[postgis-users] ST_Intersection - Slow with raster, fast with polygonized raster?

[Mike Treglia]

Hi all,

I was exploring raster/vector overlays, basically to calculate the area of
each of several landcover classes (from raster data) within an area or sets
of areas (delineated in polygon/multipolygon data). In the past, I had
ended up polygonizing the raster, making sure the polygonized layer had a
spatial index, and developing a query from there.

But I wanted to come back to the problem and see if I could do it in what
seemed like the more elegant way, using the raster directly without an
intermediate step of polygonizing it.  While I'm able to do it, at least on
a small sample area, it's at a huge performance cost.

I'm wondering if I'm just missing something I need to do to make the
raster-based version operate more efficiently. (For what it's worth,
rasters like this, are also slow to load into QGIS and such [though once
loaded, are fairly quick to render], so I wonder if that's related at all
and fixable.)

My test case is in New York City - I'm working with a six-inch resolution
landcover dataset for the entirety of NYC (
https://data.cityofnewyork.us/Environment/Land-Cover-Raster-Data-2017-6in-Resolution/he6d-2qns),
which I'm overlaying with a small subset of parcels in a small geographic
area. In this example, below, I'm just adding up the area for each
landcover class across the focal parcels for simplicity.  When using the
polygonized landcover data, it takes about 170ms. When using the raster
landcover data, it takes about 1 min 20 seconds.

Below are the queries, as well as the information about the raster table in
PostGIS, and the raster2pgsql code I used to load the layer into the
database. Curious to hear any thoughts from folks on the list. Of course if
there's anything else I can share that would be helpful, let me know

Thanks so much! Best,
Mike

Here are the queries I'm running:
1) Original Raster over Polygons- ~1 min 20 sec
---------
select
(foo.geomval).val,
sum(st_area((geomval).geom)) as area_ft2
from (
    select (st_intersection(a.rast, b.geom_2263)) as geomval
   from base_rasters.landcover6in_2017 a
   join scratch.smallsample b
   on st_intersects(a.rast, b.geom_2263)) as foo
group by val;

2) Polygonized Raster over Polygons ~0.170 ms
---------
select lc_2017_value,
sum(st_area(st_intersection(a.geom_2263, b.geom_2263))) as area_ft2
from base_rasters.landcover6in_2017_polygons a
join scratch.smallsample b
on st_intersects(a.geom_2263, b.geom_2263)
group by lc_2017_value;


Here's the information for the raster table. (I had added a second index in
hopes it might help but no, it didn't).
---------
                                Table "base_rasters.landcover6in_2017"
 Column |  Type   | Collation | Nullable |                           Default
--------+---------+-----------+----------+-------------------------------------------------------------
 rid    | integer |           | not null |
nextval('base_rasters.landcover6in_2017_rid_seq'::regclass)
 rast   | raster  |           |          |
Indexes:
    "landcover6in_2017_pkey" PRIMARY KEY, btree (rid)
    "landcover6in_2017_envidx" gist (st_envelope(rast))
    "landcover6in_2017_st_convexhull_idx" gist (st_convexhull(rast))
Check constraints:
    "enforce_height_rast" CHECK (st_height(rast) = ANY (ARRAY[128, 46]))
    "enforce_max_extent_rast" CHECK (st_envelope(rast) @
'0103000020D70800000100000005000000C3F528DC3DD72B41CCCCCCCCB46BFC40C3F528DC3DD72B413333333349B31041E27A14EE9E4930413333333349B31041E2
7A14EE9E493041CCCCCCCCB46BFC40C3F528DC3DD72B41CCCCCCCCB46BFC40'::geometry)
NOT VALID
    "enforce_nodata_values_rast" CHECK
(_raster_constraint_nodata_values(rast) = '{0.0000000000}'::numeric[])
    "enforce_num_bands_rast" CHECK (st_numbands(rast) = 1)
    "enforce_out_db_rast" CHECK (_raster_constraint_out_db(rast) =
'{f}'::boolean[])
    "enforce_pixel_types_rast" CHECK (_raster_constraint_pixel_types(rast)
= '{8BUI}'::text[])
    "enforce_same_alignment_rast" CHECK (st_samealignment(rast,
'0100000000000000000000E03F000000000000E0BFC3F528DCBDB92E413333333349B3104100000000000000000000000000000000D708000001000100'::
raster))
    "enforce_scalex_rast" CHECK (round(st_scalex(rast)::numeric, 10) =
round(0.5, 10))
    "enforce_scaley_rast" CHECK (round(st_scaley(rast)::numeric, 10) =
round(- 0.5, 10))
    "enforce_srid_rast" CHECK (st_srid(rast) = 2263)
    "enforce_width_rast" CHECK (st_width(rast) = 128)

Here's the raster2pgsql code I used to load the raster in the first place:
-------
raster2pgsql -s 2263 -d -C -t 128x128 -M -I -l 4,16
"NYC_2017_LiDAR_LandCover.tif" base_rasters.landcover6in_2017 | psql -h
localhost -U postgres -d nycdb -v ON_ERROR_STOP=1

Lastly - here's my PostGIS info (Running on Windows 10, 64 bit):
POSTGIS="3.1.1 3.1.1" [EXTENSION] PGSQL="130" GEOS="3.9.0-CAPI-1.14.1"
SFCGAL="1.3.8" PROJ="7.1.1" GDAL="GDAL 3.2.1, released 2020/12/29"
LIBXML="2.9.9" LIBJSON="0.12" LIBPROTOBUF="1.2.1" WAGYU="0.5.0 (Internal)"
TOPOLOGY RASTER
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.osgeo.org/pipermail/postgis-users/attachments/20220130/1da8d987/attachment.html>