๋ฃŒ๋กœ ํ•ฉ์„ฑํ•œ๋‹ค.

[2] ํ•ฉ์„ฑํ•œ ๊ฐ•์šฐ์ž๋ฃŒ๋ฅผ ์•ž์„œ ์ œ์•ˆํ•œ ๋ฐฉ๋ฒ•๋ก ์„ ํ†ตํ•ด ๋‘ ๊ฐœ์˜ ๋ชจ์ง‘๋‹จ์œผ๋กœ ๋ถ„๋ฅ˜ํ•˜๊ณ  ๋งค๊ฐœ๋ณ€์ˆ˜๋ฅผ ์ถ”์ •ํ•œ๋‹ค.

[3] ์ถ”์ •ํ•œ ๋งค๊ฐœ๋ณ€์ˆ˜๋กœ ๋ชจ์˜๋ฐœ์ƒ์‹œ ๊ธฐ์กด ๋ชจ์˜๋ฐœ์ƒ์ž๋ฃŒ์™€์˜ ์ƒ๊ด€์„ฑ ๋ฐ ๊ฒฝํ–ฅ์„ฑ์„ ํ†ตํ•˜์—ฌ ๋ฐฉ๋ฒ•๋ก ์„ ๊ฒ€์ฆํ•œ๋‹ค.

๋‘ ๊ฐœ ๋ชจ์ง‘๋‹จ์˜ ํ˜ผํ•ฉ์ •๋„๋ฅผ ๋‚˜ํƒ€๋‚ด๋Š” ๋งค๊ฐœ๋ณ€์ˆ˜ PICA515.gif๋Š” ์ „์ฒด์ž๋ฃŒ์—์„œ์˜ ๋ชจ์ง‘๋‹จ์˜ ๋น„์œจ์„ ์˜๋ฏธํ•˜๋ฉฐ ๋‹ค๋ฅธ ๋งค๊ฐœ๋ณ€์ˆ˜์™€ ๋‹ค๋ฅด๊ฒŒ ํ•˜๋‚˜์˜ ๊ฐ’์œผ๋กœ ๋‘ ๋ชจ์ง‘๋‹จ์˜ ๋น„์œจ์ด ์ •ํ•ด์ง„๋‹ค. ํ˜ผํ•ฉ๋น„๋ฅผ ์„ธ ๊ฐœ์˜ ๊ฒฝ์šฐ๋กœ ๋ถ„๋ฅ˜ํ•œ ๋ชจ์˜์‹คํ—˜ ๊ฒฐ๊ณผ๋Š” ๋‹ค์Œ๊ณผ ๊ฐ™๋‹ค. CaseI(PICA516.gif=0.5), CaseII(PICA527.gif=0.3), CaseIII(PICA528.gif=0.7)์ด๋‹ค.

์œ„์˜ ์ˆœ์„œ์— ๋”ฐ๋ฅธ Case๋ณ„ ๋ชจ์˜์‹คํ—˜๊ฒฐ๊ณผ๋Š” Table 1๊ณผ Fig. 2์— ์ œ์‹œํ•˜์˜€์œผ๋ฉฐ, ๋†’์€ ์ƒ๊ด€์„ฑ ๋ฐ ๊ฒฝํ–ฅ์„ฑ์„ ๋ณด์—ฌ ์ œ์•ˆํ•œ ๋ฐฉ๋ฒ•๋ก ์„ ์‹ค์ œ์ž๋ฃŒ์— ์ ์šฉํ•  ์ˆ˜ ์žˆ๋‹ค๊ณ  ํŒ๋‹จํ•˜์˜€๋‹ค.

Table 1. Estimated Parameters and Their Credible Intervals of Synthetic AMRs for Three Different Mixing Ratios ๎€‘๎€ต Table_KSCE_38_2_07_T1.jpg
Fig. 2.

A graphical Representation of Synthetic Annual Maximum Series for Different Mixing Ratios

