Upcoming sf version will break MODIS (when using s2 for geometries with ellipsoidal coordinates)

[edzer]

Please see r-spatial/sf#1649 - please consider commenting there if you get stuck! (@paleolimbot: another ✖ Loop 0 is not valid: Edge 1 crosses edge 6)

[edzer]

Still the case. Please consider using sf::st_make_valid on data with ellipsoidal coordinates.

  > getTile(Up)
  Error: Problem with `filter()` input `..1`.
  ℹ Input `..1` is `lengths(.predicate(x, y, ...)) > 0`.
  ✖ Evaluation error: Found 4 features with invalid spherical geometry.
  [13] Loop 0 is not valid: Edge 1 crosses edge 6
  [18] Loop 0 is not valid: Edge 0 crosses edge 5
  [446] Loop 0 is not valid: Edge 1 crosses edge 4
  [451] Loop 0 is not valid: Edge 2 crosses edge 5.
  Backtrace:
       █
    1. ├─MODIS::getTile(Up)
    2. ├─MODIS::getTile(Up)
    3. │ ├─MODIS::getTile(sf::st_as_sf(x), ...)
    4. │ └─MODIS::getTile(sf::st_as_sf(x), ...)
    5. │   ├─base::suppressMessages(sf::st_filter(sr, x))
    6. │   │ └─base::withCallingHandlers(...)
    7. │   ├─sf::st_filter(sr, x)
    8. │   └─sf:::st_filter.sf(sr, x)
    9. │     ├─dplyr::filter(x, lengths(.predicate(x, y, ...)) > 0)
   10. │     ├─sf:::filter.sf(x, lengths(.predicate(x, y, ...)) > 0)
   11. │     │ └─sf:::.re_sf(...)
   12. │     │   └─base::stopifnot(...)
   13. │     ├─base::NextMethod()
   14. │     └─dplyr:::filter.data.frame(...)
   15. │       └─dplyr:::filter_rows(.data, ...)
   16. │         ├─base::withCallingHandlers(...)
   17. │         └─mask$eval_all_filter(dots, env_filter)
   18. ├─base::lengths(.predicate(x, y, ...))
   19. ├─sf:::.predicate(x, y, ...)
   20. ├─sf:::st_intersects.sf(x, y, ...)
   21. │ └─sf:::st_geos_binop(...)
   22. │   └─s2:::fn(x, y, s2::s2_options(model = s2_model, ...))
   23. │     ├─s2:::cpp_s2_intersects_matrix(...)
   24. │     ├─s2::as_s2_geography(x)
   25. │     └─sf:::as_s2_geography.sf(x)
   26. │       └─sf:::st_as_s2.sf(x, ..., oriented = oriented)
   27. │         ├─sf::st_as_s2(st_geometry(x), ...)
   28. │         └─sf:::st_as_s2.sfc(st_geometry(x), ...)
   29. │           ├─s2::as_s2_geography(st_as_binary(x), ..., oriented = oriented)
   30. │           └─s2:::as_s2_geography.WKB(st_as_binary(x), ..., oriented = oriented)
   31. │             ├─s2:::new_s2_xptr(...)
   32. │             └─s2:::s2_geography_from_wkb(x, oriented = oriented, check = check)
   33. ├─base::stop(...)
   34. └─(function (e) ...
  Execution halted

You can test with s2 (1.0.5) from CRAN, and sf branch sf_1_0 from the github repo. Let me know if we can help.

[karldw]

FYI, MODIS was been removed from CRAN on June 28, 2021 because of these errors.

Users can install an archived version with remotes::install_version("MODIS", "1.2.3") (though you'll run into the issues described above unless you also install an old version of sf).

[edzer]

install an old version of sf

... or use sf_use_s2(FALSE).

[MatMatt]

Dear Edzer, Thank you for insisting on this! I have been completely out from working on the MODIS package since quite a while now. I have tried to logon to my github but it asks me for a 2f auth which I had to reset now. After getting back the access I will have a look at this!! Thanks again Matteo Am Mi., 30. Juni 2021 um 18:41 Uhr schrieb Edzer Pebesma < ***@***.***>:

…

... or use sf_use_s2(FALSE). — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#110 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AA4F4GUQBHR7H4Q2UXQT6O3TVNCM3ANCNFSM43DMV6QA> .

[edzer]

Great - let me know when you need help!

[fdetsch]

@edzer Thanks for the notification! I haven't been following the discussion in quite a while. From what I understand, would it be enough to examine any incoming geometry with sf::st_is_valid() and, if required, then invoke sf::st_make_valid()?

[fdetsch]

So to answer my own question, it seems like this approach is not viable. At least,

sf::sf_use_s2(TRUE)

data(meuse, package = "sp")
pts = sf::st_as_sf(meuse, coords = c("x", "y"), crs = 28992)

all(sf::st_is_valid(pts))
# TRUE

MODIS::getTile(pts[1, ])

using valid geometries only leads to the same error:

 Error: Problem with `filter()` input `..1`.
ℹ Input `..1` is `lengths(.predicate(x, y, ...)) > 0`.
x Evaluation error: Found 4 features with invalid spherical geometry.
[13] Loop 0 is not valid: Edge 1 crosses edge 6
[18] Loop 0 is not valid: Edge 0 crosses edge 5
[446] Loop 0 is not valid: Edge 1 crosses edge 4
[451] Loop 0 is not valid: Edge 2 crosses edge 5.

Will start investigating with temporarily setting sf_use_s2(FALSE), in particular if this impacts the operability of the pkg.