Maxar Data Access and Stereo Processing

[scottyhq]

Search

Currently we're using https://pypi.org/project/maxar-platform/ to search for Maxar Stereo data (https://developers.maxar.com/docs/developer-tools/python-sdk). It can do lots of other things too such as order data if you have a funded MGP Pro account https://pro.maxar.com/. I think anyone can register for a free account to at least perform searches?

Recently, NASA CSDA also started allowing NASA-researchers access to a subset of Maxar data (https://www.earthdata.nasa.gov/about/csda/commercial-datasets), which once you've signed all the licensing agreements, you can search via CMR https://search.earthdata.nasa.gov/search?q=CSDA . So perhaps we add that as a search option or document a means to check if a Maxar scene ID is in the CSDA holdings?

I'd really like to add an example to the docs of finding a cloud-free pair & downloading required data to pass off to ASP (preferably open-access or available via CSDA, and could be anywhere but preferably at a STV-priority site)

I know there are other open sources of Maxar Stereo data floating around (something from https://spacenet.ai/)?

Processing

It will also be good to add a bare-bones example of executing ASP to generate a DEM from the stereo pair above @ShashankBice does such an example exist? Perhaps that should actually https://github.com/uw-cryo/asp_tutorials :) But we should link to ASP Docs and other important references.

[dshean]

Thanks Scott!

For basic search, I'm hoping we can use basic MGP xpress or MGP functionality to search without creating an account or authenticating. But I fear that my no longer be possible. Perhaps we can create a shared MGP account to use for coincident users.

I'm hoping that it won't be too difficult to build out a separate workflow for CMR search, order, and download workflow (for Maxar, but also other CSDA stereo data in the future). I don't expect this to be widely used due to federal funding requirements and hoops for account creation. But future "on-demand ASP as a service" capability will need this functionality, hopefully used via a simple web client for authenticated users.

The big question in my mind is subsetting the Level-1B images stored in S3. We can do this directly with ASP, cropping the L1Bs, but I think the best approach will be with the "mapproject stereo" approach, defining the desired extent in the output projected coordinates (e.g., a 1x1 km box in UTM 10N), and relying on ASP mapproject or gdalwarp to only read the relevant portions of the tiled Level-1B geotiff stored in S3 when creating the orthoimages. But I expect there may be some challenges there.

[dshean]

I will share a doc from earlier evaluation of the available SpaceNet datasets. There are many hosted in S3 already.
the SpaceNet 4 "Off-Nadir Buildings" (AOI_6_Atlanta) is a good option, with available public lidar data.

• https://spacenet.ai/off-nadir-building-detection/
• https://spacenet.ai/atlanta/
• https://openaccess.thecvf.com/content_ICCV_2019/papers/Weir_SpaceNet_MVOI_A_Multi-View_Overhead_Imagery_Dataset_ICCV_2019_paper.pdf
• 27 WV-2 images with off-nadir angles of 7 to 54 degrees

Level-1B images ready for stereo processing are in the poorly named “metadata” directory. They should be merged with dg_mosaic. Do not use the “nadir10” and other images, which are COGs, but already orthorectified.

The other primary candidate for sample Level-1B data in the US with publicly available 3DEP and real topographic relief is the UCSD site, distributed under the MVS Challenge, CORE-3D efforts:

• https://spacenet.ai/core3d/
• https://www.spiedigitallibrary.org/conference-proceedings-of-spie/10636/106360P/Large-scale-public-lidar-and-satellite-image-data-set-for/10.1117/12.2304403.full
• 37 WV-3 images
• 48 WV-2 images

The other sites for Jacksonville, Tampa, Richmond and Omaha are all relatively flat :)

I realize they may not be 4-way coincident sites, but I think we likely want to elevate the priority for Atlanta and UCSD as STV sites given the publicly available WV data.

[dshean]

And agreed on bare-bones example for public WV stereo images in asp_tutorials. We've talked with Oleg about wrapping the "mapproject stereo" approach as a standard "mode" for ASP parallel_stereo, which would simplify user-facing wrapper scripts.

My simple example using CSM with hardcoded image paths, output GSD and output CRS is here: https://github.com/uw-cryo/wv_stereo_processing/blob/main/scripts/csm_proc/simple_mgm_csm.sh. The other wrapper scripts that attempt to determine these items automatically end up being much more complicated (and confusing for new users).

The existing ASP doc for "DigitalGlobe" was written 10 years ago and is a little confusing. We can use this effort to improve "quickstart" documentation, illustrating best practices for processing Maxar stereo images, simplifying the onboarding for new users unfamiliar with ASP.