About ECA_HWDI
Added by Tufan Turp over 4 years ago
Dear all,
I'm writing about a common problem with the limited answers in the forums. I'm trying to calculate hwdi. I have two input files: one for future period spanning 2016-2035 (daily tasmax) and one for reference 1986-2005 (daily tasmax). Normally CDO suggests to use:
cdo eca_hwdi tasmax_fut.nc tasmaxnorm_ref.nc hwdi.nc
Here, since CDO says "TXnorm is calculated as the mean of maximum temperatures of a five day window centred on each calendar day" we generally use the command ydrunmean,5 to get tasmaxnorm_ref.nc:
cdo ydrunmean,5 tasmax_ref.nc tasmaxnorm_ref.nc
If we run the previous command we will get an error because of different timeteps between two input files. My tasmax_fut.nc file has 7305 timesteps ((20*365)+5 bissextile day), whike tasmaxnorm_ref.nc has 366 timesteps.
In order to handle this problem, I applied the following steps regarding some suggestions:
cdo sub tasmax_ref.nc tasmax_ref.nc tasmax_zero.nc
cdo ydrunmean,5 tasmax_ref.nc tasmaxnorm1_ref.nc
cdo ydayadd tasmax_zero.nc tasmaxnorm1_ref.nc tasmaxnorm2_ref.nc
cdo eca_hwdi tasmax_fut.nc tasmaxnorm2.nc_ref.nc eca_output.nc
It works and give output. I have a few questions about it:
1- Is it a true approach?
2- Is 5-day window centred mean good enough for a 20-year period?
3- Can someone explain the technical idea behind it? Because similar problems occur for other indices as well, if we want to calculate an index for a longer period with respect to a single or shorter reference period.
Thanks for all comments and ideas.
Tufan
Replies (4)
RE: About ECA_HWDI - Added by Fabian Wachsmann over 4 years ago
Dear Tufan,
dealing with leap years in climate model output has caused many scientists to stumble, and as I know from discussions, there is no pretty approach on that topic.
I think one idea is to say since there are just about 25% February 29ths (or whatever your bissextile day is) in the time series so that the sample size for February 29th is just not big enough. Therefore, remove it from the time series before analysing it.
3- Can someone explain the technical idea behind it? Because similar problems occur for other indices as well, if we want to calculate an index for a longer period with respect to a single or shorter reference period.
Maybe we should make the upper approach the default in CDO?
1- Is it a true approach?
Hard to say. Is the itention of ydayadd to add a bissextile day with only zeroes to all your files? Wouldn't that result in a lowered running mean around that day?
2- Is 5-day window centred mean good enough for a 20-year period?
Define "good enough". You have 100 values per day.
There are statistical methods like bootstrapping to make the analysis more reliable. I implemented it in some etccdi operators which are available in the recent CDO versions but they take a lot of processing time. And, unfortunately, the hwdi index is not part of etccdi.
Best regards,
Fabi
RE: About ECA_HWDI - Added by Tufan Turp over 4 years ago
Dear Fabian,
Thank you very much for your comments. Your comments are always very beneficial. I agree with you. Acutally, I'm looking for using an index consecutive days period w.r.t. the reference. Alternatively, I may calculate 90th/95th percentile of the reference data. Then I can caunt the periods when it is at least 5/6-day above consecutively in the future. I may be wrong, but in my mind it says there should be one threshold value for each grid point. Hence, the timestep of the reference period must have single value not timeseries 365 or 366 or whatever.
Thanks,
Tufan
Fabian Wachsmann wrote:
Dear Tufan,
dealing with leap years in climate model output has caused many scientists to stumble, and as I know from discussions, there is no pretty approach on that topic.
I think one idea is to say since there are just about 25% February 29ths (or whatever your bissextile day is) in the time series so that the sample size for February 29th is just not big enough. Therefore, remove it from the time series before analysing it.[...]
Maybe we should make the upper approach the default in CDO?[...]
Hard to say. Is the itention of ydayadd to add a bissextile day with only zeroes to all your files? Wouldn't that result in a lowered running mean around that day?
[...]
Define "good enough". You have 100 values per day.
There are statistical methods like bootstrapping to make the analysis more reliable. I implemented it in some etccdi operators which are available in the recent CDO versions but they take a lot of processing time. And, unfortunately, the hwdi index is not part of etccdi.Best regards,
Fabi
RE: About ECA_HWDI - Added by Tufan Turp over 4 years ago
By the way, sorry for some mistakenly written words.
RE: About ECA_HWDI - Added by Ralf Mueller over 4 years ago
hi!
- I would delete all Feb 29 from the data since I don't think there is a climatological significance. (operator
del29dfeb) - there are two more general operators for consecutive events:
consectsandconsecsum(see here for more)
hth
ralf