Based on my analysis for the AMT priors paper, it's well worth implementing rapid NOAA data for all gases, but especially CH4. The main problem is how to make sure that consistent priors are generated for a given date as new NOAA data becomes available (this is more a concern for the automatic prior generation than general use case). We've already solved this for OCO-2/3 by putting in place code that allows us to truncate the input NOAA data at a specific date, and always using the observation date (possibly minus one or two months to allow for data latency) as that date.
However, doing that requires regenerating the stratospheric look-up tables, because it affects the stratospheric boundary condition. For CO2, this is annoying but manageable. For N2O and CH4, this is atrociously slow, and would make the automatic priors generation practically unusable. Thus, we probably need to work on optimizing the strat LUT generation to enable this change, and that may be best done as an overall effort to speed up core components of ginput through the use of numba or Rust and pyo3.
Based on my analysis for the AMT priors paper, it's well worth implementing rapid NOAA data for all gases, but especially CH4. The main problem is how to make sure that consistent priors are generated for a given date as new NOAA data becomes available (this is more a concern for the automatic prior generation than general use case). We've already solved this for OCO-2/3 by putting in place code that allows us to truncate the input NOAA data at a specific date, and always using the observation date (possibly minus one or two months to allow for data latency) as that date.
However, doing that requires regenerating the stratospheric look-up tables, because it affects the stratospheric boundary condition. For CO2, this is annoying but manageable. For N2O and CH4, this is atrociously slow, and would make the automatic priors generation practically unusable. Thus, we probably need to work on optimizing the strat LUT generation to enable this change, and that may be best done as an overall effort to speed up core components of ginput through the use of numba or Rust and pyo3.