powellb
commented
2 years ago I don't quite understand your description. You are saying that when you run the model, in the first non-linear model run (when it samples the model at the observation locations and times), it ignores the observations for the first 12 hours? The nonlinear model does not care about the obs_values or obs_error, per se, it just samples the variable at the proper time and location. So, there is nothing that should lead to a blowup.
One thing that I have seen happen is that you must pay careful attention to obs_time. The difference between obs_time for observations must be greater than DT of your model. For instance, let's assume your DT=3600 (1 hour). Next, your obs_time is in days from some reference. So, in this example, your model starts on day 1200 and runs until day 1205. You want observations that span 1200 to 1205, but never had a difference between the prior of less than 1/24=0.042.
Imagine now you have observations of:
obs_time = [1200, 1200, 1201, 1201, 1201, 1204, 1205]
This is fine, but if you have:
obs_time = [1200, 1200, 1200.03, 1201, 1201, 1204, 1205]
it will fail because ROMS will load the obs at t=1200, then the model steps to time 1200.042, but it passed the 1200.03 obs, and then it is stuck, it will not load any additional obs after that point.
If that isn't the issue, you just need to examine the resulting obs.nc file. Are the times, locations, etc. different? Is there an issue converting lat/lon to i,j (it should be fine, but if you cross 0 or 180 longitude, or you consistent with positive or negative values)?
There is nothing that should lead to a blowup in the first NLM run of the model. If it blows up in the inner-loop, there can be an issue with the obs_value or obs_error. If those are completely unreasonable, there could be an issue.
Hope that helps.
On Apr 4, 2022, at 3:44 AM, ymamoutos @.***> wrote:
Greetings
Recently our group switched from matlab to python using seapy package for the pre-process of the observations that are need for ROMS 4DVAR scheme (formerly PSAS now RBL4DVAR) . I am experiencing a weird issue with the python processed observation file when i use it: During the initialization of the first inner loop (outer = 001, inner =001 not 000) although obs are not rejected ROMS doesn't read their values for the first 12 hours and this - let's say - gap ends (obviously) to a blowup. Using matlab processed file with the same datasets for observations model runs smoothly. obs_error and obs_variance differ between the two files - I am setting fixed error values for the state variables in the matlab scripts - but also the number of observations is different, with the seapy produced file having more compared to matlab's.
My experience with python is limited (now I am learning) and I am trying to understand why this is happening but I had no luck. Any help or insight will be highly appreciated.
I also am using the matlab scripts from https://www.myroms.org/svn/src/matlab https://www.myroms.org/svn/src/matlab to create the observation file.
Kind Regards
Giannis Mamoutos
— Reply to this email directly, view it on GitHub https://github.com/powellb/seapy/issues/66, or unsubscribe https://github.com/notifications/unsubscribe-auth/AAANGP7DEDW4SZ3ZJAAESGTVDLW2BANCNFSM5SPSWCAA. You are receiving this because you are subscribed to this thread.
ymamoutos
ivicajan
Greetings
Recently our group switched from matlab to python using seapy package for the pre-process of the observations that are need for ROMS 4DVAR scheme (formerly PSAS now RBL4DVAR) . I am experiencing a weird issue with the python processed observation file when i use it: During the initialization of the first inner loop (outer = 001, inner =001 not 000) although obs are not rejected ROMS doesn't read their values for the first 12 hours and this - let's say - gap ends (obviously) to a blowup. Using matlab processed file with the same datasets for observations model runs smoothly. obs_error and obs_variance differ between the two files - I am setting fixed error values for the state variables in the matlab scripts - but also the number of observations is different, with the seapy produced file having more compared to matlab's.
My experience with python is limited (now I am learning) and I am trying to understand why this is happening but I had no luck. Any help or insight will be highly appreciated.
I also am using the matlab scripts from https://www.myroms.org/svn/src/matlab to create the observation file.
Kind Regards
Giannis Mamoutos