設計 確率雨量の推定(1)計算式
はじめに
確率雨量の推定方法を紹介します。 この記述の元ネタは、以下の論文によります。
- 星清:水文統計解析,開発土木研究所月報,No.540,1998年5月 (https://thesis.ceri.go.jp/db/files/0005005050.pdf)
- 星清:現場のための水文統計(1),開発土木研究所月報,No.540,1998年5月
- 星清・新目竜一・宮原雅幸:現場のための水文統計(2),開発土木研究所月報,No.541,1998年6月
ここでは、比較的良く使われると思われる、下記3種類の確率分布を考えます。 いづれも、3変数関数です。
- 3変数対数正規分布 (Log-Normal distribution with 3 parameters, LN3)
- 一般化極値分布(Generalized Extreme Value distribution, GEV)
- 対数ピアソンIII型分布 (Log-Pearson type III distribution, LP3)
本文は、たまたま昔英文のものを作っていたので、それをコピーしています。 Jupyter の Markdown で正常に表示されることを確認して「はてな」に持ってきましたが、「はてな」では正常表示するのに苦労しました。 とりあえず、今後の参考のため、「はてな」の数式表示で気がついたことをメモしておきます。
- 別行だての式では、
{align}
、{gather}
が使える。{equation}
も使えるが{displaystyle}
を使わないと行内表示となるためここでは使っていない。 - 別行だて、行内表示とも、
[ ..... ]
は\[ ..... \]
のように、必ずエスケープする。 - 行内表示ではアンダーバー(
_
)は必要に応じてエスケープする。 - 不等号は、うまく表示されない場合は、空白で囲む。
- TeX記法での
\text{ ..... }
の中で[tex: ]
は使えない。 - 水圧鉄管の記事では、
\text{ ... $...$ ... }
は使えた。
以下、数式を羅列します。
Calculation Formulas
Basis
PDF and CDF
The relationship between PDF and CDF is shown below:
PDF (probability density function) | |
CDF (cumulative distribution function) |
According to the definition of CDF, we can understand that : CDF is equal to non-exceedance probability for .
Retuen period
For Annual Maximum Series data (AMS):
Plotting position formulas
A general formula for computing plotting position is defined as follow:
Number of sample | |
Rank order in ascending order | |
Value of sample with rank order in ascending order | |
Constant for a paticular plotting position formula () | |
Plotting position (like a non-exceedance probability) |
Formula | Weibull | Blom | Cunnane | Gringorten | Hazen |
---|---|---|---|---|---|
0 | 0.375 | 0.40 | 0.44 | 0.5 |
Probability Distribution Model
Log-Normal Distribution with 3 parameters (LN3-distribution)
The parameter estimation method in the case of Hydrologic statistic $x$ following Log-Normal distribution with 3 parameters is shown below.
Probability Density Function
Cumulative Distribution Function
Quantile for non-exceedance probability
Estimation of parameters (Moment method)
To get the unbiased skewness [tex\gamma_x], the following formula for Log-Normal distribution by Bobee and Robitaille is adopted. Note that to the power of 3 is multiplied by .
The relationship between moments and parameters is shown below:
The parameters which we want to know are , and in above equations. The third equation in above is expressed as follow,
The real positive solution for above cubic equation can be obtained by Cardano's method as follow,
As a result, parameters can be estimated using following equations with which is a real positive silution of cubic equation shown above.
Log-Pearson type III distribution (LP3-distribution)
The parameter estimation method in the case of Hydrologic statistic following Log-Pearson type III distribution is shown below.
Probability Density Function
\tex:c] is a location parameter, is a scale parameter, is a shape parameter.
Cumulative Distribution Function
Quantile for non-exceedance probability
Estimation of parameters (Moment method)
To get the unbiased skewness , the following formula for Pearson type III distribution by Bobee and Robitaille is adopted. Note that to the power of 2 is multiplied by .
The relationship between moments and parameters is shown below:
As a result, parameters can be estimated using following equations.
Note that if is less than , shall be less than and the value of is for the value of .
Generalized Extreme Value distribution (GEV-distribution)
The parameter estimation method in the case of Hydrologic statistic following Generalized Extreme Value distribution is shown below. Gumbel distribution is the same as Generalized Extreme value distribution with .
Probability Density Function
is a location parameter, is a scale parameter, is a shape parameter.
Cumulative Distribution Function
Quantile for non-exceedance probability
Estimation of parameters (L-moments method)
where, is a -th value of sample in ascending order. It means that is the minimum value of sample data and is the maximum value of sample data.
The values of can be calculated using following relationship by deeming .
Parameters can be estimated using following relationship between L-Moments and parameters.