I am looking for gridded/raster estimates of daily mean temperature (ideally ambient/at 2 meter) for a public health analysis. Specifically, I want to correlate daily mortality counts with daily mean ambient temperature (globally, between the years 1990-2016). I want to use gridded data rather than nearby temperature stations to allow for some extrapolation to areas without stations with long time series (e.g. large swaths of sub-Saharan Africa).

The accuracy of temperature in terms of degrees is somewhat less important relative to the accuracy of day-to-day variability for this project.

It seems like the GEFS Reforcast and NCEP Reanalysis products both provide daily mean temperature at a global extent, albeit at different spatial resolutions. I've looked at the documentation and underlying papers, but my earth science background is weak and I can't really suss out the difference between the two products.

Is one product substantially better than the other for global mean ambient temperature estimates? Is there a simple difference between the two products that I've missed?

Parameters I care about (roughly in order):

  1. Daily measurement of mean temperature with global coverage
  2. High spatial resolution
  3. Accuracy of day to day variation (e.g. if day 1 was X degrees hotter than day 2, is that reflected?)
  4. Accuracy of daily mean estimate

1 Answer 1


The fundamental difference is that you are comparing an ensemble reforecast with a reanalysis product. The assumption here is that the reanalysis provides the best possible information based on all available measurements using state of the art modelling.

The reforecast on the other hand simulates the forecast process based on current models and all available observations, i.e. it only utilizes observations up to the forecast base time, then runs a forecast just like a forecast system does in real life. This might still be good but you are looking at forecasts with a lead time of 0 to 24 hours if you are using the first day.

Also, GEFS is an ensemble system. That means that it performs several forecasts for the same days using different starting parameters. In this way, it is possible to estimate the uncertainty of the forecast system. E.g. if for a certain day the available forecasts deviate a lot, you can assume that there is higher forecast uncertainty.

As a summary, I'd say that using reforecast data would require some more justification than using reanalysis data. Using the ensemble information may be a good reason to use GEFS reforecast but would also add another layer of complexity.

Finally, allow me some general comments:

  • Have you considered using data from other services, most notably ECMWF? The ERA-Interim reanalysis is quite commonly used and has very good reputation. Or, as gansub pointed out, you can consider the reanalysis by JMA.
  • No matter which product you are using, do not take the results as "truth". You may find that interpolating down to an individual location requires more thought than people usually expect. For example, what happens if your location is a land location but falls into a sea pixel? This would greatly affect simulated 2m temperatures and the diurnal cycle. Using a neighbouring pixel in such cases may get you better results.
  • Speaking of the diurnal cycle, do you really care about daily means? In this kind of studies, I have seen people distinguish between day and night-time temperatures, which appeared to me as quite crucial.

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.