Map Great Britain with Polygons

[crangelsmith]

Functionality to consistently and efficiently map Great Britain and create a list of polygons.

Code in here can help:

• https://urbangrammarai.xyz/signature_ai/create_chips/chips_glasgow.html

[andrewphilipsmith]

Also, create an array for storing output results aligned with the same grid.

[andrewphilipsmith]

Three key vector datasets in operation in the pipeline.

The country boundary: We can check with Dani if he has a preferred source for this. But I suspect that this can be quite a simplified geometry - it just has to be detailed enough to decide if a chip is inside or outside. In the absence of another source, this one will suffice: https://github.com/wmgeolab/geoBoundaries/blob/main/releaseData/gbOpen/GBR/ADM0/geoBoundaries-GBR-ADM0_simplified.geojson It would be mildly preferable to convert and store this in BNG, but it is fairly trivial to reproject on demand.

The grid of chips

• This should be stored in BNG.
• It should be derived procedurally, inline with BNG and the chip-width (320m?)
• Each chip ID can be derived from the easting + northing of a fixed point on the chip (eg the bottom left corner) and the chip-width. The later is required in case we re-run the pipeline with different chip-width values - the chipID should be distinct.
• Once derived it should then be clipped to the country boundary. The country boundary should be (re-)projected to BNG for this operation.
• TBD this grid could be stored as a single geodataframe, with a column pointing to the filename of each individual chip.

We will have two sister data structures representing this grid:

1. A xarray.DataArray (stored as a COG with rioxarray) --> Frame for all predictions and uncertainty score on the inference stage
2. A geopandas.GeoDataFrame (stored as GeoParquet) with the structure: --> Maps directly on to chip objects to pass through inference

| ID | geometry | ... |

Importantly, 1. and 2. will be fully synchronised in terms of the coordinates. To get this, it might be worth to build 1. and, from there, construct 2. using the to_series feature in DataArray objects

The grid of regions / download images This is the grid of images as they requested from GEE. My understanding is that pyveg represents this as a centre-point and width in degrees (the region_size parameter).

• Orientated to BNG. Squares with a fixed region_size.
• regionID - could this be generated as an aggregate of the grid-of-chips.
• The size of the regions should be optimised to have the maximum size (area), but within the GEE image size (Mb) limit.
• The grid of images should be clipped to the grid of chips. The grid of chips could be reprojected to WGS84 for this operation.
• TBD this grid could be stored as a single geodataframe, with a column pointing to the filename of each individual GEE download.

Does this make sense? I would find it helpful to agree on names for these. I’m going to suggest “chips_grid” and “images_grid” - but I’m open to better suggestions ;-)