EarthEnv landcover + WorldClim 2.1 + CHELSA V1

[tpoisot]

The new data have the same API as worldclim, but the files are much larger. In fact, the tests do not run as GitHub actions using the whole layer, and it takes a solid machine to run them at all. They work, but they can't be tested automatically -- as a quick check, I still run them using the new ability to only load parts of a dataset.

Because these new data do not always have a NaN type, the absence of data is now represented as nothing - this had all sort of consequences for almost everything in the package, none of which broke the API at all. But internally, this is a very extensive rewrite.

This PR also comes with improvements, specifically the sliding window analysis, the ability to broadcast over layers, and the ability to get a subset of the data from a file. The last point in particular makes a lot of data loading much faster.

• earthenv - closes Landcover dataset #20
• sliding window - closes Sliding window analysis #22
• subset of the data - closes Only load parts of the data #23
• correct buffer size for Only load parts of the data #23 - closes Size of layers are inverted #25
• worldclim 2.1
• export CHELSA bioclim function (high-res)
• version bump (minor)
• updated documentation
• testing
• broadcasting
• fix the issue with latitudes when loading data subsets - closes Loading parts of the datasets give different regions on different datasets #24

[codecov-commenter]

Codecov Report

Merging #21 into master will increase coverage by 0.30%.
The diff coverage is 79.87%.

@@            Coverage Diff             @@
##           master      #21      +/-   ##
==========================================
+ Coverage   81.09%   81.40%   +0.30%     
==========================================
  Files          11       16       +5     
  Lines         201      285      +84     
==========================================
+ Hits          163      232      +69     
- Misses         38       53      +15     

Flag	Coverage Δ
#unittests	81.40% <79.87%> (+0.30%)	⬆️

Flags with carried forward coverage won't be shown. Click here to find out more.

Impacted Files	Coverage Δ
src/operations/sliding.jl	0.00% <0.00%> (ø)
src/datasets/chelsa.jl	33.33% <33.33%> (ø)
src/datasets/landcover.jl	33.33% <33.33%> (ø)
src/datasets/sources.jl	83.33% <83.33%> (ø)
src/lib/overloads.jl	83.56% <92.85%> (-0.05%)	⬇️
src/SimpleSDMLayers.jl	80.00% <100.00%> (+13.33%)	⬆️
src/datasets/download_layer.jl	100.00% <100.00%> (ø)
src/datasets/geotiff.jl	100.00% <100.00%> (ø)
src/datasets/raster.jl	100.00% <100.00%> (ø)
src/datasets/worldclim.jl	100.00% <100.00%> (ø)
... and 12 more

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update cd79fc1...8b3e1a2. Read the comment docs.

[tpoisot]

@gabrieldansereau Can you make sure that nothing in this PR is going to break what you are about to present at JuliaCon? This is a fairly big rewrite internally, but it didn't break any of the existing unit tests.

[gabrieldansereau]

I'll make sure it works while finishing my poster. I mean to show something similar to the examples in the documentation. As long as these are not broken, what I present shouldn't be either.

[gabrieldansereau]

@tpoisot Is this normal? The following

temperature = worldclim(1);
temperature_europe = temperature[left=-11.0, right=31.1, bottom=29.0, top=71.1];
size(temperature)
size(temperature_europe)

returns (2160, 1080) & (507, 128) with this version, but returns (1080, 2160) & (254, 254) on v0.2.2. Should the dimensions really be inversed in the first case and different in the second?

The coarsen example in the documentation is broken because of this.

[tpoisot]

@gabrieldansereau I have no idea why that's the case. I'll check this evening.

[tpoisot]

So, this is because the code of geotiff changed, and I would tempted to believe the new geotiff code, as using

worldclim(1; left=-11.0, right=31.1, bottom=29.0, top=71.1)

gives the correct result in terms of which area is covered. Let me check the size issue, it's weird.

[tpoisot]

Mmhhh. You're right @gabrieldansereau , it looks like the size is inverted:

julia> latitudes(temperature) |> length
2160

julia> longitudes(temperature) |> length
1080

julia> stride(temperature)
(0.16666666666666666, 0.04158950617283951)

In particular, the stride should have the same values - let me add a test to check this and then fix the bug.

[tpoisot]

GREAT now the github actions are not running. This is just fantastic. I'm ready to start a new life hoarding goats or something - nevermind, https://www.githubstatus.com/ says it's a site wide problem.

[gabrieldansereau]

All fine for me! Works with my JuliaCon poster. I had no problems running the examples and playing with it a bit after the small fix to similar().
Thanks for the fix to worldclim() & layer sizes!

[gabrieldansereau]

similar() didn't work without this for climate layers & all. eltype(temperature.grid) is always Union{Nothing, Float32}, which has no method for zero()