Cartopy: Errors in coastline filling in Lambert Conformal conic projection

[dennissergeev]

I noticed that coastlines are not filled correctly when I use Lambert conic projection. For example,

import cartopy
import cartopy.crs as ccrs
import matplotlib.pyplot as plt

fig = plt.figure()                                                

ax = fig.add_subplot(111, projection=ccrs.LambertConformal())                         
feature = cartopy.feature.NaturalEarthFeature(name='coastline', category='physical',
                                              scale='110m',
                                              edgecolor='#000000', facecolor='#AAAAAA')
ax.add_feature(feature)
plt.show()

results in

It happens at all coastline scales and with different extents of the map, so I guess it's a bug. Maybe it's related to #579

[dopplershift]

I'm getting the same problem with the oceans on master right now:

proj = cartopy.crs.LambertConformal(central_latitude=25, central_longitude=-97,
                                    standard_parallels=[25])
ax = plt.axes(projection=proj)
ax.add_feature(cartopy.feature.OCEAN)
ax.coastlines()
ax.set_extent([-140, -70, 20, 60])

Does not occur if we use PlateCaree() instead:

[dopplershift]

Also works fine if we use standard_parallels=[35] for the Lambert projection.

[akrherz]

May I add LambertAzimuthalEqualArea to the list or reproducers with current master + matplotlib2

proj = cartopy.crs.LambertAzimuthalEqualArea(central_latitude=45,
                                             central_longitude=-100)
ax = plt.axes(projection=proj)
ax.add_feature(cartopy.feature.OCEAN)
ax.coastlines()
ax.set_extent([-140, -70, 20, 60])

[byersiiasa]

I have just encountered a similar issue, with the same script on 2 different machines, but between using Python 3.6 (error) and Python 2.7: both with Cartopy 0.15.1
ax.add_feature(cfeature.BORDERS, linewidth=0.3, color='#778899')

NB however it may due to some package differences:
For example:
matplotlib = 1.5.3 on P2.7
matplotlib = 2.1.0 on P3.6

Python 2.7

Python 3.6

I have limited use rights so I'm afraid its not easy for me to go installing/switching packages etc to test further

[kaedonkers]

A user posted a question on Stack Overflow with the same issue of adding oceans to LambertConformal:
https://stackoverflow.com/questions/48503737/cartopy-oceans-cover-land

User demonstrated that it was a problem with the projection, not the features they were adding.

[byersiiasa]

@kaedonkers @dopplershift
In my case i wasn't using Lambert. Either Robinson or PlateCarree. So either it is a different issue, or something that extends beyond just the Lambert projections.

[byersiiasa]

I have just tested the following code:

import cartopy.feature as cfeature  
import cartopy.crs as ccrs  
import matplotlib.pyplot as plt  
fig = plt.figure()  
ax = plt.axes(projection= ccrs.PlateCarree())  
ax.set_global()  
ax.coastlines( linewidth=0.3, color='#778899')  
ax.add_feature(cfeature.BORDERS, linewidth=0.3, color='#778899')

Python 3.6 (Anaconda) (2 different machines)
Matplotlib 2.1.2, Cartopy 0.15.1

Python 2.7 (Anaconda)
Matplotlib 2.1.2 (and Matplotlib 1.5.3); Cartopy 0.15.1

[QuLogic]

@byersiiasa your difference between 2 and 3 is likely because of keyword argument & dictionary ordering. If you pass a more explicit facecolor='#778899', then both 2 and 3 show weird looking patches. If you pass edgecolor='#778899', then both 2 and 3 look fine. I assume you'd like the latter. I'm not so sure this is really related to this issue as I don't think the BORDERS feature is intended to be filled.

Edit: that inconsistency is fixed by #1029.

[byersiiasa]

Thank you @QuLogic ! That was exactly the issue! Perfect timing.

[QuLogic]

So I've spent a while looking into this and the 'bug' is that coastlines are _not_ intended to be filled. They appear to be a bunch of line strings only with no polygonal topology. I tried several methods to turn them into polygons, but they are simply not designed for it. If you just convert blindly, several line strings are not long enough, and others are only half a continent (Asia is split in a North and South-ish half).

The best way to achieve OP's desired result is to add coastlines with no fill and land with fill:

import cartopy.feature as cfeature
import cartopy.crs as ccrs
import matplotlib.pyplot as plt

fig = plt.figure()
ax = fig.add_subplot(111, projection=ccrs.LambertConformal())

feature = cfeature.NaturalEarthFeature(
    name='coastline', category='physical',
    scale='110m',
    edgecolor='#000000', facecolor='none')
ax.add_feature(feature)

feature = cfeature.NaturalEarthFeature(
    name='land', category='physical',
    scale='110m',
    facecolor='#AAAAAA')
ax.add_feature(feature)

plt.show()

or using builtin features:

import cartopy.feature as cfeature
import cartopy.crs as ccrs
import matplotlib.pyplot as plt

fig = plt.figure()
ax = fig.add_subplot(111, projection=ccrs.LambertConformal())

ax.add_feature(cfeature.COASTLINES, edgecolor='#000000')
ax.add_feature(cfeature.LAND, facecolor='#AAAAAA')

plt.show()

I'm tempted to close this "won't fix", or rather "can't fix", because this is simply the way the Natural Earth dataset is defined and there's not much we can do to clean it up automatically. If you want to pursue cleaning that up, that would be best done with them https://github.com/nvkelso/natural-earth-vector/

[QuLogic]

As for the bug in the first comment, https://github.com/SciTools/cartopy/issues/803#issuecomment-272286945, and third comment, https://github.com/SciTools/cartopy/issues/803#issuecomment-333909742, the OCEANS _are_ true polygons. The first case is a symptom of the same problem as #1131 and is fixed by #1146. I'm not sure about the second one just yet, but I'm going to open it as a separate issue.

Since the original report is not technically a bug (as above), I'm going to close this issue.

[QuLogic]

Oh, also the Stack Overflow post is the same as #1131 as well, and is fixed by #1146.

[dennissergeev]

Thanks, @QuLogic!