Some Questions about make a forecasting system

[cheng_dada]

I need to make a forecasting system now, and I have some questions about it. Hope all the teachers can give me some help.

I read some literature and learned that the forecast system usually provides an available initial field for the real simulation of the model through spinup or real analysis field difference. Now I plan to use HYCOM interpolation to get my initial field.

Question 1: Considering the stability issue, the models I run before usually run for 3-4 months and only use the results of the last month. However, some forecasting systems I have learned will predict the results of the next 120 hours each time it is started. Does this mean that ROMS only runs for 120 hours each time it is started? Doesn't this need to consider the stability of the model?

Question 2: In addition, I learned that many forecasting systems will perform a spinup at 00:00 every day to prevent errors in multiple integration iterations from making the forecast deviate. Is this step necessary? Because the initial field of my forecast started at 12:00 is obtained by the assimilation of the analysis field and observation data obtained at that time from the last forecast。

Thank you teachers for your answers.

[arango]

All the modern operational forecasting systems for the atmosphere and ocean do not do spinup. They require initial ocean analysis conditions that combine the model with the observations. They use data assimilation, which is a pretty advanced subject. The limitation of true ocean prediction is the availability of forcing field forecasts from atmospheric models. Also, if you use the regional application, you need the large-scale forecast from the basin or global ocean models for lateral boundary conditions. Recall that in the real-time forecast, you are predicting the future state of the atmosphere and ocean, and they are coupled!

This subject is too complex for someone starting to learn about ocean prediction. It takes years of expertise and several scientists and data systems to get operational ocean forecasting going.

[philip]

What are you trying to forecast?

If you are only forecasting total water levels or flooding, the problem is somewhat simpler and less sensitive to the use of data assimilation. Instead it requires more attention to met forcing, ensembling, river inputs, and perhaps postprocessing bias corrections based on observations. But even this sub-field is moving in the direction of data assimilation.

There are 3D and 2D model-based water level forecast systems without data assimilation that are widely used by the US National Weather Service —
https://ams.confex.com/ams/2020Annual/w ... S-2020.pdf
https://www.sciencedirect.com/science/a ... 4722000834

More broadly, if one is seeking to use COAWST to simulate storm impacts, it is also a somewhat different problem.

Feel free to reach out if you’d like more information.

[cheng_dada]

What are you trying to forecast?

If you are only forecasting total water levels or flooding, the problem is somewhat simpler and less sensitive to the use of data assimilation. Instead it requires more attention to met forcing, ensembling, river inputs, and perhaps postprocessing bias corrections based on observations. But even this sub-field is moving in the direction of data assimilation.

There are 3D and 2D model-based water level forecast systems without data assimilation that are widely used by the US National Weather Service —
https://ams.confex.com/ams/2020Annual/w ... S-2020.pdf
https://www.sciencedirect.com/science/a ... 4722000834

More broadly, if one is seeking to use COAWST to simulate storm impacts, it is also a somewhat different problem.

Feel free to reach out if you’d like more information.

Thanks you for your answer. My previous description was not clear enough. I wanted to build a model to predict the current within 200km of the shore. Data assimilation is not something I need to worry about, it will be done by other teams. I just need to build a predictive model that doesn't assimilate the actual data. So, I intend to use HYCOM's forecast results as background and boundary fields, and ERA-Interim data as forcing fields.