Support for non-uniform observations?

[Balinus]

Hello!

Before I try to call an external model in C++, I wanted to know if there is a support for non-uniform observations? For example, here's our constraint with a hydrological model. We want to assimilate snow data that are gathered "randomly" by human.

• Our observations are not "periodic" : most years, but not always, observations are available on February 15th, March 1st-15th, April 1st (those date, for a given year, can be displaced by some days. i.e. instead of Feb 15, March1 and 15, it could be Feb 16, Feb 28, March 16, ...)
• Rest of the year, there are no observations
• Some year, we have continuous obs between February and May

Cheers|

[Alexander-Barth]

Well for the high-level "driver" functions uniform observations are indeed required but not the underlying anlysis routines. It should be easy to change to make H and R time variant, but what could be a sensible API? The observations yo is currently just an array space x time. I see these possibilities:

1. get an extra time dimension of H and R (like the observations)
2. H and R are vectors of matrices (potentially lazily generated)
3. H and R could be function of the time (or time index), when called e.g. R(t) would return a matrix

Maybe we need also to revise the way observations are handled.

[Balinus]

Thanks!

I think option 3 makes sense. We have Matlab code that does this behaviour: based on t we extract the distributions of errors based on the time. Since this is snow, such an error goes to 0 for some conditions. Otherwise, notably with EnKF, there is constant creation of snow.

[Alexander-Barth]

I updated the driver functions as shown here:

https://alexander-barth.github.io/DataAssim.jl/dev/

H will be handled the same way as the model (because for some assimilation schemes all the same methods are used). For y (and R) I use a vector of vector (or matrices). But all elements can also be generated on the fly.

[Balinus]

Thanks! I will take a look at those as soon as I find a way to call the external model 😄