Regions¶
A latitude and longitude box¶
europe = climval.Box(lat=(35, 72), lon=(-10, 40), name="Europe")
result = climval.compare(ref, cand, "tas", region=europe)
lat is (south, north). lon is (west, east), read eastward from west
to east, in either convention: (-10, 40) and (350, 40) are the same box.
west > east crosses the antimeridian:
pacific = climval.Box(lat=(-20, 20), lon=(160, -150)) # 160E eastward to 150W
band = climval.Box(lat=(60, 90)) # all longitudes
A dict works too: region={"lat": (35, 72), "lon": (-10, 40)}.
The box is cut before regridding: the target grid exactly, the grid that is regridded with a margin of one cell (and, for bilinear and nearest, wrapped columns when it is global) so the points at the edge have neighbours. The step records how many points each side kept.
regionmask regions¶
With regionmask installed
(pip install "climval[regions]"), name a region as "<collection>:<region>":
result = climval.compare(ref, cand, "tas", region="ar6.land:NEU") # IPCC AR6 Northern Europe, land
result = climval.compare(ref, cand, "tas", region="ar6.all:MED") # Mediterranean, land and sea
The collection is an attribute path under regionmask.defined_regions
(ar6.land, ar6.all, ar6.ocean, srex, giorgi and others); the region
is its abbreviation, name or number. climval cuts the region's bounding box
before regridding and applies the polygon mask after regridding, on the
target grid; the region_mask step records how many cells lie inside.
You can also pass a regionmask region object with climval.MaskRegion:
The IPCC AR6 reference regions are described by Iturbide et al. (2020), https://doi.org/10.5194/essd-12-2959-2020.
Land or sea only¶
For a land-only comparison without a named region, mask the reference with a land fraction before the comparison; the paired mask then removes the same points from the candidate: