def process(
forecast: cli.inputcube,
reliability_table: cli.inputcubelist = None,
):
"""
Calibrate a probability forecast using the provided reliability calibration
table. This calibration is designed to improve the reliability of
probability forecasts without significantly degrading their resolution. If
a reliability table is not provided, the input forecast is returned
unchanged.
The method implemented here is described in Flowerdew J. 2014. Calibrating
ensemble reliability whilst preserving spatial structure. Tellus, Ser. A
Dyn. Meteorol. Oceanogr. 66.
Args:
forecast (iris.cube.Cube):
The forecast to be calibrated.
reliability_table (iris.cube.Cube or iris.cube.CubeList):
The reliability calibration table to use in calibrating the
forecast. If input is a CubeList the CubeList should contain
separate cubes for each threshold in the forecast cube.
Returns:
iris.cube.Cube:
Calibrated forecast.
"""
from improver.calibration.reliability_calibration import ApplyReliabilityCalibration
if reliability_table is None:
return forecast
plugin = ApplyReliabilityCalibration()
return plugin(forecast, reliability_table)
def setUp(self):
"""Set up data for testing the interpolate method."""
self.reliability_probabilities = np.array([0.0, 0.4, 0.8])
self.observation_frequencies = np.array([0.2, 0.6, 1.0])
self.plugin = Plugin()
def test_values(self):
"""Test expected values are returned when two or more bins are
available for interpolation."""
expected_0 = (
np.array([0.0, 0.25, 0.5, 0.75, 1.0]),
np.array([0.0, 0.0, 0.25, 0.5, 0.75]),
)
expected_1 = (
np.array([0.0, 0.25, 0.5, 0.75, 1.0]),
np.array([0.25, 0.5, 0.75, 1.0, 1.0]),
)
plugin = Plugin()
threshold_0 = plugin._calculate_reliability_probabilities(
self.reliability_cube[0])
threshold_1 = plugin._calculate_reliability_probabilities(
self.reliability_cube[1])
assert_array_equal(threshold_0, expected_0)
assert_array_equal(threshold_1, expected_1)
def process(
forecast: cli.inputcube,
reliability_table: cli.inputcube = None,
*,
minimum_forecast_count=200,
):
"""
Calibrate a probability forecast using the provided reliability calibration
table. This calibration is designed to improve the reliability of
probability forecasts without significantly degrading their resolution. If
a reliability table is not provided, the input forecast is returned
unchanged.
The method implemented here is described in Flowerdew J. 2014. Calibrating
ensemble reliability whilst preserving spatial structure. Tellus, Ser. A
Dyn. Meteorol. Oceanogr. 66.
Args:
forecast (iris.cube.Cube):
The forecast to be calibrated.
reliability_table (iris.cube.Cube):
The reliability calibration table to use in calibrating the
forecast.
minimum_forecast_count (int):
The minimum number of forecast counts in a forecast probability
bin for it to be used in calibration. If the reliability
table for a forecast threshold includes any bins with
insufficient counts that threshold will be returned unchanged.
The default value of 200 is that used in Flowerdew 2014.
Returns:
iris.cube.Cube:
Calibrated forecast.
"""
from improver.calibration.reliability_calibration import ApplyReliabilityCalibration
if reliability_table is None:
return forecast
plugin = ApplyReliabilityCalibration(
minimum_forecast_count=minimum_forecast_count)
return plugin(forecast, reliability_table)
class Test__interpolate(unittest.TestCase):
"""Test the _interpolate method."""
def setUp(self):
"""Set up data for testing the interpolate method."""
self.reliability_probabilities = np.array([0.0, 0.4, 0.8])
self.observation_frequencies = np.array([0.2, 0.6, 1.0])
self.plugin = Plugin()
def test_unmasked_data(self):
"""Test unmasked data is interpolated and returned as expected."""
expected = np.array([0.4, 0.6, 0.8])
forecast_threshold = np.array([0.2, 0.4, 0.6])
result = self.plugin._interpolate(
forecast_threshold,
self.reliability_probabilities,
self.observation_frequencies,
)
assert_allclose(result, expected)
def test_masked_data(self):
"""Test masked data is interpolated and returned with the original
mask in place."""
expected = np.ma.masked_array([np.nan, 0.6, 0.8], mask=[1, 0, 0])
forecast_threshold = np.ma.masked_array([np.nan, 0.4, 0.6],
mask=[1, 0, 0])
result = self.plugin._interpolate(
forecast_threshold,
self.reliability_probabilities,
self.observation_frequencies,
)
assert_allclose(result, expected)
def test_clipping(self):
"""Test the result, when using data constructed to cause extrapolation
to a probability outside the range 0 to 1, is clipped. In this case
an input probability of 0.9 would return a calibrated probability
of 1.1 in the absence of clipping."""
expected = np.array([0.4, 0.6, 1.0])
forecast_threshold = np.array([0.2, 0.4, 0.9])
result = self.plugin._interpolate(
forecast_threshold,
self.reliability_probabilities,
self.observation_frequencies,
)
assert_allclose(result, expected)
def test_reshaping(self):
"""Test that the result has the same shape as the forecast_threshold
input data."""
expected = np.array([[0.2, 0.325, 0.45], [0.575, 0.7, 0.825],
[0.95, 1.0, 1.0]])
forecast_threshold = np.linspace(0, 1, 9).reshape((3, 3))
result = self.plugin._interpolate(
forecast_threshold,
self.reliability_probabilities,
self.observation_frequencies,
)
self.assertEqual(result.shape, expected.shape)
assert_allclose(result, expected)