Figure_KSCE_38_2_07_F2.jpg

4. ๋ถ„ํฌํ˜•์— ๋”ฐ๋ฅธ ๋งค๊ฐœ๋ณ€์ˆ˜ ์ถ”์ • ๋ฐ ๋นˆ๋„ํ•ด์„ ํ‰๊ฐ€ ๊ฒฐ๊ณผ

ํ™•๋ฅ ๋ฐ€๋„ํ•จ์ˆ˜๋ฅผ ์ถ”์ •ํ•˜๋Š” ๋ฐฉ๋ฒ•์—๋Š” ๋งค๊ฐœ๋ณ€์ˆ˜์ (parametric)๋ฐฉ๋ฒ•๊ณผ ๋น„๋งค๊ฐœ๋ณ€์ˆ˜์ (non-parametric)๋ฐฉ๋ฒ•์œผ๋กœ ๊ตฌ๋ถ„๋œ๋‹ค. ๋งค๊ฐœ๋ณ€์ˆ˜์  ๋ฐฉ๋ฒ•์€ ์‚ฌ์ „์— ํŠน์ • ํ™•๋ฅ ๋ฐ€๋„ํ•จ์ˆ˜์— ๋Œ€ํ•œ ๋ชจ๋ธ์„ ์ •ํ•ด๋†“๊ณ  ์ž๋ฃŒ๋“ค๋กœ๋ถ€ํ„ฐ ๋ชจ๋ธ์˜ ๋งค๊ฐœ๋ณ€์ˆ˜๋งŒ ์ถ”์ •ํ•˜๋Š” ๋ฐฉ์‹์ด๋‹ค. ๋ณธ ์—ฐ๊ตฌ์—์„œ ์ œ์•ˆํ•˜๊ณ  ์žˆ๋Š” ํ˜ผํ•ฉํ™•๋ฅ ๋ถ„ํฌํ•จ์ˆ˜๊ฐ€ ์ด์— ํ•ด๋‹นํ•œ๋‹ค. ๊ทธ๋Ÿฌ๋‚˜ ํ˜„์‹ค์ ์œผ๋กœ ํŠน์ • ํ™•๋ฅ ๋ถ„ํฌ๋ฅผ ์‚ฌ์ „์— ์ธ์ง€ํ•˜๋Š” ๊ฒƒ์€ ์‰ฝ์ง€ ์•Š๋‹ค. ๋”ฐ๋ผ์„œ ์‚ฌ์ „ ์ •๋ณด๋‚˜ ์ง€์‹ ์—†์ด ๊ด€์ธก๋œ ์ž๋ฃŒ๋งŒ์œผ๋กœ ํ™•๋ฅ ๋ฐ€๋„ํ•จ์ˆ˜๋ฅผ ์ถ”์ •ํ•˜๋Š” ๋น„๋งค๊ฐœ๋ณ€์ˆ˜์  ํ•ต๋ฐ€๋„์ถ”์ •(Kernel density estimation, KDE) ๋ฐฉ๋ฒ•์„ ์‹œ๊ฐ์ ์ธ ๊ฒ€ํ†  ์ธก๋ฉด์—์„œ ๋Œ€์กฐ๊ตฐ์œผ๋กœ ์‚ฌ์šฉํ•˜์˜€๋‹ค. ์ฆ‰, ๋งค๊ฐœ๋ณ€์ˆ˜์  ๋ฐฉ๋ฒ•์œผ๋กœ ์–ป์€ ํ™•๋ฅ ๋ฐ€๋„ํ•จ์ˆ˜์™€ KDE๋ฅผ ํ†ตํ•ด ์–ป์€ ํ™•๋ฅ ๋ฐ€๋„ํ•จ์ˆ˜๊ฐ€ ์ผ์น˜ํ•˜๋Š” ์ •๋„๋ฅผ ๊ฐ€์ง€๊ณ  ๊ฐ„์ ‘์ ์œผ๋กœ ๋ถ„ํฌ์ถ”์ •์˜ ์ •๋„๋ฅผ ํŒ๋‹จํ•˜์˜€๋‹ค. ์ „์ฃผ์ง€์ ์˜ ์—ฐ์ตœ๋Œ€๊ฐ•์šฐ๋Ÿ‰์— ๋Œ€ํ•ด ์ง€์†์‹œ๊ฐ„๋ณ„๋กœ ์ ์šฉํ•œ ๊ฒฐ๊ณผ๋Š” ๋‹ค์Œ๊ณผ ๊ฐ™๋‹ค.

์šฐ์„  ์ง€์†์‹œ๊ฐ„๋ณ„๋กœ ์ถ”์ •๋œ ๋งค๊ฐœ๋ณ€์ˆ˜๋ฅผ ๋น„๊ตํ•˜์—ฌ Table 2์— ์ •๋ฆฌํ•˜์—ฌ ๋‚˜ํƒ€๋‚ด์—ˆ๋‹ค. ๋ชจ๋“  ์ง€์†์‹œ๊ฐ„์— ๋Œ€ํ•ด์„œ ์ฒซ ๋ฒˆ์งธ Gumbel ๋ถ„ํฌ์˜ ๊ทœ๋ชจ๋งค๊ฐœ๋ณ€์ˆ˜๊ฐ€ ๋‘ ๋ฒˆ์งธ Gumbel ๋ถ„ํฌ์— ๋น„ํ•ด ์ƒ๋Œ€์ ์œผ๋กœ ์ž‘๊ฒŒ ์ถ”์ •๋˜๊ณ  ์žˆ์œผ๋ฉฐ, ๋‹จ์ผ Gumbel ๋ถ„ํฌ์™€ ๋น„๊ตํ•ด๋ณด๋ฉด ๋‘ ๋ฒˆ์งธ Gumbel ๋ถ„ํฌ๊ฐ€ ์ „๋ฐ˜์ ์œผ๋กœ ์œ ์‚ฌ์„ฑ์ด ํฌ๋‹ค. ๊ทธ๋Ÿฌ๋‚˜ ์ง€์†์‹œ๊ฐ„์ด ์งง์€ 1-3์‹œ๊ฐ„์˜ ๊ฒฝ์šฐ์—๋Š” ๋‹จ์ผ Gumbel ๋ถ„ํฌ์™€ ์ฐจ์ด๊ฐ€ ํฌ๊ฒŒ ๋‚˜ํƒ€๋‚˜๊ณ  ์žˆ์œผ๋ฉฐ ์ด๋Š” ํ˜ผํ•ฉ๋น„ PICA538.gif์™€๋„ ์—ฐ๊ด€์„ฑ์ด ํฌ๋‹ค. ๋˜ํ•œ, ์ง€์†์‹œ๊ฐ„ 6์‹œ๊ฐ„์—์„œ์˜ ํ˜ผํ•ฉ๋น„์™€ ์ฒซ ๋ฒˆ์งธ Gumbel ๋ถ„ํฌ์˜ ์œ„์น˜, ๊ทœ๋ชจ๋งค๊ฐœ๋ณ€์ˆ˜๊ฐ€ ๊ธ‰๊ฐํ•˜๋Š” ๊ฒƒ์œผ๋กœ ๋‚˜ํƒ€๋‚˜๊ณ  ์žˆ๋‹ค. ์ด๋Š” ์ฒซ ๋ฒˆ์งธ Gumbel ๋ถ„ํฌ์— ์†ํ•˜๋Š” ์ž๋ฃŒ๋“ค์˜ ์ƒ๋Œ€์ ์ธ ํฌ๊ธฐ์™€ ๊ฐœ์ˆ˜๊ฐ€ ์ค„์–ด๋“ค๋ฉด์„œ ์ถ”์ •์˜ ๋ถˆํ™•์‹ค์„ฑ๋„ ์ค„์–ด๋“œ๋Š” ๊ฒƒ์œผ๋กœ ํŒ๋‹จํ•  ์ˆ˜ ์žˆ๋‹ค. ์ฆ‰, ์ง€์†์‹œ๊ฐ„์ด ์งง์€ 1~3์‹œ๊ฐ„์˜ ๊ฒฝ์šฐ ํ˜ผํ•ฉ๋น„ PICA539.gif๊ฐ€ 34% ์ด์ƒ์œผ๋กœ ์ถ”์ •๋˜์—ˆ์œผ๋ฉฐ, ์ด๋Š” 2๊ฐœ์˜ Gumbel ๋ถ„ํฌ๊ฐ€ ๊ฑฐ์˜ ๋Œ€๋“ฑํ•˜๊ฒŒ ๊ฐ€์ค‘์น˜๋ฅผ ๊ฐ–๋Š” ๊ฒƒ์œผ๋กœ ๋ณผ ์ˆ˜ ์žˆ๋‹ค. ์ด๋Ÿฌํ•œ ์ ์—์„œ ๋‹จ์ผ Gumbel ๋ถ„ํฌ์™€๋Š” ๋ถ„ํฌ์˜ ํ˜•ํƒœ ๋ฐ ๋งค๊ฐœ๋ณ€์ˆ˜ ์ธก๋ฉด์—์„œ ์ƒ์ด์„ฑ์ด ํฌ๋‹ค๊ณ  ํ•  ์ˆ˜ ์žˆ๋‹ค. ์ด๋Ÿฌํ•œ ํŠน์ง•์€ Fig. 3์—์„œ ๋‚˜ํƒ€๋‚ธ ํ™•๋ฅ ๋ฐ€๋„ํ•จ์ˆ˜์˜ ๋น„๊ต ๊ทธ๋ฆผ์—์„œ๋„ ํ™•์ธ๋  ์ˆ˜ ์žˆ๋‹ค. ์ฆ‰, ์ง€์†์‹œ๊ฐ„์ด ์งง์€ 1~3์‹œ๊ฐ„์˜ ๊ฒฝ์šฐ ๋‘ ๋ฒˆ์งธ ์ฒจ๋‘์˜ ์œ„์น˜๊ฐ€ ์ฒซ ๋ฒˆ์งธ ์ฒจ๋‘์™€ ์ƒ๋Œ€์ ์œผ๋กœ ๊ฐ€๊นŒ์šด ๋ฐ˜๋ฉด ์ง€์†์‹œ๊ฐ„์ด ๊ธด ๊ฒฝ์šฐ์—๋Š” ๋‘ ๋ฒˆ์งธ ์ฒจ๋‘๊ฐ€ Upper Tail ์ชฝ์— ์œ„์น˜ํ•˜๊ณ  ์žˆ๋Š” ์ ์„ ํ™•์ธํ•  ์ˆ˜ ์žˆ๋‹ค.

