{
"cells": [
{
"cell_type": "markdown",
"metadata": {},
"source": [
"# Surface Water-Mass Transformation \n",
"\n",
"In this notebook, we compute surface water-mass transformation rates (both in the net and partitioned into contributions from heat and salt fluxes) for the southern ocean, south of 60$^\\circ$S. \n",
"\n",
"Here we are looking specifically at ACCESS-OM2-01 output (0.1$^\\circ$ resolution, latest spinup), as this configuration has demonstrated considerable success when simulating high latitude dense water formation processes. "
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## 1. Defining surface water-mass transformation\n",
"\n",
"The surface water-mass transformation framework described here follows [Newsom *et al* (2016)](https://journals.ametsoc.org/doi/full/10.1175/JCLI-D-15-0513.1) and [Abernathey *et al* (2016)](https://www.nature.com/articles/ngeo2749). Surface water-mass transformation may be defined as the volume flux into a given density class ($\\sigma$) from lighter density classes ($\\sigma'<\\sigma$) due to surface buoyancy forcing. Integrated over a region of the ocean surface, this volume flux can be expressed as,\n",
"\n",
"$$\n",
"\\begin{equation}\n",
"\\Omega(\\sigma, t) = \\frac{\\partial}{\\partial\\sigma} \\iint_{\\sigma'<\\sigma} \\Big(\\frac{\\partial\\sigma}{\\partial\\theta}\\theta + \\frac{\\partial\\sigma}{\\partial S}S\\Big) \\, \\mathrm{d}x \\, \\mathrm{d}y\n",
"\\end{equation}\n",
"$$\n",
"\n",
"where $t$ is time, and the terms in the integrand are the potential temperature ($\\theta$) flux and salinity ($S$) flux components of the surface buoyancy flux. The linearity of this expression means we can extract the relative contributions of heat ($\\Omega_\\text{heat}$) and salinity ($\\Omega_\\text{fw}$) fluxes to surface water-mass transformation, highlighting driving mechanisms."
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"## 2. Requirements\n",
"\n",
"**The cosima-cookbook**: The `conda/analysis3-22.10` (or later) module on NCI (or your own up-to-date cookbook installation).\n",
"\n",
"**Model diagnostics**: for this example I use some of the long RYF9091 spinup output, if you want to use do surface water-mass transformation analysis on a different experiment you'll need to check you have saved the required diagnostics.\n",
"\n",
"**Warning:** if you are using a surface heat flux diagnostic computed online by an ACCESS-OM2 run, make sure that this diagnostic is computed correctly! Some issues were found in the online computation of `net_sfc_heating` (see this [git issue](https://github.com/COSIMA/access-om2/issues/139) from March 2019). In this example we compute net surface heat fluxes offline from component fluxes to avoid these problems - if in doubt use the component fluxes as done here. If you dont have these variables saved and can't re-run... `net_sfc_heating` isn't really so bad for near Antarctic surface water-mass transformation calculations, since transformation rates are strongly dominated by freshwater fluxes, so the errors in heat fluxes confer only a small error to transformation rates.\n",
"\n",
"We'll need at least monthly resolution (in `diag_table` language...): `surface_temp`, `surface_salt`, `pme_river`, `sfc_salt_flux_restore`,`sfc_hflux_from_runoff`, `sfc_hflux_coupler`, `sfc_hflux_pme` and `frazil_3d_int_z` (although the latter is not strictly a surface flux, instead frazil heat fluxes are vertically integrated over the column. BUT this variable is highly surface intensified and, in most cases, conceptually relevant to what we want to know from SWMT analysis) plus some basic gridding information."
]
},
{
"cell_type": "markdown",
"metadata": {
"tags": []
},
"source": [
"## 3. Computing surface water-mass transformation"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"First, import some required modules and functions."
