T4.4.2: Some core Fortran scripts to be embedded within NEMOVAR-related balancing scheme to assimilate spectral Rrs.
This repository contains some Fortran modules that include the capability to implement spectral remote sensing reflectance (Rrs) increments and translate them into increments of ERSEM model state variables via a balancing scheme. Unfortunately due to delays caused by the Met Office HPC MonSOON system transition this code has not been tested in 3D, but an analogous balancing approach has been developed in 1D set-up, based on the EAT software (Bruggeman et al. 2024, https://gmd.copernicus.org/articles/17/5619/2024/). It has been demonstrated that in 1D, at the L4 location in the western English Channel, this approach (based on assimilating 6 wavebands of the OC-CCI satellite multi-spectral satellite products: 412nm, 443nm, 490nm, 510nm, 560nm, 665nm) performs well, in the sense that over the 2002-2017 period it improves the spectral Rrs forecast (lead time dependent on the next data availability, on average ~5 days) relative to the free run by 12-60% (depending on the wavelength). The skill was measured by the median absolute differences relative to the (relatively high-quality) OC-CCI satellite Rrs. This is a clear indicator of (overall) improved representation of the ERSEM optically-active tracers (after NECCTON WP5 developments these consist of phytoplankton, detritus, CDOM and SPM), as the Rrs depends solely on their (improved) concentrations. The approach works by using model Rrs outputs derived as in NECCTON WP5 (fabm-spectral, T5.2.4), based on the approximation assuming that optically active layer is contained within the mixed layer. The increments for Rrs are subsequently translated into the increments for the ERSEM optically active tracers (total phytoplankton, detritus, CDOM and SPM concentrations) based on third order polynomials that were fitted in the 1D simulations. How independent these relationships are on the L4 location needs to be still tested. These increments are then sub-divided into the functional types (ERSEM state variables) as per the ``standard'' approach based on background ratios (see Skakala et al. 2018, https://agupubs.onlinelibrary.wiley.com/doi/full/10.1029/2018JC014153). The analytical (3-rd order polynmial) model is used, as it can be easily coded within the existing balancing module, however more complicated approaches e.g. involving ML inversion models can be also developed.
Baseline: in 1D this approach was compared with satellite chlorophyll assimilation run (chlorophyll derived from the satellite Rrs assimilated here).
Metrics: Comparing Rrs forecast at the next assimilation time with satellite OC-CCI Rrs, using RMSE as metric.