0

Is there a way to determine the files, and perhaps time-steps within the files, that are used to construct a "best-time-series" within THREDDS? We currently serve output from a numerical model via THREDDS. We have output each day (ideally) that contains 3-hourly output for eight days. The TDS best-time-series uses the values from the current day forecast for each day. In cases where the model output was not produced on a certain day, the TDS uses the forecast from the day before. This seems to fall apart on our server at a specific time.

Attached are two images. Both contain the values of a certain variable (temperature in degrees K) at a certain point (lat=0, lon=180). The black line was created using the TDS best-time-series, and the red/green lines were created using the daily files. The figure for January seems to match my understanding; the TDS line tracks the values from the first file for Jan 1, then the second file for Jan 2 (black overlays red first then green). Jan_2013

In comparison, the second figure for March 2013 shows a totally different line for the TDS and the individual files. I would like to determine from where the TDS "best-time-series" is getting these values. March_2013

JimP
  • 83
  • 5

1 Answers1

0

I think I've discovered the problem. The client s/w used to create the above plots (GrADS, http://cola.gmu.edu/grads/grads.php) assumes a linear time axis. This can lead to problems like the above if the time-series is based on an aggregation and there are missing files. GrADS will assume times based on initial time + n * dt. A missing file in an aggregation goes unnoticed, and the next valid file/time is used.

As an example, if the aggregation is done over monthly files (jan.dat, feb.dat, etc.) and "mar.dat" is missing, GrADS will simply use the apr.dat file as the next time (march), then may.day as apr.dat and so on.

The data set used in the initial question had two months of missing output, thus the mismatch between the "best time series" and the actual files.

JimP
  • 83
  • 5