Fig. 3.

Comparison of Probability Density Functions between Univariate and Mixed Gumbel Distributions at Jeonju Station. All the Parameters for the Univariate and Mixed Gumbel Distribution were Estimated Within a Bayesian Framework

Figure_KSCE_38_2_07_F3.jpg
Table 2. Comparison of the Estimated Parameters between Univariate and Mixed Gumbel Distributions Table_KSCE_38_2_07_T2.jpg

์ง€์†์‹œ๊ฐ„๋ณ„๋กœ ๋‹จ์ผ Gumbel ๋ถ„ํฌ์™€ ํ˜ผํ•ฉ Gumbel ๋ถ„ํฌ์˜ ํŠน์„ฑ์„ ์ž์„ธํžˆ ์‚ดํŽด๋ณด๋ฉด ๋‹ค์Œ๊ณผ ๊ฐ™๋‹ค. ์ง€์†์‹œ๊ฐ„ 1, 6, 9, 12์‹œ๊ฐ„์— ๋Œ€ํ•œ ์ž๋ฃŒํŠน์„ฑ์ด ๋‘ ๊ฐœ์˜ ์ฒจ๋‘๊ฐ€ ๋šœ๋ ทํ•˜๊ณ , ์ง€์†์‹œ๊ฐ„ 2, 3, 18, 24์‹œ๊ฐ„์— ๋Œ€ํ•œ ์ž๋ฃŒํŠน์„ฑ์—์„œ๋Š” ์ „์ž์— ๋น„ํ•ด ๋šœ๋ ทํ•˜์ง€ ์•Š๋‹ค. ๋ชจ๋“  ์ง€์†์‹œ๊ฐ„์— ๋Œ€ํ•œ ์ž๋ฃŒํŠน์„ฑ๋“ค์ด ๊ผฌ๋ฆฌ๋ถ€๋ถ„์—์„œ์˜ ์ฒจ๋‘ ํ˜•ํƒœ๋ฅผ ๊ฐ€์ง€๊ณ , ๋‹จ์ผ Gumbel ๋ถ„ํฌ ํŠน์„ฑ์ƒ ๊ผฌ๋ฆฌ๋ถ€๋ถ„์˜ ํ™•๋ฅ ๊ฑฐ๋™์„ ํ‘œํ˜„ํ•˜๊ธฐ ํž˜๋“ค๋‹ค. ๊ทธ ๊ฒฐ๊ณผ ๋นˆ๋„ํ•ด์„์‹œ ํ˜ผํ•ฉ Gumbel์— ๋น„ํ•ด ๊ณผ์†Œ์ถ”์ •๋˜๋Š” ๊ฒฝํ–ฅ์„ ๋ณด์ธ๋‹ค. ์ค‘์•™๋ถ€๋ถ„ ์ฒจ๋‘, ๊ผฌ๋ฆฌ๋ถ€๋ถ„ ์ฒจ๋‘๋กœ ๊ตฌ๋ถ„ํ•˜์—ฌ ๋ถ„์„ํ•ด ๋ณผ ๋•Œ ์ง€์†์‹œ๊ฐ„ 6, 9, 12์‹œ๊ฐ„์— ๋Œ€ํ•œ ์ค‘์•™์ฒจ๋‘์˜ ํ™•๋ฅ ์ด ํ˜ผํ•ฉ Gumbel ๋ถ„ํฌํ˜•์—์„œ๋Š” ์‹ค์ œ์ž๋ฃŒ๋ณด๋‹ค ๋‹ค์†Œ ์ž‘๊ฒŒ ์ถ”์ •๋˜๊ณ , ๋‹จ์ผ Gumbel ๋ถ„ํฌํ˜•์—์„œ๋Š” ๋‹ค์†Œ ํฌ๊ฒŒ ์ถ”์ •๋˜์—ˆ๋‹ค. ์ด์™ธ์˜ ๋ชจ๋“  ์ง€์†์‹œ๊ฐ„์— ๋Œ€ํ•ด์„œ๋Š” ํ˜ผํ•ฉ Gumbel ๋ถ„ํฌํ˜•์ด ์ž๋ฃŒ์˜ ๋ถ„ํฌ ํŠน์„ฑ์„ ๋ณด๋‹ค ์‹ค์งˆ์ ์œผ๋กœ ๋ฌ˜์‚ฌํ•˜๊ณ  ์žˆ๋Š” ๊ฒƒ์œผ๋กœ ํ™•์ธํ•  ์ˆ˜ ์žˆ๋‹ค.