]
},
{
"cell_type": "code",
"execution_count": 1,
"metadata": {},
"outputs": [],
"source": [
"%matplotlib inline\n",
"%config InlineBackend.figure_format='retina'\n",
"\n",
"import xarray as xr\n",
"import numpy as np\n",
"import dask.array\n",
"import cosima_cookbook as cc\n",
"from gsw import alpha, SA_from_SP, p_from_z, CT_from_pt, beta, sigma1 \n",
"# note these gsw numpy functions mean some care is needed to avoid memory issues in very high-res analysis \n",
"\n",
"## plotting\n",
"import matplotlib.pyplot as plt\n",
"from matplotlib import gridspec\n",
"import cmocean.cm as cmo\n",
"import matplotlib.colors as col\n",
"import cartopy.crs as ccrs\n",
"import matplotlib.path as mpath\n",
"from matplotlib import rc\n",
"rc('text', usetex=True)\n",
"rc('xtick', labelsize=25) \n",
"rc('ytick', labelsize=25) \n",
"rc('axes', labelsize=25) "
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"You should also set up a `dask` client - this code uses the vanilla method for doing this. I suggest using Gadi and an entire node (48 cores) for this calculation."
]
},
{
"cell_type": "code",
"execution_count": 2,
"metadata": {},
"outputs": [
{
"data": {
"text/html": [
"
\n",
" Local directory: /jobfs/77955591.gadi-pbs/dask-worker-space/worker-72a55hyl\n",
"
\n",
"
\n",
"\n",
" \n",
"\n",
" \n",
"\n",
"
\n",
" \n",
"
\n",
"
\n",
" \n",
"\n",
" \n",
"
\n",
"\n",
" \n",
"
\n",
"
\n",
" \n",
" \n",
"\n",
"
\n",
"
"
],
"text/plain": [
""
]
},
"execution_count": 2,
"metadata": {},
"output_type": "execute_result"
}
],
"source": [
"## once you've set up a dask-worker, connect to it, click the dashboard link to check worker status\n",
"from dask.distributed import Client\n",
"client = Client()\n",
"client"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"I'll be using output from the ACCESS-OM2-01 RYF9091 spinup. "
]
},
{
"cell_type": "code",
"execution_count": 3,
"metadata": {},
"outputs": [],
"source": [
"session = cc.database.create_session()\n",
"expt = '01deg_jra55v13_ryf9091'"
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"#### The function below will compute surface water-mass transformation rates for the southern ocean, averaged over an integer number of years of monthly resolution output. Below I have flagged some possible modifications you may want/need to employ: \n",
"\n",
"- If you are not computing net surface heat fluxes from its components (here we define `net_surface_heating = sfc_hflux_from_runoff + sfc_hflux_coupler + sfc_hflux_pme + frazil_3d_int_z`), the function below will need to be modified to account for this.\n",
"- If you are not looking at the southern ocean, alter the `yt_ocean` slicing regime.\n",
"- You may want to use a different density binning array (`isopycnal_bins`) if the phenomena you're interested in occurs in a shifted density space.\n",
"- We use $\\sigma_1$ here because we are interested in dense waters that subduct to the Antarctic continental shelf <1000 m depth. You may want to use $\\sigma_0$, $\\sigma_2$ or something else. If so, alter the `gsw` function used, and define a more appropriate `isopycnal_bins` array.\n",
"- If you're looking at a non-interval number of years, the time averaging scheme will need to be modified.\n",
"- If you're online computed temperatures (SST) are potential temp (as opposed to conservative temp), use the gsw function `CT_from_pt` before the `sigma1` function as `sigma1` takes conservative temperature input (more info on the TEOS-10 gsw functions [here](http://www.teos-10.org/pubs/gsw/v3_04/html/gsw_contents.html)). This is the case for older ACCESS-OM2-01 runs (e.g. spinup6).\n",
"- On this `dask` setup and with the memory constraints of the VDI we don't think you would be able to do much more than a 5 year average of 0.1$^\\circ$ data using this function. You are memory-limited by the numpy TEOS-10 gsw functions. If you need to average over more years, you mey have to do the TEOS-10 gsw translations separately and save these prior to computing transformation rates."
]
},
{
"cell_type": "markdown",
"metadata": {},
"source": [
"When analysing 0.1$^\\circ$ output we tend to compute + save and reload surface water-mass transformation diagnostics in separate steps. In part this is because the computations take a while (~13min for 5 years of data below), but we also find that there's some instability (my workers are always killed) if we instead `return net_transformation, heat_transformation, salt_transformation` and then manipulate these variables straight away (a mystery for another mind). Saving and reloading works fine however."
]
},
{
"cell_type": "code",
"execution_count": 4,
"metadata": {},
"outputs": [
{
"data": {
"text/html": [
"