[feedback][user_data] Maurer/NLDAS forcings and benchmark models

[Frederik Kratzert]

Dear Andy and Nans,

just to keep track on possible updates before a CAMELS v2 release, and also with keeping the CAMELS benchmark modeling effort in mind. Maybe a brief summary, what is meant with "CAMELS benchmark modeling". A number of modeling groups used the same forcings (Maurer) and same calibration/validation period to train their models. These models will soon be released and allow for a much broader model benchmarking as only to compare against SAC-SMA. The set of benchmark models include basin-wise calibrated models, as well as CONUS-wide calibrated models.

However, there is a number of problems with releasing this data:

- The selected period are calibration: 1.10.1999 - 30.9.2009 and validation: 1.10.1989 - 30.9.1999. However, the CAMELS data set currently only contains Maurer forcings until 31.12.2008.

- Some of these models use daily min/max temperature as inputs. However, the CAMELS data set only contains daily mean temperature for Maurer forcing data (as well as for NLDAS).

Obviously, everybody take the gridded data products and derive the missing period on their own. However, this introduces a lot of uncertainty and potential for errors, since the shape files may vary. All in all, I think that if the aim is to release a true set of benchmark models it should be guaranteed that they use the _exact_ same inputs and all of these inputs are available to anybody for training/calibrating their own models. 

So in summary, we need:

- update of CAMELS data set to include Maurer (and NLDAS) forcing data until at least 30.9.2009

- Make sure that all benchmark models used the exact same data, if not we have to report this at least in a possible release of the benchmark models

As always, happy to help where I can

Best,

Freddy

[Frederik Kratzert]

Update: So there has been some confusion (on my side) regarding the calibration period of the benchmark models. Seems like all benchmark models were calibrated on the period 1.10.1999 until 30.09.2008 (not 2009) thus only 9 years of data, not 10 years like the validation period.

Therefore the only thing missing in the CAMELS data set are the min/max temperatures for the NLDAS forcing data which are e.g. used by VIC (one of the benchmark models).

This reduces the problem considerably!