ST_Transform() function unavailable

[hjort]

I noted that ST_Transform() function is still unavailable in the package.
It is indeed a very useful and used function in GIS applications.

How hard is it to implement ST_Transform()?

ST_Transform

Regards,
Hjort

[randallwhitman]

For projection from WGS84 to Web Mercator, and de-projection from Web Mercator to WGS84, only, probably moderate effort.
For transformations and projections between dozens of datums, and hundreds of projections, a very large effort.

[hjort]

Sorry about the naïve question, but based on the approach Postgis employs (with a "spatial_ref_sys" table containing the available reference systems along with its guide parameters), isn't it possible to have a bunch of .properties or .json files with those SRS and an unique Java code to form a ST_Transform() function?

[randallwhitman]

Yes, some separation of code/logic from data/transform-definitions can be done. Careful design would be necessary about handling the transformation-data. Would it be stored at a location such that all processing nodes can download/side-load the data by URL? Would it be in JAR files that can be automatically distributed to processing nodes by Hadoop-Map-Reduce/Spark framework? Other?

[hjort]

Yeah, it really seems that it isn't a trivial task. :(

Postgis comes with a spatial_ref_sys table [1] populated with 3,000+ most common SRSs [2].
But, under the hood, it uses PROJ to execute these geospatial coordinates transformations [3,4], which is in C.
Apparently there was an initiative to port it to Java, Proj4J, but its repo is untouched for at least 7 years [5].

1: https://postgis.net/docs/manual-1.4/ch04.html#spatial_ref_sys
2: https://epsg.org/home.html
3: https://proj.org/
4: https://github.com/OSGeo/PROJ
5: https://github.com/Proj4J/proj4j

[clumzzey]

but since st_setsrid and st_srid are available it would be good to have the transformation available. It is the function I use very often and if I need a workaround (take it to python for example) for this, I could just as easy start doing all my spatial stuff there.

[randallwhitman]

The Spatial-framework-for-Hadoop is open-source and contributions are welcome.

You might also take a look at the links at - https://github.com/Esri/spatial-framework-for-hadoop#see-also

[clumzzey]

You make a fair point. My java skills are not good enough to do this so maybe I should have been just grateful for the things you do add to the table instead of complaining about what is missing. In my experience (n=1) transformations are like 90% from or to WGS84 format (4326). Would the solution become less complex if in first instance only from and to 4326 would be supported (st_tranformto4326 and st_transformfrom4326) ? If for each srid a seperate properties file is created like hjort suggests I would be happy to help along to create property files for the ones I need. Transformations between projections not being 4326 would be a 2-way step with the risk of rounding errors.
As for your earlier question, I would prefer the option "in JAR files that can be automatically distributed to processing nodes by Hadoop-Map-Reduce/Spark framework" since opening up a cluster to the internet might sometimes be a hassle.

Sorry for my n00bness if this is really a silly remark.

[randallwhitman]

Initial design discussed here separates properties data from code implementation. On the side of what the code would need to do, are two types of operations:

1. Projection (and de-projection) between a geographic coordinate system (GCS) and projected coordinate systems (PCS) that use the same datum; and
2. Transformation between different datums.

It would be less effort to implement only one of those, than to implement both. Implementation of either should be preceded by a complete design specification.