A spatial join uses binary predicates such as intersects and crosses to combine two GeoDataFrames based on the spatial relationship between their geometries.
A common use case might be a spatial join between a point layer and a polygon layer where you want to retain the point geometries and grab the attributes of the intersecting polygons.
From https://github.com/geopandas/geopandas/blob/master/examples/spatial_joins.ipynb. Re-ran and tweaked by Emilio Mayorga, 12/10/2015, on my geopandas conda environment.
In [1]:
from IPython.core.display import Image
Image(url="https://dl.dropboxusercontent.com/u/6769420/sjoin_test.png")
Out[1]:
We currently support the following methods of spatial joins. We refer to the left_df and right_df which are the correspond to the two dataframes passed in as args.
In a LEFT OUTER JOIN (how='left'), we keep all rows from the left and duplicate them if necessary to represent multiple hits between the two dataframes. We retain attributes of the right if they intersect and lose right rows that don't intersect. A left outer join implies that we are interested in retaining the geometries of the left.
This is equivalent to the PostGIS query:
SELECT pts.geom, pts.id as ptid, polys.id as polyid
FROM pts
LEFT OUTER JOIN polys
ON ST_Intersects(pts.geom, polys.geom);
geom | ptid | polyid
--------------------------------------------+------+--------
010100000040A9FBF2D88AD03F349CD47D796CE9BF | 4 | 10
010100000048EABE3CB622D8BFA8FBF2D88AA0E9BF | 3 | 10
010100000048EABE3CB622D8BFA8FBF2D88AA0E9BF | 3 | 20
0101000000F0D88AA0E1A4EEBF7052F7E5B115E9BF | 2 | 20
0101000000818693BA2F8FF7BF4ADD97C75604E9BF | 1 |
(5 rows)In a RIGHT OUTER JOIN (how='right'), we keep all rows from the right and duplicate them if necessary to represent multiple hits between the two dataframes. We retain attributes of the left if they intersect and lose left rows that don't intersect. A right outer join implies that we are interested in retaining the geometries of the right.
This is equivalent to the PostGIS query:
SELECT polys.geom, pts.id as ptid, polys.id as polyid
FROM pts
RIGHT OUTER JOIN polys
ON ST_Intersects(pts.geom, polys.geom);
geom | ptid | polyid
----------+------+--------
01...9BF | 4 | 10
01...9BF | 3 | 10
02...7BF | 3 | 20
02...7BF | 2 | 20
00...5BF | | 30
(5 rows)In an INNER JOIN (how='inner'), we keep rows from the right and left only where their binary predicate is True. We duplicate them if necessary to represent multiple hits between the two dataframes. We retain attributes of the right and left only if they intersect and lose all rows that do not. An inner join implies that we are interested in retaining the geometries of the left.
This is equivalent to the PostGIS query:
SELECT pts.geom, pts.id as ptid, polys.id as polyid
FROM pts
INNER JOIN polys
ON ST_Intersects(pts.geom, polys.geom);
geom | ptid | polyid
--------------------------------------------+------+--------
010100000040A9FBF2D88AD03F349CD47D796CE9BF | 4 | 10
010100000048EABE3CB622D8BFA8FBF2D88AA0E9BF | 3 | 10
010100000048EABE3CB622D8BFA8FBF2D88AA0E9BF | 3 | 20
0101000000F0D88AA0E1A4EEBF7052F7E5B115E9BF | 2 | 20
(4 rows)Downloaded sample NYC Boroughs shape file from http://www.nyc.gov/html/dcp/download/bytes/nybb_14aav.zip
In [2]:
%matplotlib inline
import os
from shapely.geometry import Point
from geopandas import GeoDataFrame, read_file
from geopandas.tools import overlay, sjoin
# NYC Boros
#zippath = os.path.abspath('../examples/nybb_14aav.zip')
zippath = os.path.abspath('/home/mayorga/Desktop/MySciPythonApps/nybb_14aav.zip')
polydf = read_file('/nybb_14a_av/nybb.shp', vfs='zip://' + zippath)
# Generate some points
b = [int(x) for x in polydf.total_bounds]
N = 8
pointdf = GeoDataFrame([
{'geometry' : Point(x, y), 'value1': x + y, 'value2': x - y}
for x, y in zip(range(b[0], b[2], int((b[2]-b[0])/N)),
range(b[1], b[3], int((b[3]-b[1])/N)))])
In [3]:
pointdf
Out[3]:
In [4]:
pointdf.plot();
In [5]:
polydf
Out[5]:
In [6]:
polydf.plot();
In [7]:
join_left_df = sjoin(pointdf, polydf, how="left")
join_left_df
# Note the NaNs where the point did not intersect a boro
Out[7]:
In [8]:
join_left_df.plot();
In [9]:
join_right_df = sjoin(pointdf, polydf, how="right")
join_right_df
# Note Staten Island is repeated
Out[9]:
In [10]:
join_right_df.plot();
In [11]:
join_inner_df = sjoin(pointdf, polydf, how="inner")
join_inner_df
# Note the lack of NaNs; dropped anything that didn't intersect
Out[11]:
In [12]:
join_inner_df.plot();
We're not limited to using the intersection binary predicate. Any of the Shapely geometry methods that return a Boolean can be used by specifying the op kwarg.
In [13]:
sjoin(pointdf, polydf, how="left", op="within")
Out[13]:
In [ ]: