Postgresql 优化PostGIS查询，ST_相交

我正在尝试在两个表之间运行空间查询。表1（prism_ppt_monthly-详情见下文）为月度降水数据。表2（美国地质勘探局盆地边界-详见下文）为水文盆地边界的多边形

我想为每个流域创建一个总降水量的时间序列。我有一个查询可以做到这一点（详情见下文），但一次计算需要将近4.75秒。考虑到我有1440个月的降水数据和近40个流域，这个查询需要：4.75秒*1440*40=77小时

下面是有关查询和表的信息。我在每个表上都有空间索引（gist），并对每个表进行了真空分析。任何关于我如何能够加快这件事的想法都将不胜感激

查询：

EXPLAIN ANALYZE
SELECT filename,date_from,date_to,site_no,sqmi,(ST_SummaryStats(rast)).* FROM prism_ppt_monthly, usgs_basin_boundary WHERE ST_Intersects(rast,ST_Transform(geom,4269)) LIMIT 1;
                                                                             QUERY PLAN                                                                         
    ------------------------------------------------------------------------------------------------------------------------------------------------------------
     Limit  (cost=65964.53..66000.10 rows=1 width=81) (actual time=4764.969..4764.972 rows=1 loops=1)
       ->  Nested Loop  (cost=65964.53..66782.60 rows=23 width=81) (actual time=4764.963..4764.963 rows=1 loops=1)
             Join Filter: _st_intersects(st_transform(usgs_basin_boundary.geom, 4269), prism_ppt_monthly.rast, NULL::integer)
             ->  Hash Semi Join  (cost=65964.53..66610.73 rows=47 width=126256) (actual time=4587.961..4587.961 rows=1 loops=1)
                   Hash Cond: ((usgs_basin_boundary.site_no)::text = df_flow.code)
                   ->  Seq Scan on usgs_basin_boundary  (cost=0.00..639.09 rows=2509 width=126256) (actual time=0.007..1.279 rows=595 loops=1)
                   ->  Hash  (cost=65963.94..65963.94 rows=47 width=9) (actual time=4585.313..4585.313 rows=47 loops=1)
                         Buckets: 1024  Batches: 1  Memory Usage: 2kB
                         ->  HashAggregate  (cost=65963.00..65963.47 rows=47 width=9) (actual time=4585.126..4585.215 rows=47 loops=1)
                               ->  Seq Scan on df_flow  (cost=0.00..63244.20 rows=1087520 width=9) (actual time=5.826..2367.593 rows=1087520 loops=1)
             ->  Index Scan using prism_ppt_monthly_rast_gist on prism_ppt_monthly  (cost=0.00..0.40 rows=1 width=64) (actual time=0.034..0.034 rows=1 loops=1)
                   Index Cond: ((rast)::geometry && st_transform(usgs_basin_boundary.geom, 4269))
     Total runtime: 4765.151 ms

表1：

\d+ prism_ppt_monthly
                                            Table "public.prism_ppt_monthly"
      Column   |  Type   |                            Modifiers                            | Storage  | Description 
    -----------+---------+-----------------------------------------------------------------+----------+-------------
     rid       | integer | not null default nextval('prism_ppt_monthly_rid_seq'::regclass) | plain    | 
     rast      | raster  |                                                                 | extended | 
     filename  | text    |                                                                 | extended | 
     date_from | date    |                                                                 | plain    | 
     date_to   | date    |                                                                 | plain    | 
    Indexes:
        "prism_ppt_monthly_pkey" PRIMARY KEY, btree (rid)
        "prism_ppt_monthly_rast_gist" gist (st_convexhull(rast))
    Check constraints:
        "enforce_height_rast" CHECK (st_height(rast) = 621)
        "enforce_max_extent_rast" CHECK (st_coveredby(st_convexhull(rast), '0103000020AD10000001000000050000005555555555415FC01E01000000F8484060A9AAAAAA9E50C01E01000000F8484060A9AAAAAA9E50C0F5FFFFFFFF0F38405555555555415FC0F5FFFFFFFF0F38405555555555415FC01E01000000F84840'::geometry))
        "enforce_nodata_values_rast" CHECK (_raster_constraint_nodata_values(rast)::numeric(16,10)[] = '{-9999}'::numeric(16,10)[])
        "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) = '{32BF}'::text[])
        "enforce_same_alignment_rast" CHECK (st_samealignment(rast, '0100000000365755555555A53F365755555555A5BF5555555555415FC01E01000000F8484000000000000000000000000000000000AD10000001000100'::raster))
        "enforce_scalex_rast" CHECK (st_scalex(rast)::numeric(16,10) = 0.04166666666667::numeric(16,10))
        "enforce_scaley_rast" CHECK (st_scaley(rast)::numeric(16,10) = (-0.04166666666667)::numeric(16,10))
        "enforce_srid_rast" CHECK (st_srid(rast) = 4269)
        "enforce_width_rast" CHECK (st_width(rast) = 1405)
    Has OIDs: no

表2：

\d+ usgs_basin_boundary
                                                 Table "public.usgs_basin_boundary"
  Column  |            Type             |                             Modifiers                             | Storage  | Description 
----------+-----------------------------+-------------------------------------------------------------------+----------+-------------
 gid      | integer                     | not null default nextval('usgs_basin_boundary_gid_seq'::regclass) | plain    | 
 site_no  | character varying(15)       |                                                                   | extended | 
 sqmi     | numeric                     |                                                                   | main     | 
 abs_diff | numeric                     |                                                                   | main     | 
 geom     | geometry(MultiPolygon,5070) |                                                                   | main     | 
Indexes:
    "usgs_basin_boundary_pkey" PRIMARY KEY, btree (gid)
    "usgs_basin_boundary_shape_gist" gist (geom)
Has OIDs: no

---

未使用美国地质勘探局盆地边界上的索引。geom，因为您正在调用

ST_Transform（geom，4269）

您应该对转换结果创建索引（）

---

这并不能解决你所有的问题，但我只是在你的问题中偶然发现了一些东西：

SELECT (...) WHERE ST_Intersects(rast,ST_Transform(geom,4269)) LIMIT 1;

重要的是这一点：

ST_Transform(geom,4269)

<>你在你的查询中间投影一个几何体。虽然这当然是可能的，但这可能不是一个好的做法。您可以预先将'geom'投影到另一个表中的'SRID:4269'。之后，您只需从另一个表访问变换后的几何体

---

将geom转换为新表后，可能还需要在该表上创建索引。这可能会稍微提高性能。

您确定解释计划和查询匹配吗？我不知道df\U流是从哪里来的。

ST_Transform(geom,4269)