์ตœ์ ๋ถ„ํฌํ˜•์„ ์„ ์ •ํ•˜๋Š” ๊ธฐ์ค€์—๋Š” AIC (Akaike Information Criterion), BIC (Bayesian Information Criterion), DIC (Deviance Information Criterion) ๋“ฑ์ด ์žˆ๋‹ค. ๋ณธ ์—ฐ๊ตฌ์—์„œ๋Š” ์ž๋ฃŒ์˜ ๊ฐœ์ˆ˜PICA53A.gif, ๋งค๊ฐœ๋ณ€์ˆ˜์˜ ๊ฐœ์ˆ˜PICA54B.gif, ์šฐ๋„PICA54C.gif๋ฅผ ๊ณ ๋ คํ•˜์—ฌ ์šฐ์„ ์ˆœ์œ„๋ฅผ ์„ ์ •ํ•˜๋Š” BIC, ์ฆ‰ Eq. (6)๋ฅผ ์‚ฌ์šฉํ•˜์˜€๋‹ค. PICA54D.gif์€ ๋ชจ๋ธ์˜ ์ตœ์šฐ๋„๊ฐ’(the maxi-mized value of the liklehood function)์ด๋ฉฐ, Eq. (7)๋กœ ๋‚˜ํƒ€๋‚ด๋ฉฐ, PICA55D.gif๋Š” ์ตœ์šฐ๋„ํ•จ์ˆ˜์˜ ๋งค๊ฐœ๋ณ€์ˆ˜๋ฅผ ๋‚˜ํƒ€๋‚ธ๋‹ค.

PICA55E.gif (6)

PICA56F.gif (7)

์ ์€ ๋งค๊ฐœ๋ณ€์ˆ˜์™€ ํฐ ์šฐ๋„๊ฐ’์„ ํ† ๋Œ€๋กœ BIC๊ฐ’์ด ์ž‘๊ฒŒ ์‚ฐ์ •๋  ๋•Œ ํ†ต๊ณ„์ ์œผ๋กœ ์šฐ์ˆ˜ํ•œ ๋ถ„ํฌํ˜•์ด๋ผ ํŒ๋‹จํ•  ์ˆ˜ ์žˆ๋‹ค. ํŠนํžˆ ๊ณผ์ ํ•ฉ(over-fitting) ๋ฌธ์ œ์— ์žˆ์–ด์„œ ๋งค๊ฐœ๋ณ€์ˆ˜์˜ ๊ฐœ์ˆ˜์— ๋Œ€ํ•œ ๋ฒŒ์ (Penalty)์ด ํฌ๊ฒŒ ์ž‘์šฉํ•˜๋ฏ€๋กœ ํ˜ผํ•ฉ๋ถ„ํฌํ˜•์ด ์ „๋ฐ˜์ ์œผ๋กœ ๋ถˆ๋ฆฌํ•˜๋‹ค. ๊ทธ๋Ÿฌ๋‚˜ Table 3์— ์ •๋ฆฌํ•œ ๊ฒƒ์ฒ˜๋Ÿผ ํ˜ผํ•ฉ๋ถ„ํฌํ˜•์˜ ์šฐ๋„(likelihood)๊ฐ€ ๋‹จ์ผ๋ถ„ํฌํ˜•์— ๋น„ํ•ด ๊ฐœ์„ ๋˜๋Š” ํšจ๊ณผ๊ฐ€ ๋งค๊ฐœ๋ณ€์ˆ˜ ๊ฐœ์ˆ˜์˜ ์ฆ๊ฐ€์— ๋น„ํ•ด ํฌ๊ฒŒ ๋‚˜ํƒ€๋‚˜๊ณ  ์žˆ๋‹ค. ๋”ฐ๋ผ์„œ ๋ชจ๋“  ๊ฒฝ์šฐ์— ์žˆ์–ด์„œ ํ˜ผํ•ฉ Gumbel ๋ถ„ํฌ์˜ BIC๊ฐ’์ด ๋” ์ž‘๊ฒŒ ์‚ฐ์ •๋˜๊ณ  ์žˆ๋‹ค. ์ฆ‰, ํ†ต๊ณ„์  ์ธก๋ฉด์—์„œ ํ˜ผํ•ฉ Gumbel ๋ถ„ํฌํ˜•์ด ๋” ์ ํ•ฉํ•˜๋‹ค๊ณ  ํ‰๊ฐ€๋  ์ˆ˜ ์žˆ๋‹ค. ๋˜ํ•œ, ๋™์ผ ์ง€์†์‹œ๊ฐ„์—์„œ ๋‚ฎ์€ ์žฌํ˜„๊ธฐ๊ฐ„์ผ ๊ฒฝ์šฐ ๊ฐ•์šฐ๊ฐ•๋„๊ฐ€ ํ˜ผํ•ฉ๋ถ„ํฌ์™€ ๋‹จ์ผ๋ถ„ํฌ๊ฐ„์˜ ์ฐจ์ด๊ฐ€ ์ž‘์ง€๋งŒ, ๋†’์€ ์žฌํ˜„๊ธฐ๊ฐ„์œผ๋กœ ๊ฐˆ์ˆ˜๋ก ๊ทธ ์ฐจ์ด๊ฐ€ ์ปค์ง€๋Š” ๊ฒƒ์„ Table 3์˜ ์žฌํ˜„๊ธฐ๊ฐ„๋ณ„ ํ™•๋ฅ ๊ฐ•์ˆ˜๋Ÿ‰๊ณผ Fig. 4์˜ Intensity- Duration-Frequency (IDF) ๊ณก์„ ์—์„œ ํ™•์ธํ•  ์ˆ˜ ์žˆ๋‹ค.

