diff --git a/metpy/plots/tests/test_skewt.py b/metpy/plots/tests/test_skewt.py index 1bb2ea66984..bac12244d67 100644 --- a/metpy/plots/tests/test_skewt.py +++ b/metpy/plots/tests/test_skewt.py @@ -17,7 +17,7 @@ MPL_VERSION = matplotlib.__version__[:3] -@pytest.mark.mpl_image_compare(tolerance=0.021, remove_text=True) +@pytest.mark.mpl_image_compare(tolerance=0.224, remove_text=True) def test_skewt_api(): """Test the SkewT API.""" with matplotlib.rc_context({'axes.autolimit_mode': 'data'}): @@ -87,7 +87,8 @@ def test_profile(): return np.linspace(1000, 100, 10), np.linspace(20, -20, 10), np.linspace(25, -30, 10) -@pytest.mark.mpl_image_compare(tolerance={'2.0': 1.12}.get(MPL_VERSION, 0.), remove_text=True) +@pytest.mark.mpl_image_compare(tolerance={'2.0': 1.12}.get(MPL_VERSION, 0.2432), + remove_text=True) def test_skewt_shade_cape_cin(test_profile): """Test shading CAPE and CIN on a SkewT plot.""" p, t, tp = test_profile @@ -104,7 +105,8 @@ def test_skewt_shade_cape_cin(test_profile): return fig -@pytest.mark.mpl_image_compare(tolerance={'1.4': 1.70}.get(MPL_VERSION, 0.), remove_text=True) +@pytest.mark.mpl_image_compare(tolerance={'1.4': 1.70}.get(MPL_VERSION, 0.2432), + remove_text=True) def test_skewt_shade_area(test_profile): """Test shading areas on a SkewT plot.""" p, t, tp = test_profile @@ -131,7 +133,8 @@ def test_skewt_shade_area_invalid(test_profile): skew.shade_area(p, t, tp, which='positve') -@pytest.mark.mpl_image_compare(tolerance={'1.4': 1.75}.get(MPL_VERSION, 0.), remove_text=True) +@pytest.mark.mpl_image_compare(tolerance={'1.4': 1.75}.get(MPL_VERSION, 0.2432), + remove_text=True) def test_skewt_shade_area_kwargs(test_profile): """Test shading areas on a SkewT plot with kwargs.""" p, t, tp = test_profile @@ -208,7 +211,7 @@ def test_skewt_barb_color(): return fig -@pytest.mark.mpl_image_compare(tolerance={'2.0': 0.2}.get(MPL_VERSION, 0), remove_text=True) +@pytest.mark.mpl_image_compare(tolerance=0.2, remove_text=True) def test_skewt_barb_unit_conversion(): """Test that barbs units can be converted at plot time (#737).""" u_wind = np.array([3.63767155210412]) * units('m/s')