# 書報討論 新震模信息法與地震群湯模型之研究
> [name=TingKai Kuo] [time=Mon, Mar 18, 2019 14:10 PM] [color=#907bf7]

## SOG (soup of group)
### grained media
### clustering and splitting
- $v_coal$ coalition rate
- $v_frag$ fragment rate
- terrorism event (Bohorquez, 2009)
- idea: 2d ARCH
- Number of clusters N
- $n_s$ : # of size-s cluster
- simulation
- steady state (all n_s constant)
- $n_s \sim s^{-5/2}$
- Magnitude (M)
- logE = 5.24+1.44M (Bath, 1966)
- Intensity (~ acceleration)
- logN = a - bM (Gutenberg-Richter relation)
- turning point $M_c$
- SOC and Earthquake
- Fault system
- power law F(m>M)~$10^{-bM}$
- F(e>E) ~ $E^{-b}$
- p(e) ~ $e^{-2}$
- $S_max$個小單元可造成 $M_max$
- exist S cluster cause M
- 剩下的小單元所能產生的最大地震規模 $M_1 = 1og_{10}(10^{M_{max}}-10^M)$
- spacing time of earthquake

## Convensional method
### Pattern Informatics of Nonlinear threshold system
- Karhunen-Loeve expension
- Rundle et al., 2000a (Fault in Carlifornia)
- 將空間上的錯動轉化成矩陣表示(use KLE)
- use state vector to represent the phase change of earthquake activity
- machanism
- $\delta$應力 -> $\delta$相位 > high value > prediction
- Critical transition
- locking to slide
- 不穩定均衡
- 恢復力下降使得變異度增加
- calculation
- grid 0.1deg x 0.1 deg (~ 10km x 10km)
- trancated scale: Mc
- time parameter
- 孕震時間
- 變化區間 預測區間
- reference time $t_b$
- Demonstration
- Receptor operating characteristics (ROC, maching learning common used) (logorithmic scale)
- Top 100 value spatial distribution
- AUC(area under curve)

- novel earthquake modeling informatics
- conventional + SOG
- Scaled segmentic choosing 分不同的規模段
- scale window
- stride
- joint probability
- Data
- Central Weather Bureau 地震目錄
- Depth distribution plot
- <= 25 km (淺層 90%)
- M >= 5.5
- 1984/1/31 ~2016/12/31

- absolute PI value (normalized by the total spatial-temporal maximum)

- Pros and Cons
- Pros
- need not to know underground structure
- Cons
- spatial-temporal accuracy
- data-intensed