Table 3. Estimations of Design Rainfalls of Univariate and Mixed Gumbel Distribution and Their Comparison Corresponding to Different Durations Table_KSCE_38_2_07_T3.jpg
Fig. 4.

IDF Curves of Univariate and Mixed Gumbel Distributions

Figure_KSCE_38_2_07_F4.jpg

5. ๊ฒฐ ๋ก 

์šฐ๋ฆฌ๋‚˜๋ผ์—์„œ๋Š” ๋นˆ๋„ํ•ด์„ ์‹œ ์ผ๋ฐ˜์ ์œผ๋กœ ํ™•๋ฅ ๊ฐ€์ค‘๋ชจ๋ฉ˜ํŠธ๋ฒ•์„ ์ ์šฉํ•œ Gumbel ๋ถ„ํฌํ˜•์„ ์‚ฌ์šฉํ•˜๊ณ  ์žˆ์ง€๋งŒ, ์—ฌ๋Ÿฌ ๋ฐœ์ƒ์š”์ธ์œผ๋กœ ์ด๋ฃจ์–ด์ง„ ์ž๋ฃŒ์˜ ๋ชจ์ง‘๋‹จ์ถ”์ •์€ ํ˜ผํ•ฉ๋ถ„ํฌํ˜•์ด ์ ํ•ฉํ•˜๋‹ค๋Š” ๋งŽ์€ ์—ฐ๊ตฌ์‚ฌ๋ก€๊ฐ€ ์žˆ๋‹ค. ๋ณธ ์—ฐ๊ตฌ์—์„œ๋Š” ๋ถˆํ™•์‹ค์„ฑ์— ๋Œ€ํ•œ ์‹ ๋ขฐ์„ฑ์„ ํ™•๋ณดํ•˜๋Š”๋ฐ ์šฐ์ˆ˜ํ•œ ์„ฑ๋Šฅ์„ ๋ฐœํœ˜ํ•˜๋Š” Bayesian ๊ธฐ๋ฒ•๊ณผ ์—ฐ๊ณ„ํ•œ ๋นˆ๋„ํ•ด์„ ๊ธฐ๋ฒ•์„ ๊ฐœ๋ฐœํ•˜์˜€๋‹ค. ๋ณธ ์—ฐ๊ตฌ์—์„œ๋Š” ์ „์ฃผ์ง€์—ญ ๊ฐ•์šฐ์ž๋ฃŒ์— ๋Œ€ํ•ด์„œ ๋ชจํ˜•์˜ ์ ํ•ฉ์„ฑ์„ ํ‰๊ฐ€ํ•˜์˜€์œผ๋ฉฐ, BIC๋ฅผ ํ†ตํ•ด ๊ธฐ์กด ๋ฐฉ๋ฒ•๋ก ๊ณผ ์ •๋Ÿ‰์ ์ธ ๋น„๊ต ๊ฒ€ํ† ๋ฅผ ์ˆ˜ํ–‰ํ•˜์˜€๋‹ค. ๋ณธ ์—ฐ๊ตฌ๋ฅผ ํ†ตํ•˜์—ฌ ๋„์ถœ๋œ ๊ฒฐ๋ก ์€ ๋‹ค์Œ๊ณผ ๊ฐ™๋‹ค.

(1)์ผ๋ฐ˜์ ์ธ ๊ฐ•์šฐ๋นˆ๋„ํ•ด์„๋ฒ•์€ ์—ฐ์ตœ๋Œ€์น˜๊ฐ•์šฐ๊ฐ€ ๋‹จ์ผ ๋ชจ์ง‘๋‹จ์„ ์ด๋ฃฌ๋‹ค๊ณ  ๊ฐ€์ •ํ•˜์ง€๋งŒ ์ „์ฃผ์ง€์—ญ์˜ ๊ฐ•์šฐ์ž๋ฃŒ๋ฅผ ํ‰๊ฐ€ํ•ด ๋ณด๋ฉด ํ•˜๋‚˜์˜ ํŠน์„ฑ์„ ๊ฐ€์ง„ ๋ชจ์ง‘๋‹จ์ด๋ผ๊ณ  ๋ณด๊ธฐ ์–ด๋ ต๋‹ค. ์ฆ‰, ๋ฐœ์ƒ ๋น„์œจ๋ฉด์—์„œ ์ฐจ์ด๊ฐ€ ์žˆ์ง€๋งŒ, ์ตœ์†Œ 2๊ฐœ ์ด์ƒ์˜ ํŠน์„ฑ์„ ๊ฐ€์ง„ ๋ชจ์ง‘๋‹จ์œผ๋กœ ๊ตฌ๋ถ„๋˜๊ณ  ์žˆ๋‹ค. ์ด๋Ÿฌํ•œ ์ž๋ฃŒ๋ฅผ ํ†ต๊ณ„์ ์œผ๋กœ ํ•˜๋‚˜์˜ ๋ถ„ํฌํ˜•์œผ๋กœ๋งŒ ์ ํ•ฉ์‹œํ‚ค๋Š”๋ฐ ํ•œ๊ณ„๊ฐ€ ์žˆ์œผ๋ฉฐ ๋ณธ ์—ฐ๊ตฌ์—์„œ ์ œ์‹œํ•œ ํ˜ผํ•ฉ Gumbel ๋ถ„ํฌํ˜•์€ ์‹ ๋ขฐํ•  ๋งŒํ•œ ์ถ”์ •์น˜๋ฅผ ์ œ๊ณตํ•ด์ฃผ๊ณ  ์žˆ๋‹ค.

