Comparing Directional Data (e.g. wind direction)¶

Comparing directional data is easy from version 1.0 if the quantity is defined as directional. This happens automatically if data is loaded from a dfs file with EUM unit in "degrees" or if loaded from a xarray dataset with attribute "units" set to "degrees". The quantity can also be created as directional manually by ms.Quantity(..., is_directional=True).

In the below example, the EUM unit is "degrees".

In [1]:

import modelskill as ms
import mikeio

In [2]:

fn = "../tests/testdata/wave_dir.dfs0"
ds = mikeio.read(fn)
ds

Out[2]:

<mikeio.Dataset>
dims: (time:9026)
time: 2006-12-20 22:00:00 - 2007-12-31 23:00:00 (9026 records)
geometry: GeometryUndefined()
items:
  0:  China_Model: Sign. Wave Height <Significant wave height> (meter)
  1:  China_Measured: Sign. Wave Height <Significant wave height> (meter)
  2:  China_Model: Mean Wave Direction <Mean Wave Direction> (degree)
  3:  China_Measured: Mean Wave Direction <Mean Wave Direction> (degree)

In [3]:

cmp = ms.from_matched(ds, obs_item=3, mod_items=[2])
cmp

Out[3]:

<Comparer>
Quantity: Mean Wave Direction [degree]
Observation: China_Measured: Mean Wave Direction, n_points=9026
 Model: China_Model: Mean Wave Direction, rmse=71.358

In [4]:

cmp.quantity

Out[4]:

Quantity(name='Mean Wave Direction', unit='degree', is_directional=True)

Circular metrics are used to compare directional data if the quantity is defined as directional.

In [7]:

cmp.skill(metrics=["c_rmse","c_max_error"]).round(1)

Out[7]:

	n	c_rmse	c_max_error
observation
China_Measured: Mean Wave Direction	9026	60.7	179.9

In [8]:

cmp.plot.timeseries();

In [9]:

cmp.plot.kde();   # note: the KDE estimate is not directional! (yet)

In [10]:

cmp.plot.scatter();  # note: regression line and Q-Q are not shown 