summary: "This article explains how to batch import rasters into a Postgres database. It also covers the use of PostGIS functions to query these rasters and create new ones. These techniques are particularly useful for applications such as QGIS."
It's possible to download the entire SRTM (Shuttle Radar Topography Mission) satellite imagery dataset and insert it into a Postgres database for personal use. This can be helpful for if you need global elevation data for your analysis and don't want to be limited by third-parties APIs.
The SRTM is a near-global dataset of elevation data with a resolution of 1-arc-second (30m). More information is available from [USGS](https://www.usgs.gov/centers/eros/science/usgs-eros-archive-digital-elevation-shuttle-radar-topography-mission-srtm?qt-science_center_objects=0#qt-science_center_objects).
In this guide we will go through downloading the data, inserting it into a Postgres database with PostGIS, then querying the final result to create a DEM for any country or region.
This guide assumes you are using Linux (this may also apply to other Unix like systems such as MacOS) and have a Postgres database with the PostGIS extension installed. More information [here](https://postgis.net/documentation/getting_started/).
## Download data
We'll download the data from the [OpenTopography](https://portal.opentopography.org/raster?opentopoID=OTSRTM.082015.4326.1) AWS S3 bucket. You should create a free account with them before downloading. Also, remember to provide citations where necessary.
To download the data, first we need to install the [AWS](https://docs.aws.amazon.com/cli/latest/userguide/getting-started-install.html) CLI utility:
Next we can run the following command to download the rasters. This will take some time depending on your internet speed. The final size is about 127GB and took me about 1 hour to download on a 600MB connection.
-`-d` This drops an existing table if it already exists. It's useful to include this in case we need to run the command again.
-`-M` This runs `VACUUM ANALYZE` on the table after the raster tiles have been inserted into the database. This optimizes the table by reclaiming storage space and then collecting table statistics.
-`-C` This applies raster constraints to the table by using the `AddRasterConstraints` function in PostGIS. This adds several types of constraints to the table which ensures data integrity and enforces rules for consistency.
-`-I` Creates a spatial index (GiST) on the table.
-`-s` Assign SRID (Spatial Reference System Identifier) to the raster.
-`-t` Tile size. This splits the raster into tiles of the specified size. Each tile is a row in the final table.
-`-F` Creates a filename column. As the raster is split into many smaller tiles, this is useful to help us identify the original file the tile belonged to.
-` *.tif` Selects all of the .tif files in the directory.
-`dem.srtm` Defines the schema and name of the table. Remember to create the `dem` schema in the database before running the command.
This process will take a couple of hours. It could take longer if running over a network, so it'd be best to run this command on the same machine as the Postgres instance.
Once this process has completed, we can query the table to see how many tiles we have:
{{<highlightsql>}}
select count(*) from dem.srtm
{{</highlight>}}
{{<highlightplain-text>}}
count
---------
12009480
{{</highlight>}}
So we have just over 12 million 128x128 tiles in the database.
## Raster Clipping
Now we have global SRTM data loaded in our local database, we can extract any of the tiles for further analysis and access this using any application.
Here I'll demonstrate using the following PostGIS commands: `st_clip`, `st_intersects` and `st_union` to create a digital elevation table for singapore and access this from from within QGIS.
I already have countries vector data in my Postgres database that I extracted from OpenStreetMaps. You can download a pg_dump of this [here](/data/countries.sql) to insert into your database. Alternatively, I have a post [here](/blogs/import-osm-countries-data) that explains the process to do this manually.
To create our table, we'll use the following SQL. I'll break down each section step-by-step.
First we'll use the `st_intersects` command to select all of the raster tiles that intersect with the Singapore. Note, we'll need to create a table, spatial index then apply constraints for this to work in QGIS.
{{<highlightsql>}}
-- drop existing table and create a new table
drop table if exists dem.singapore_srtm;
create table dem.singapore_srtm as
-- cte to create polygon of singapore
with country as (select geom as geom from dev.countries where country = 'Singapore')
-- select tiles that intersect with the Singapore polygon using a left join and st_intersects function
select srtm.rast, srtm.rid
from dem.srtm srtm
join country ply on st_intersects(ply.geom, st_convexhull(srtm.rast));
-- create index and apply constraints
create index on "dem"."singapore_srtm" using gist (st_convexhull("rast"));
Hmm, this doesn't look quite right! As you can see, the clipped tiles only seem to apply within the tile boundaries outside of the clip region. These areas have been set to `nodata` (which is expected). However, outside of that, QGIS is rendering those pixels as something else (it's rendering them as having a value of 0).
The reason for this is because we are not strictly importing a single raster into QGIS, rather we are still importing all of the indivitual tiles. Since GeoTIFFs are rasters that can only exist as a grid of pixels (i.e. a rectangle), QGIS must therefore render the entire raster extent, which leads to the unusual rendering issue above.
We can solve this by using the `st_union` function in our query. This takes all of the individual raster tiles and unions them into a single raster. All of the `nodata` values for the extent of the clipped region will also be unioned, thus creating a single rectangular raster that we can import into QGIS.
Our final query will look like this:
{{<highlightsql>}}
-- drop existing table and create a new table
drop table if exists dem.singapore_srtm;
create table dem.singapore_srtm as
-- cte to create polygon of singapore
with country as (select geom as geom from dev.countries where country = 'Singapore'),
-- cte to clip tiles to the singapore polygon
clipped_tiles as (
select st_clip(srtm.rast, ply.geom, true) as rast, srtm.rid
from dem.srtm srtm
join country ply on st_intersects(ply.geom, st_convexhull(srtm.rast))
)
-- union tiles into a single raster
select st_union(rast) as rast
from clipped_tiles;
-- create index and apply constraints
create index on "dem"."singapore_srtm" using gist (st_convexhull("rast"));