(2)ํ˜ผํ•ฉ Gumbel ํ˜ผํ•ฉ๋ถ„ํฌํ˜•์ด ๋‹จ์ผ Gumbel ๋ถ„ํฌํ˜•์— ๋น„ํ•ด ๋งค๊ฐœ๋ณ€์ˆ˜๊ฐ€ ๋งŽ์•„ BIC๋ฅผ ๋น„๊ตํ•  ๋•Œ ๋ถˆ๋ฆฌํ•˜๊ฒŒ ์ž‘์šฉํ•œ๋‹ค. ๊ทธ๋Ÿฌ๋‚˜ ํ˜ผํ•ฉ๋ถ„ํฌ๋ฅผ ํ†ตํ•œ ์ž๋ฃŒ์˜ ์ ํ•ฉ ํŠน์„ฑ์„ ๋‚˜ํƒ€๋‚ด๋Š” ์šฐ๋„๊ฐ’์˜ ์ฆ๊ฐ€๊ฐ€ ๋งค๊ฐœ๋ณ€์ˆ˜ ์ฆ๊ฐ€๋กœ ์ธํ•œ ๋ฒŒ์ (Penalty)๋ณด๋‹ค ์ƒ๋Œ€์ ์ธ ์šฐ์œ„๋ฅผ ๋ณด์ด๊ณ  ์žˆ์–ด ๋ชจ๋“  ๊ฒฝ์šฐ์— ์žˆ์–ด ํ˜ผํ•ฉ Gumbel ๋ถ„ํฌํ˜•์˜ ์ ํ•ฉ์„ฑ์ด ์šฐ์ˆ˜ํ•œ ๊ฒƒ์œผ๋กœ ํ‰๊ฐ€๋˜์—ˆ๋‹ค. ์ฆ‰, ๋‹จ์ผ Gumbel ๋ถ„ํฌํ˜•๋ณด๋‹ค ๊ทน์น˜๊ฐ•์šฐ๋Ÿ‰ ์ž๋ฃŒ๋ฅผ ๋‚˜ํƒ€๋‚ด๋Š”๋ฐ ์ ํ•ฉํ•˜๋‹ค๊ณ  ํŒ๋‹จํ•  ์ˆ˜ ์žˆ๋‹ค.

(3)๋ชจ๋“  ์ง€์†์‹œ๊ฐ„์— ๋Œ€ํ•œ ๊ทน์น˜๋นˆ๋„ํ•ด์„์—์„œ ํ˜ผํ•ฉ๋ถ„ํฌ๊ฐ€ ๋‹จ์ผ๋ถ„ํฌ๋ณด๋‹ค ํฐ ๊ฐ’์„ ๊ฐ€์ง„๋‹ค. ์ด๋Š” ํ˜„์žฌ์˜ ์„ค๊ณ„๊ธฐ์ค€์— ์˜ํ•œ ์„ค๊ณ„ ์‹œ ๊ทน์น˜ ํ™•๋ฅ ๊ฐ•์ˆ˜๋Ÿ‰์˜ ๊ณผ์†Œ์ถ”์ •์œผ๋กœ ์ด์–ด์ ธ ์ˆ˜๊ณต๊ตฌ์กฐ๋ฌผ์˜ ์•ˆ์ „๋„ ์ธก๋ฉด์—์„œ ๋ถˆ๋ฆฌํ•  ๊ฒƒ์œผ๋กœ ํŒ๋‹จ๋œ๋‹ค. ์ œ์•ˆ๋œ ํ˜ผํ•ฉ Gumbel ๋ถ„ํฌํ˜•์€ ๋‹จ์ผ Gumbel ๋ถ„ํฌํ˜•์ด ๋ฐ˜์˜ํ•˜์ง€ ๋ชปํ•˜๋Š” ๊ผฌ๋ฆฌ๋ถ€๋ถ„์˜ ์‹ ๋ขฐ์„ฑ ์žˆ๋Š” ์ถ”์ •์ด ๊ฐ€๋Šฅํ•˜์˜€์œผ๋ฉฐ, ์ด๋Ÿฌํ•œ ์ ์—์„œ ์„ค๊ณ„๊ฐ•์ˆ˜๋Ÿ‰ ์ถ”์ • ์‹œ ๋ณด๋‹ค ์‹ ๋ขฐ์„ฑ ์žˆ๋Š” ์ ‘๊ทผ์ด ๊ฐ€๋Šฅํ•  ๊ฒƒ์œผ๋กœ ํŒ๋‹จ๋œ๋‹ค.

