Suggestions for vector example datasets?

[ctoney]

Currently, the only example dataset for vector is the MTBS fire polygon layer for Yellowstone National Park, USA. This is used throughout the documentation and in tests. There are a handful of other vector layers in inst/extdata, but these are only of interest for tests.

I've started development of a vignette for the Vector API to be included in the forthcoming 2.0 release. It would be nice to have one or two other example datasets for vector.

Currently, all of the raster and vector datasets used in package examples are in the conterminous US. Having more geographic variation would be desirable. A pair of vector layers that might be used together in geoprocessing examples could also be helpful.

Example datasets need to be small of course, ideally 200 to 300 kB at the most (smaller if possible). A dataset stored in zipped format is an option to reduce file size (works well for GPKG but other formats too).

Any suggestions much appreciated.

[mdsumner]

Oh sorry I had lost track of this issue had intended to do this, I couldn't find the issue when I I looked. I had a good idea of some kind ....

[ctoney]

No problem. I should have added some explanation instead of just closing this.

I added inst/extdata/ynp_features.zip in #684. It is an SOZip file containing ynp_features.gpkg along with a readme. It complements the existing MTBS fire perimeters sample data. Includes the YNP boundary, roads and points_of_interest, all in geographic CS (NAD83). So it adds a line and point layer that could be used with the fire polygons, but makes interesting examples on its own (e.g., one could map the location of vault vs. flush toilet points of interest 😄). I used it in the example and tests for ogr_reproject() in #684.

Although GDAL can access the GPKG directly without explicit /vsizip/:

Starting with GDAL 3.7, the driver can also support reading and creating .gpkg.zip files containing one .gpkg file.

since gdalraster currently supports GDAL >= 3.1, we'll access it like this for now:

f <- system.file("extdata/ynp_features.zip", package = "gdalraster")
ynp_dsn <- file.path("/vsizip", f, "ynp_features.gpkg")

ogr_ds_layer_names(ynp_dsn)
#> [1] "points_of_interest" "roads"              "ynp_bnd"

It has a some slight weirdness retained from the source data, e.g., points and lines are xyz (POINT25D, LINESTRING25D) but z is not populated. Haven't looked too closely at the roads, but they seem overly discontinuous, i.e., many separate features for the same road name. Not sure if I'll clean it up a bit or just leave as-is since that's how it comes from the official source (https://public-nps.opendata.arcgis.com/). Possibly useful in some examples anyway.

Doesn't achieve the goal of having more geographic/hemispheric variation for the examples and vignette, so still interested in that as long as it doesn't grow inst/extdata too much.

[mdsumner]

one thought was maybe a GTI vector dataset that links to COP30 DEM tiles?

If I crop to Australia as Parquet this is about 35K:

v <- vect("/vsicurl/https://github.com/mdsumner/cog-example/raw/refs/heads/main/gti/cop30.fgb")
v <-  crop(v, ext(100, 160, -45, -10))

(whole world is 907K). That actual file needs cleaning up to be a really valid and efficient GTI (I created it before I really understood what's required), but that could be a bit of fun. Just, it's a bit early for Parquet - if we zipped a more accessible format we could do it, and even on installs without GTI we could demonstrate a bit of the ideas crossing vector/raster concepts.

[mdsumner]

ah, that won't work because as I just noticed the underlying tifs have changed recently, so this has be generated fresh from the VRT - nvm

old then new

https://opentopography.s3.sdsc.edu/raster/COP30/COP30_hh/Copernicus_DSM_10_S11_00_E105_00_DEM.tif
https://opentopography.s3.sdsc.edu/raster/COP30/COP30_hh/Copernicus_DSM_COG_10_S11_00_E105_00_DEM.tif