GOA-CLIM / ROMS_to_Index

Functions to develop environmental indices from Regional Ocean Modelling System models

Overview

This repo overviews the process for making environmental indices from the ROMS NEP data including: data overview, index creation, and delta correction. We don't proved generic functions because each GOA-CLIM modeler may have indices specific to a particular spatial and temporal domain. We therefore provide an example Quarto that folks can copy and derive their desired indices.

Data overview

Projections of environmental variables biomass are derived from a downscaled Intergovernmental Panel on Climate Change (IPCC) projection model from the coupled model intercomparison project (CMIP): GFDL-ESM2M. The IPCC projection model was downscaled using the North East Pacific (NEP) Regional Ocean Modeling System (ROMS) developed for the Gulf of Alaska.

ROMS simulations are referred to as the following:

• hindcast: representing the final year of the spinup forced with observed oceanographic conditions to better represent historical conditions (1990 to 2020),

• projection: GFDL-ESM2M downscaled projection (2015 to 2099)

• historical: representing model spinup (1980 to 2014).

Projections are available in the following $CO_2$ pathways:

• ssp126
• ssp245
• ssp585

Variables from ROMS simulations averaged for each layer and grid cell on a monthly time scale were averaged or summed across the following vertical distributions:

• Surface (from 0 to 10 m depth)
• Bottom (deepest ROMS point assumed to be representative of bottom conditions)
• Midwater (from 10 m depth to just above the bottom layer - deepest ROMS point)

NMFS areas are clipped to isobaths of interest and the index is calculated as an average for the intersection between the NMFS area and the isobath. The isubaths we use are the 300 m and the 1,000 m isobaths. Areas are calculated from st_area in the sf package. For the 300 m isobath:

• 610 (area = 57225003746 $m^2$)
• 620 (area = 62597059226 $m^2$)
• 630 (area = 98582220025 $m^2$)
• 640 (area = 32560976631 $m^2$)
• 650 (area = 36726651409 $m^2$)

And for the 1,000 m isobath:

• 610 (area = 63986698621 $m^2$)
• 620 (area = 69583703140 $m^2$)
• 630 (area = 105918077937 $m^2$)
• 640 (area = 37270389681 $m^2$)
• 650 (area = 43952466109 $m^2$)

The "raw" monthly data are located in the /Data/NEP_10k_revised_indices folder. They represent the monthly indices either averaged across each depth and spatial domain (nep_avg_…) or summed across each depth domain and then averaged across the spatial domain (nep_sum…).

Delta correction overview

Delta correction is done to correct for shifts in the time series between the hindcast and projection. For normally distributed variables:

$$X^{proj'}_t=\bar{X}^{hind}_{\bar{T}}+ \left(\frac{\sigma^{hind}_{\bar{T}}}{\sigma^{hist}_{\bar{T}}} * \left(X^{proj}_t-\bar{X}_{\bar{T}}^{hist} \right)\right)$$

where $X^{proj'}_t$ is the bias corrected projection index in time-step $t$, $\bar{X}^{hind}_{\bar{T}}$ is the average index value from the hindcast during reference period $T$, $\sigma^{hind}_{\bar{T}}$ is the standard deviation of the index from the hindcast during the reference period (1990-2014), $\sigma^{hist}_{\bar{T}}$ is the standard deviation of the index from the historical run during the reference period, $X^{proj}_t$ is the non bias-corrected projection index, $\bar{X}_{\bar{T}}^{hist}$ is the average value from historical run during the reference period.

For log-normally distributed variables the formula can be adjusted as follows:

$$X^{proj'}_t=\exp\left(\bar{logX}^{hind}_{\bar{T}}+\left(\frac{\sigma^{hind}_{\bar{T}}}{\sigma^{hist}_{\bar{T}}} * \left(logX^{proj}_t-\bar{logX}_{\bar{T}}^{hist} \right)\right)\right)$$

where $T$, $\sigma^{hind}_{\bar{T}}$ is the standard deviation of the log index from the hindcast during the reference period, $\sigma^{hist}_{\bar{T}}$ is the standard deviation of the log index from the historical run during the reference period.

Because the smallest time step in month, the mean and variance terms should be calculated as follows:

Monthly indices: time-step $t$ should be year $y$ and month $m$ (e.g. $t=y,m$). A function to do the bias correction is in the R/Delta_correction.r.

Users can decide to apply a delta correction method that does not use the ratio between the standard deviations as scaling factor. Scaling by the standard deviation may result in artefacts being introduced, for example if one of the $\sigma$ values is particularly high. This should not happen much at the spatial and temporal resolution used here, but it has been a known issue in Atlantis (which uses daily time steps and a small spatial scale, resulting in fewer ROMS points per calculation) and in some ACLIM applications. The default is use_sd = TRUE.

When doing the delta correction, users can decide whether they want the end product to be a time series that consists of the bias-corrected historical run and projection, or historical run, hindcast, and projection. The second option splices the hindcast (which incorporates observations) into the full bias-corrected time series. Analysts that are interested in exploring the effects of recent climate events in the GOA (e.g., the heat wave) may want to use include_hindcast = TRUE. If projections are the focus, this becomes less important.

Index derivation

Environmental indices can be derived for each modelers following their specific needs. We don't develop all indices here as individual modeler's needs may vary. But, indices integrated across each spatial NMFS management area should be area weighted using the following areas.

A quarto document document exampling some index generation can be found here

Locations of monthly average output files on loon

File naming convention: The suffix moave_{yr}_{mo} indicates monthly average, year, and month, e.g. /ahr0/hermann/goa-output/monthly_aves/cgoa_moave_2000_01.nc is the monthly average for January of year 2000.

The "master directories" and grid files are here:

/ahr0/hermann/goa-output/

The latitude, longitude, and depth locations of each grid point are in these files:

• zpoints_cgoa.nc

• zpoints_nep.nc

For variables like temperature and salinity, the relevant variables for latitude, longitude and depth are lat_rho, lon_rho, zval_rho

Note there are values for u, v, and w variables, which are on slightly different (staggered) grids.

The major subdirectories of /ahr0/hermann/goa-output/ are as follows:

• cgoa 3km model hindcast: monthly_aves

• nep 10km model hindcast: monthly_aves_nep_hind

• nep 10km model GFDL "historical" run: monthly_aves_nep_wb_hist

• nep 10km model GFDL ssp126 projection: monthly_aves_nep_wb_ssp126

• nep 10km model GFDL ssp245 projection: /ahr2/hermann/monthly_aves_nep_wb_ssp245 (note different path)

• nep 10km model GFDL ssp585 projection: monthly_aves_nep_wb_ssp585

Aclim links

Can be used to develop indices from 10K resolution ROMS models given an input of a spatial field following: https://kholsman.github.io/ACLIM2/#23_ROMSNPZ_variables