๋ณธ ์—ฐ๊ตฌ์— ์ ์šฉ๋œ Bayesian๊ธฐ๋ฒ•๊ณผ ์—ฐ๊ณ„ํ•œ ํ˜ผํ•ฉ๋ถ„ํฌ ๋งค๊ฐœ๋ณ€์ˆ˜ ์ถ”์ • ๋ฐ ๋นˆ๋„ํ•ด์„ ๊ธฐ๋ฒ•์€ ๋‹จ์ผ ๋ถ„ํฌํ˜•์— ๋ฐ˜์˜๋˜์ง€ ๋ชปํ•˜๋Š” ๊ผฌ๋ฆฌ(tail)๋ถ€๋ถ„์˜ ์ด์ค‘์ฒจ๋‘ ๊ฑฐ๋™์„ ํšจ๊ณผ์ ์œผ๋กœ ๋ชจ์˜ํ•  ์ˆ˜ ์žˆ๋Š” ๋ฐฉ๋ฒ•์œผ๋กœ์„œ ๋ณธ ์—ฐ๊ตฌ์—์„œ๋Š” ์žฅ๋งˆ์™€ ํƒœํ’์œผ๋กœ ์ธํ•œ ์šฐ๋ฆฌ๋‚˜๋ผ ์—ฌ๋ฆ„์ฒ  ๊ฐ•์šฐ์˜ ๊ฐ•์šฐ์ง€์†์‹œ๊ฐ„ ๋ฐ ๊ฐ•์šฐ๊ฐ•๋„์˜ ๋ณ€ํ™”์— ๋”ฐ๋ฅธ ์ด์ค‘์ฒจ๋‘ ํŠน์ง•์„ ๊ณ ๋ คํ•œ ๋นˆ๋„ํ•ด์„์‹œ ์‹ ๋ขฐ์„ฑ์žˆ๋Š” ๊ฒฐ๊ณผ๋ฅผ ๋„์ถœํ•˜๋Š”๋ฐ ํ™œ์šฉํ•˜์˜€๋‹ค. ๊ฒฐ๊ณผ์ ์œผ๋กœ ๊ธฐ์กด์˜ ๋‹จ์ผ๋ถ„ํฌ๋ฅผ ํ†ตํ•œ ์„ค๊ณ„๊ฐ•์ˆ˜๋Ÿ‰ ์ถ”์ •์‹œ ๋ณด๋‹ค ์‹ ๋ขฐ์„ฑ์žˆ๋Š” ์ ‘๊ทผ์ด ๊ฐ€๋Šฅํ•˜๊ฒŒ ๋์œผ๋ฉฐ, ์šฐ๋ฆฌ๋‚˜๋ผ ๊ทน์น˜๊ฐ•์šฐ์ž๋ฃŒ ๋ถ„์„์‹œ ๊ธฐ์กด ๋‹จ์ผ๋ถ„ํฌ๊ธฐ๋ฐ˜์˜ ๋นˆ๋„ํ•ด์„๊ธฐ๋ฒ• ๋Œ€์•ˆ์œผ๋กœ ์ ์šฉ์ด ๊ฐ€๋Šฅํ•  ๊ฒƒ์œผ๋กœ ํŒ๋‹จ๋œ๋‹ค. ์ด๋Ÿฌํ•œ ์ ์—์„œ ํ–ฅํ›„ ์—ฐ๊ตฌ๋กœ์จ ์„œ๋กœ ๋‹ค๋ฅธ ๋ถ„ํฌํ˜•์„ ๋ชจํ˜•์— ์ ์šฉํ•˜์—ฌ ์ตœ์ ํ™”๋œ ๋ชจํ˜•์„ ๊ตฌ์ถ•ํ•˜๊ณ  ๊ฐ•์šฐ์˜ ์‹œ๊ฐ„์— ๋”ฐ๋ฅธ ๋ณ€๋™์„ฑ์„ ๊ณ ๋ คํ•  ์ˆ˜ ์žˆ๋Š” ๋น„์ •์ƒ์„ฑ ๋นˆ๋„๋ถ„์„์„ ์ˆ˜ํ–‰ํ•˜์—ฌ ๊ณผ๊ฑฐ๋ถ€ํ„ฐ ํ˜„์žฌ์— ์ด๋ฅด๋Š” ์—ฌ๋ฆ„์ฒ  ์žฅ๋งˆ ๋ฐ ํƒœํ’์— ๋”ฐ๋ฅธ ๊ฐ•์šฐ๋ณ€๋™์„ฑ์„ ๊ณ ๋ คํ•œ ๋ณด๋‹ค ์‹ ๋ขฐ์„ฑ์žˆ๋Š” ์„ค๊ณ„๊ฐ•์šฐ๋Ÿ‰ ์ถ”์ •์— ํ™œ์šฉํ•  ์ˆ˜ ์žˆ์„ ๊ฒƒ์œผ๋กœ ํŒ๋‹จ๋œ๋‹ค.

