Scaling Domains

This demo notebook showcases some key functions for scaling coordinates. We often need to use semantic coordinate values that are actually useful for computation. So for example, lat/lon coordinates are often in the spherical domain. But for many physical problems, we need them in meters. Similarly with time coordinates: we often have np.datetime64 coordinate values but we need them in seconds.

Note: There are some more meaningful rescaling that involve the mean and variance (StandardScaler) or specific min/max values (MinMaxScaler) but this is an arbitrary transformation that isn’t specific to coordinate values.

import autoroot
import typing as tp
from dataclasses import dataclass
import numpy as np
import pandas as pd
import xarray as xr
import einops
import xarray_dataclasses as xrdataclass
from oceanbench._src.datasets.base import XRDAPatcher

%load_ext autoreload
%autoreload 2

file = "/gpfswork/rech/cli/uvo53rl/projects/jejeqx/data/natl60/NATL60-CJM165_GULFSTREAM_ssh_y2013.1y.nc"
!ls $file

da = xr.open_dataset(file, decode_times=False)
da["time"] = pd.to_datetime(da.time)
da = da.sortby("time")

Spatial Rescaling

Degrees to Meters

from oceanbench._src.geoprocessing.spatial import latlon_deg2m

da_scaled = latlon_deg2m(da, mean=False)
da_scaled

Temporal Rescaling

DateTime 2 Seconds

from oceanbench._src.geoprocessing.temporal import time_rescale
import pandas as pd

t0 = "2012-12-15"
freq_dt = 1
freq_unit = "D"


da_scale = time_rescale(da_scaled, freq_dt=freq_dt, freq_unit=freq_unit, t0=t0)

da_scale