No available images on CRS change

I have been following this example:
https://eo-learn.readthedocs.io/en/latest/examples/io/SentinelHubIO.html

It works fine but when I change the line

roi_bbox = BBox(bbox=[5.60, 52.68, 5.75, 52.63], crs=CRS.WGS84)

to

roi_bbox = BBox(bbox=[1475000.0, 5100000.0, 1495000.0, 5120000.0], crs=CRS(‘2193’))

there are no available images found. My guess is that CRS(‘2193’) which is NZGD2000 / New Zealand Transverse Mercator 2000 is not compatible.

Although on this page https://docs.sentinel-hub.com/api/latest/api/process/crs/ its listed in the ‘other’ section. Can I not use this CRS format?

The error message I get is:

ValueError: During execution of task SentinelHubInputTask: No available images for requested time range: (datetime.datetime(2018, 4, 1, 0, 0), datetime.datetime(2018, 5, 1, 23, 59, 59))

Page 1 / 1

Hi,

If I transform your bounding box back to WGS84 I get:

BBox(bbox=[1475000.0, 5100000.0, 1495000.0, 5120000.0], crs=CRS('2193')).transform(CRS.WGS84)

> BBox(((-118.79345083627057, -57.54181875157152), (-118.65579881368123, -57.3346324139308)), crs=CRS('4326'))

This location is somewhere in the south Pacific ocean and there are no images available for the location. Are you sure you have correct coordinates?

Thanks for your response maleksandrov.

I am getting confused here, I see what you are saying, it does drop the WGS84 coordinates into the ocean - well away from the original location. But is there something wrong with the conversion as the original coordinates are just off the coast of New Zealand. See image

Further to this, if I use this website

epsg.io

Transform coordinates - GPS online convertor

Transform your coordinates online easily with epsg.io

the original coordinates transform to 171.434508, -44.2425095 (not -118.79345083627057, -57.54181875157152). Is there something wrong with the BBox.transform?

You are right, this is indeed weird. I checked your coordinates in QGIS and got the same location as you.

The implementation of BBox.transform in the background uses pyproj package. Here is the code that does a transformation using only pyproj:

import pyproj



POINT = 1475000.0, 5100000.0



transformer = pyproj.Transformer.from_crs(

    crs_from=pyproj.CRS(2193),

    crs_to=pyproj.CRS(4326)

)



transformer.transform(*POINT)



> (-57.54181875157152, -118.65579881368123)

or even

import pyproj



POINT = 1475000.0, 5100000.0



transformer = pyproj.Transformer.from_proj(

    proj_from=pyproj.Proj(2193),

    proj_to=pyproj.Proj(4326)

)



transformer.transform(*POINT)



> (-57.54181875157152, -118.65579881368123)

This produces the same wrong result (BBox.transform just makes sure to use lng-lat order instead of the default lat-lng). I tried running this code with the latest pyproj version 3.1.0 and an older 2.2.0 version and in both cases got the same result.

Hence, pyproj has a wrong/different interpretation of some less common CRS. I can even confirm we recently noticed a similar pyproj issue with a different CRS (EPSG:3346). Probably it is worth opening an issue at pyproj repository.

I investigated the issue further and it seems that is actually this issue. Adding the parameter always_xy=True to transformer initialization solves the problem and correctly transforms coordinates to

171.43450807766266, -44.242509416017654

Therefore, it seems that we have to start using this parameter also in sentinelhub-py. This will be fixed in the next release.

Ok great. Thanks a lot for your help maleksandrov.

The fix for this has just been released in the new sentinelhub-py version 3.3.2.

In sentinelhub-py transformation methods and functions we now have an optional always_xy which is by default set to True but users can change that. The default option seems to work correctly for all our use cases.