def read(dictArgs):
""" read data from model and obs files, process data and return it """
dsmodel = xr.open_mfdataset(dictArgs["infile"],
combine="by_coords",
decode_times=False)
if dictArgs["obsfile"] is not None:
# priority to user-provided obs file
dsobs = xr.open_mfdataset(dictArgs["obsfile"],
combine="by_coords",
decode_times=False)
else:
# use dataset from catalog, either from command line or default
cat_platform = "catalogs/obs_catalog_" + dictArgs["platform"] + ".yml"
catfile = pkgr.resource_filename("om4labs", cat_platform)
cat = intake.open_catalog(catfile)
dsobs = cat[dictArgs["dataset"]].to_dask()
# read in model and obs data
datamodel = read_data(dsmodel, dictArgs["possible_variable_names"])
dataobs = read_data(dsobs, dictArgs["possible_variable_names"])
# subset data
if dictArgs["depth"] is None:
dictArgs["depth"] = dictArgs["surface_default_depth"]
if dictArgs["depth"] is not None:
datamodel = subset_data(datamodel, "assigned_depth", dictArgs["depth"])
dataobs = subset_data(dataobs, "assigned_depth", dictArgs["depth"])
# reduce data along depth (not yet implemented)
if "depth_reduce" in dictArgs:
if dictArgs["depth_reduce"] == "mean":
# do mean
pass
elif dictArgs["depth_reduce"] == "sum":
# do sum
pass
# reduce data along time, here mandatory
if ("assigned_time"
in datamodel.dims) and (len(datamodel["assigned_time"]) > 1):
warnings.warn("input dataset has more than one time record, " +
"performing non-weighted average")
datamodel = simple_average(datamodel, "assigned_time")
if ("assigned_time" in dataobs.dims) and len(dataobs["assigned_time"]) > 1:
warnings.warn("reference dataset has more than one time record, " +
"performing non-weighted average")
dataobs = simple_average(dataobs, "assigned_time")
datamodel = datamodel.squeeze()
dataobs = dataobs.squeeze()
# check final data is 2d
assert len(datamodel.dims) == 2
assert len(dataobs.dims) == 2
# check consistency of coordinates
assert np.allclose(datamodel["assigned_lon"], dataobs["assigned_lon"])
assert np.allclose(datamodel["assigned_lat"], dataobs["assigned_lat"])
# homogeneize coords
dataobs = copy_coordinates(datamodel, dataobs,
["assigned_lon", "assigned_lat"])
# restrict model to where obs exists
datamodel = datamodel.where(dataobs)
# dump values
model = datamodel.to_masked_array()
obs = dataobs.to_masked_array()
x = datamodel["assigned_lon"].values
y = datamodel["assigned_lat"].values
# compute area
if "areacello" in dsmodel.variables:
area = dsmodel["areacello"].values
else:
if model.shape == (180, 360):
area = compute_area_regular_grid(dsmodel)
else:
raise IOError("no cell area provided")
return x, y, area, model, obs
def read(dictArgs):
""" read data from model and obs files, process data and return it """
if dictArgs["config"] is not None:
# use dataset from catalog, either from command line or default
cat = open_intake_catalog(dictArgs["platform"], dictArgs["config"])
ds_static = cat["ocean_static_1x1"].to_dask()
if dictArgs["static"] is not None:
ds_static = xr.open_dataset(dictArgs["static"])
# Compute basin codes
codes = generate_basin_codes(ds_static, lon="lon", lat="lat")
codes = np.array(codes)
# depth coordinate
if "deptho" in list(ds_static.variables):
depth = ds_static.deptho.to_masked_array()
elif "depth" in list(ds_static.variables):
depth = ds_static.depth.to_masked_array()
else:
raise ValueError("Unable to find depth field.")
depth = np.where(np.isnan(depth), 0.0, depth)
depth = depth * -1.0
dsmodel = xr.open_mfdataset(dictArgs["infile"],
combine="by_coords",
use_cftime=True)
if dictArgs["obsfile"] is not None:
# priority to user-provided obs file
dsobs = xr.open_mfdataset(dictArgs["obsfile"],
combine="by_coords",
decode_times=False)
else:
# use dataset from catalog, either from command line or default
cat = open_intake_catalog(dictArgs["platform"], "obs")
dsobs = cat[dictArgs["dataset"]].to_dask()
# read in model and obs data
datamodel = read_data(dsmodel, dictArgs["possible_variable_names"])
dataobs = read_data(dsobs, dictArgs["possible_variable_names"])
# reduce data along time, here mandatory
if ("assigned_time"
in datamodel.dims) and (len(datamodel["assigned_time"]) > 1):
warnings.warn("input dataset has more than one time record, " +
"performing non-weighted average")
datamodel = simple_average(datamodel, "assigned_time")
if ("assigned_time" in dataobs.dims) and len(dataobs["assigned_time"]) > 1:
warnings.warn("reference dataset has more than one time record, " +
"performing non-weighted average")
dataobs = simple_average(dataobs, "assigned_time")
datamodel = datamodel.squeeze()
dataobs = dataobs.squeeze()
# check final data is 3d
assert len(datamodel.dims) == 3
assert len(dataobs.dims) == 3
# check consistency of coordinates
assert np.allclose(datamodel["assigned_lon"], dataobs["assigned_lon"])
assert np.allclose(datamodel["assigned_lat"], dataobs["assigned_lat"])
# homogeneize coords
dataobs = copy_coordinates(datamodel, dataobs,
["assigned_lon", "assigned_lat"])
# restrict model to where obs exists
datamodel = datamodel.where(dataobs)
# dump values
model = datamodel.to_masked_array()
obs = dataobs.to_masked_array()
x = datamodel["assigned_lon"].values
y = datamodel["assigned_lat"].values
z = datamodel["assigned_depth"].values
# convert z to negative values
z = z * -1
# compute area
if "areacello" in dsmodel.variables:
area = dsmodel["areacello"].values
else:
if (model.shape[-2], model.shape[-1]) == (180, 360):
area = compute_area_regular_grid(dsmodel)
else:
raise IOError("no cell area provided")
# date range
dates = date_range(dsmodel)
return y, z, depth, area, codes, model, obs, dates
def test_copy_coordinates():
from om4labs.om4common import copy_coordinates
da = xr.DataArray(data=np.empty((360, 2)), dims=("lon", "bnds"))
da = copy_coordinates(grid_1x1["lon_bnds"], da, ["lon", "bnds"])
assert list(da.coords).sort() == list(grid_1x1["lon_bnds"].coords).sort()