S.A.B.E.R

Semi Analytical Bayesian Error Retrieval

A radiative transfer model to retrieve inherent optical proprieties, water depth and bottom reflectance from remote sensing reflectance. Implement Markov Chain Monte Carlo

Creator and original developer: Soham Mukherjee Packaging: Raphael Mabit

Outline of the code structure

This package follow the recommendations of https://r-pkgs.org/ and the tidyverse style guide https://style.tidyverse.org/.

The code is written with a functional approach. The "business" logic (the low level functions) are written in file names starting with (fct_*), more generic function are written under (utils_*) files.

The central piece of code that stitches together the forward models with the objective function is objective_factory high level function. It allows to easily combine any forward model with any objective functions, even those that a user might add.

For users to add their own forward models and objectives function, we adopted the use of registries. A registry is an environment in which specific function are stored and can be retrieved by name in the higher level function arguments. The three registry currently in use are .input_preparer_registry, .forward_model_registry, .objective_function_registry.

Questions

• Where does the pure water IOPs comes from ? a = Pope 97, b = Morel 74

• In pure water IOP Why different pure water bb values for case one or two waters ? pure water IOP does not depends on water type.

• Where does the parametric formula to retrieve spectral slope of CDOM + NAP comes from, QAA ? QAA v5 or 6

• Use saber_forward_parametric_conc_wise.R as authoritative final function.

• The inversion of fraction rb_class is not constrained to unity ? Seems like a difficult problem ...

• What about the geometry in inversion ? With whihch value do we parametrize forward models ?

• Is it really rrs_bottom [sr^-1], is it not just Rb [dimensionless] ? Check the SVC method !

• What's the logic between negative and positive log-ll? minimizing the error vs maximizing the likelihood?

• We should probably not use extrapolation of spectrum. for example the phyto absorption spectrum in iop_from_oac