NCEP's Twentieth Century Reanalysis (V2) provides daily variables from 1871/01/01 to 2012/12/31. Monolayer analysis variables are made available in a regular grid and forecast variables in a gaussian grid. This distinction is made very clear in their list on their webpage, although I think the grid is not the main problem here, but the terms in bold.

Daily precipitation rate, for example, is a forecast variable. This makes you think weather you may use forecast and analysis interchangeably. What are the main differences? Any pros or cons? Are all of these still considered reanalysis? Any assumputions I need to make?

This question came up as a consequence of the discussion held here.

Thank you in advance.


Some variables in reanalyses have to be forecast and cannot be assimilated to allow the model to converse mass, momentum and energy. Precipitation is a good example. The model will assimilate observations which give information about moisture (someone who knows more about assimilation may want to chirp in here). This will be used in the analysis (e.g. 00z) and is forecast forward. Daily precipitation is therefore given as the accumulation of precipitation from 00z analysis -> 12z forecast and the 12z analysis -> 24z forecast.

There is a hierarchy of precipitation datasets each with pros and cons:

  • Gauge data is often considered the best but it very patchy (not to mentions instrument issues). You can use kriging to make spatial maps. This paper may be of interest.
  • Satellite datasets such as GPCP and TRMM give better spatial information but subject to issues over land, for example (I think). The are also subject to trackiness (only get data underneath/near the satellite).
  • Reanalyses can provide spatial and temporal homogeneity but it is a model and 'all models are wrong, but some are useful'.
  • Datasets are emerging combining both gauge, satellite and reanalyses such as in this paper.
  • $\begingroup$ So this means that the daily accumulated precipitation is entirely given by the model, which receives daily "initial condition" analysis variables, such as moisture. Therefore the amount of rainfall is very dependent on model parameterizations and so on. Am I getting it right? $\endgroup$
    – ouranos
    Aug 31 '17 at 20:12
  • $\begingroup$ I intend to use NCEP's daily precipitation rate because it is one of the only sources of daily precipitation from mid 1800's. Another option is ERA 20C (1900 to present). Any other suggestions would be very welcome. $\endgroup$
    – ouranos
    Aug 31 '17 at 20:13
  • $\begingroup$ Yes. You are right how precip is sensitive to the model's physics package etc.. I think they are your only bets if you are interested that far back. A good approach might be to use both and compare their climatology to climatology from more recent reanalyses. There are some online tools you can use as a first order check esrl.noaa.gov/psd/cgi-bin/data/testdap/plot.comp.pl Might give you some indication on which one is 'better' to use for your application. $\endgroup$
    – Ray Bell
    Aug 31 '17 at 20:27

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.