I am running ROMS with is4dvar data assimilation. Between the several datasets we assimilate, there are 3 HF radar antennas that provide radial velocities. The area covered by the radial velocities is just a very small portion of the model domain. These data are assimilated after filtering out the high frequency variability and adding the tides using TPXO. I am conducting a series of tests myself, and it would be very helpful to learn from your experience. In particular, I would like to hear about what impacts should be expected from assimilating HF radar radial velocities without filtering out the high frequency, using a variational assimilation method.
My small doubt . What was the purpose of adding tides after filtered out the high frequency signals from HF radar. Because tides are high frequency signals
..
Joao Marcos A. C. Souza: unfortunately, I have few experience with ocean dynamics. Most part of the time, I have worked with atmospheric dynamics. However, I can add a comment:
1. The insertion of data can come from a small region into domain, but the influence of the observation is larger than the collected data region, IF the co-variance matrix has entries different from zero out of the main diagonal.
My question: in your experiments, is there some positive impact by assimilating the HD radar data?
Thank you for all the comments.
The reason I filter out the high frequency is to remove processes that my grid and model configuration (eg.: forcings) are not capable of representing. Since I am using tides in the simulation, I have to add this information back to the data for the assimilation.
The impact of the HF radar radial velocities assimilation very important. The exclusion of this data from the assimilation process leads to ~50% larger errors in the SLA. This becomes even more important if we focus on the surface currents.
I have completed some experiments and am now finishing a paper on the subject. To give you an idea of the conclusions, the assimilation of raw HF-radar data gives better results for coarser grids - reducing the rms errors and time shifts in relation to the data. If we use a finer grid, the improvement is less important.
Cheers
Joao
Coarser grid simulation is not able to pick up the ocean real characteristics whereas the finer grid simulations provide a better simulation more close to reality. This may be the reason for getting more improvement in coarser grid simulation.
I would like to know the possible higher frequency signals in HF radar data other than tides.
Hi Smitha;
Yes, that is what I meant... But I would say it differently - Coarser grids are unable to resolve all the small scale processes observed in the real ocean. It is a matter of adequating your grid to the question you are interested on. No grid is capable of representing all the scales observed in the real ocean, and that is why parametrizations and closure sub-models are needed. You should keep in mind the computational cost involved in data assimilation simulations, that will limit the size of the computational grid.
To put things clear, comparing 4 experiments:
1- Coarser grid (~6km) with detided HF radar assimilation;
2- Coarser grid (~6km) with raw HF radar assimilation;
3- Finer grid (~2km) with detided HF radar assimilation;
4- Finer grid (~2km) with raw HF radar assimilation;
I observed that 2 and 4 performed better than 1 and 3, respectivelly. But the improvement of 2 in relation to 1 is bigger. The 2km grid was nested inside the 6km.
In my case, the tidal and sub-inertial frequency signals dominate the circulation - what make these grids appropriate.
- Badges
- Science topic
- Similar topics
- Geoscience
- Atmospheric Sciences
- Atmosphere