References

1 
Gelman, A., Chew, G. L. and Shnaidman, M. (2004). โ€œBayesian analysis of serial dilution assays.โ€ Biometrics, Vol.  60, No. 2, pp. 407-417.10.1111/j.0006-341X.2004.00185.xGelman, A., Chew, G. L. and Shnaidman, M. (2004). โ€œBayesian analysis of serial dilution assays.โ€ Biometrics, Vol.  60, No. 2, pp. 407-417.DOI
2 
George, E. I. and McCulloch, R. E. (1993). โ€œVariable selection via Gibbs sampling.โ€ Journal of the American Statistical Association, Vol. 88, No. 423, pp. 881-889.10.1080/01621459.1993.10476353George, E. I. and McCulloch, R. E. (1993). โ€œVariable selection via Gibbs sampling.โ€ Journal of the American Statistical Association, Vol. 88, No. 423, pp. 881-889.DOI
3 
Heo, J. H., Kim, G. D. and Han, J. H. (1999). โ€œDerivation of rainfall Intensity-Duration-Frequency equation based on the approproate probability distribution.โ€ Journal of Korea Water Resources Association, Vol. 32, No. 3, pp. 247-254 (in Korean).Heo, J. H., Kim, G. D. and Han, J. H. (1999). โ€œDerivation of rainfall Intensity-Duration-Frequency equation based on the approproate probability distribution.โ€ Journal of Korea Water Resources Association, Vol. 32, No. 3, pp. 247-254 (in Korean).Google Search
4 
Ho, C. H., Lee, J. Y., Ahn, M. H. and Lee, H. S. (2003). โ€œA sudden change in summer rainfall characteristics in Korea during the late 1970s.โ€ International Journal of Climatology, Vol. 23, No. 1, pp. 117-128.10.1002/joc.864Ho, C. H., Lee, J. Y., Ahn, M. H. and Lee, H. S. (2003). โ€œA sudden change in summer rainfall characteristics in Korea during the late 1970s.โ€ International Journal of Climatology, Vol. 23, No. 1, pp. 117-128.DOI
5 
Kwon, H. H. and Moon, Y. I. (2006). โ€œImprovement of overtopping risk evaluations using probabilistic concepts for existing dams.โ€ Stochastic Environmental Research and Risk Assessment, Vol. 20, No. 4, p. 223.10.1007/s00477-005-0017-2Kwon, H. H. and Moon, Y. I. (2006). โ€œImprovement of overtopping risk evaluations using probabilistic concepts for existing dams.โ€ Stochastic Environmental Research and Risk Assessment, Vol. 20, No. 4, p. 223.DOI
6 
Kwon, H. H., Brown, C. and Lall, U. (2008a). โ€œClimate informed flood frequency analysis and prediction in Montana using hierarch-ical Bayesian modeling.โ€ Geophysical Research Letters, Vol. 35, No. 5.10.1029/2007GL032220Kwon, H. H., Brown, C. and Lall, U. (2008a). โ€œClimate informed flood frequency analysis and prediction in Montana using hierarch-ical Bayesian modeling.โ€ Geophysical Research Letters, Vol. 35, No. 5.DOI
7 
Kwon, H. H., Khalil, A. F. and Siegfried, T. (2008b). โ€œAnalysis of extreme summer rainfall using climate teleconnections and typhoon characteristics in South Korea.โ€ JAWRA Journal of the American Water Resources Association, Vol. 44, No. 2, pp. 436-448.10.1111/j.1752-1688.2008.00173.xKwon, H. H., Khalil, A. F. and Siegfried, T. (2008b). โ€œAnalysis of extreme summer rainfall using climate teleconnections and typhoon characteristics in South Korea.โ€ JAWRA Journal of the American Water Resources Association, Vol. 44, No. 2, pp. 436-448.DOI
8 
Kwon, H. H., Kim, J. G., Lee, J. S. and Na, B. K. (2012). โ€œUncertainty assessment of single event rainfall-runoff model using bayesian model.โ€ Journal of Korea Water Resources Association, Vol. 45, No. 5, pp. 505-516 (in Korean).10.3741/JKWRA.2012.45.5.505Kwon, H. H., Kim, J. G., Lee, J. S. and Na, B. K. (2012). โ€œUncertainty assessment of single event rainfall-runoff model using bayesian model.โ€ Journal of Korea Water Resources Association, Vol. 45, No. 5, pp. 505-516 (in Korean).DOI
9 
Kwon, H. H., Moon, Y. I. and Khalil, A. F. (2007). โ€œNonparametric Monte Carlo simulation for flood frequency curve derivation: an application to a Korean watershed.โ€ JAWRA Journal of the American Water Resources Association, Vol. 43, No. 5, pp. 1316-1328.10.1111/j.1752-1688.2007.00115.xKwon, H. H., Moon, Y. I. and Khalil, A. F. (2007). โ€œNonparametric Monte Carlo simulation for flood frequency curve derivation: an application to a Korean watershed.โ€ JAWRA Journal of the American Water Resources Association, Vol. 43, No. 5, pp. 1316-1328.DOI
10 
Lee, J. J., Lee, J. S., Kim, B. I. and Park, J. Y. (2000). โ€œDerivation of probable rainfall intensity formula of individual zone based on the representative probability distribution.โ€ Journal of Korea Water Resources Association, Vol. 33, No. S1, pp. 124-129 (in Korean).Lee, J. J., Lee, J. S., Kim, B. I. and Park, J. Y. (2000). โ€œDerivation of probable rainfall intensity formula of individual zone based on the representative probability distribution.โ€ Journal of Korea Water Resources Association, Vol. 33, No. S1, pp. 124-129 (in Korean).Google Search
11 
Lee, J. J., Lee, S. W. and Kwak, C. J. (2009). โ€œApplication of jackknife method for determination of representative probability distribution of annual maximum rainfall.โ€ Journal of Korea Water Resources Association, Vol. 42, No. 10, pp. 857-866 (in Korean).10.3741/JKWRA.2009.42.10.857Lee, J. J., Lee, S. W. and Kwak, C. J. (2009). โ€œApplication of jackknife method for determination of representative probability distribution of annual maximum rainfall.โ€ Journal of Korea Water Resources Association, Vol. 42, No. 10, pp. 857-866 (in Korean).DOI
12 
Park, C. Y., Moon, J. Y., Cha, E. J., Yun, W. T. and Choi, Y. E. (2008). โ€œRecent changes in summer precipitation characteristics over South Korea.โ€ Journal of the Korean Geographical Society, Vol. 43, No. 3, pp. 324-336.Park, C. Y., Moon, J. Y., Cha, E. J., Yun, W. T. and Choi, Y. E. (2008). โ€œRecent changes in summer precipitation characteristics over South Korea.โ€ Journal of the Korean Geographical Society, Vol. 43, No. 3, pp. 324-336.Google Search
13 
Redner, R. A. and Walker, H. F. (1984). โ€œMixture densities, maximum likelihood and the EM algorithm.โ€ SIAM Review, Vol. 26, No. 2, pp. 195-239.10.1137/1026034Redner, R. A. and Walker, H. F. (1984). โ€œMixture densities, maximum likelihood and the EM algorithm.โ€ SIAM Review, Vol. 26, No. 2, pp. 195-239.DOI
14 
Shin, J. Y. and Lee, T. (2014). โ€œParameter estimation of the mixture normal distribution for hydro-meteorological variables using meta- heuristic maximum likelihood.โ€ Journal of Korean Society of Hazard Mitigation, Vol. 14, No. 4, pp. 93-99 (in Korean).10.9798/KOSHAM.2014.14.4.93Shin, J. Y. and Lee, T. (2014). โ€œParameter estimation of the mixture normal distribution for hydro-meteorological variables using meta- heuristic maximum likelihood.โ€ Journal of Korean Society of Hazard Mitigation, Vol. 14, No. 4, pp. 93-99 (in Korean).DOI
15 
Yoon, P. Y., Kim, T. W., Yang, J. S. and Lee, S. O. (2012). โ€œEstimating quantiles of extreme rainfall using a mixed Gumbel distribution model.โ€ Journal of Korea Water Resources Association, Vol. 45, No. 3, pp. 263-274 (in Korean).10.3741/JKWRA.2012.45.3.263Yoon, P. Y., Kim, T. W., Yang, J. S. and Lee, S. O. (2012). โ€œEstimating quantiles of extreme rainfall using a mixed Gumbel distribution model.โ€ Journal of Korea Water Resources Association, Vol. 45, No. 3, pp. 263-274 (in Korean).DOI