0
$\begingroup$

Currently, I am working on historic climate variable data of precipitation and temperature. I want to resample my data to find out variability in precipitation and temperature. But I read somewhere that the bootstrap approach only works for hydrologic data for reproducing the historical frequency distribution of streamflow. However, it does not work well for precipitation and temperature data.

Please guide me in this respect.

$\endgroup$
1
$\begingroup$

This should've been a comment but turned into a longer post... I would imagine that it has to do with stationarity of the data series. In my experience, the bootstrap approach has worked well for stationary data. For nonstationary data, there are, however, non-stationary bootstrap procedures (see tsboot and meboot packages in R) which try to incorporate some of the correlation structure into the sampling procedures. There was a nice article discussing some of the dangers of using the block bootstrap which I will try and find / post later.

$\endgroup$
  • $\begingroup$ Here's the article which I found from this question $\endgroup$ – rrrrr Jul 28 '16 at 14:36
  • $\begingroup$ @Ejaz: It could be true what you say but that wouldn't be due to precipitation and temperature data in general but rather due to the underlying distribution of the specific data considered.. So, if the "sample" ( which in bootstrapping is your population ) is somehow odd or contaminated, then the boostrap distribution is going to be wrong. So, that may be what the person who said that meant. I couldn't tell if you were dealing with a time-series but, if so, then that adds complication because of the error structure. In that case, read things by Davidson and Mckinnon. They're extremely clear. $\endgroup$ – mlofton Sep 25 at 7:36
  • $\begingroup$ Note that the link posted by @rrrrr points to a paper co-authored by Davidson. $\endgroup$ – mlofton Sep 25 at 7:41

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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