1
$\begingroup$

The question is a bit weird so i'll open it up. So i have a table of return periods for different amounts of rain. The table has been made using GEV distribution on known data and then the mean and confidence intervals are calculated for different return periods. GEV distribution of precipitation

I need to recreate this curve closely as possible in another program that does not have GEV-distribution so i will have to use lognormal in this case. How can i get the parameters for a lognormal distribution that would be close to the GEV curve. I also don't have the original datapoints, just the data from the curve.

Return periods

$\endgroup$

1 Answer 1

0
$\begingroup$

With the R code below, I found the lognormal distribution which minimizes $$\sum(\log_{10}(\text{actual period})-\log_{10}(\text{predicted period}))^2.$$ This gives $\mu=2.994$, $\sigma=.0516$, and the following curiously well-fitting graph: periods and predicted periods vs precipitation

periods <- c(10, 25, 50, 100, 200, 300, 500, 10^3, 10^4, 
             10^5, 10^6, 10^7, 10^8, 10^9)
precips <- c(44, 53, 60, 68, 76, 81, 87, 96, 131, 174, 225, 
             288, 365, 458)

periodfits <- function(params){1/(1-plnorm(precips, 
                       params[[1]], params[[2]]))}
score      <- function(params){sum((log10(periods) - 
                       log10(periodfits(params)))^2)}

par <- optim(c(1,1), score)$par
print(par)
plot(  precips, log10(periods))
points(precips, log10(periodfits(par)), col=2)
$\endgroup$
2
  • $\begingroup$ What if i need to change the periods to probability. So changing to 10^-1, 10^-2., 10^-3..etc. Can't seem to get it to work as it does not want to go under 0 if using the lognormal scale. $\endgroup$ Aug 16, 2021 at 13:51
  • $\begingroup$ If the transformations $\mu_{prob}=-\mu_{period}$ or $prob=1/period$ are causing difficulties, I'm not sure what to say. But if there's anything in the answer that you found useful, then an upvote would be nice! $\endgroup$
    – Matt F.
    Aug 16, 2021 at 17:17

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.