♻️ Use xarray.merge with join="override" in collate functions

Refactoring the xarray collate functions to use `xr.merge` instead of the dictionary style way of appending data variables to an xarray.Dataset. Solution adapted from 7787f8e in #62 that is more robust to images being cut off due to rounding issues as with 6b18934 in #31. Downside is the need to verbosely rename the xarray.DataArray objects, and handle some conflicting coordinate labels.

docs/stacking.md

@@ -322,7 +322,7 @@ dp_sen1_copdem
 Next, use {py:class}`torchdata.datapipes.iter.Collator` (functional name:
 `collate`) to convert 🤸 the tuple of {py:class}`xarray.DataArray` objects into
 an {py:class}`xarray.Dataset` 🧊, similar to what was done in
-{doc}`./object-detection-boxes`.
+{doc}`./vector-segmentation-masks`.
 
 ```{code-cell}
 def sardem_collate_fn(sar_and_dem: tuple) -> xr.Dataset:
@@ -333,12 +333,13 @@ def sardem_collate_fn(sar_and_dem: tuple) -> xr.Dataset:
     # Turn 2 xr.DataArray objects into 1 xr.Dataset with multiple data vars
     sar, dem = sar_and_dem
 
-    # Initialize xr.Dataset with VH and VV channels
-    dataset: xr.Dataset = sar.sel(band="vh").to_dataset(name="vh")
-    dataset["vv"] = sar.sel(band="vv")
-
-    # Add Copernicus DEM mosaic as another layer
-    dataset["dem"] = dem.squeeze()
+    # Create datacube with VH and VV channels from SAR + Copernicus DEM mosaic
+    da_vh: xr.DataArray = sar.sel(band="vh", drop=True).rename("vh")
+    da_vv: xr.DataArray = sar.sel(band="vv", drop=True).rename("vv")
+    da_dem: xr.DataArray = (
+        dem.sel(band="data").drop_vars(names=["proj:epsg", "platform"]).rename("dem")
+    )
+    dataset: xr.Dataset = xr.merge(objects=[da_vh, da_vv, da_dem], join="override")
 
     return dataset
 ```
@@ -366,7 +367,6 @@ Visualize the DataPipe graph ⛓️ too for good measure.
 torchdata.datapipes.utils.to_graph(dp=dp_vhvvdem_dataset)
 ```
 
-
 ### Rasterize target labels to datacube extent 🏷️
 
 The landslide polygons 🔶 can now be rasterized and added as another layer to

docs/vector-segmentation-masks.md

@@ -88,7 +88,7 @@ geographic coordinates by default (OGC:CRS84). To make the pixels more equal 
 area, we can project it to a 🌏 local projected coordinate system instead.
 
 ```{code-cell}
-def reproject_to_local_utm(dataarray: xr.DataArray, resolution: float=100.0) -> xr.DataArray:
+def reproject_to_local_utm(dataarray: xr.DataArray, resolution: float=80.0) -> xr.DataArray:
     """
     Reproject an xarray.DataArray grid from OGC:CRS84 to a local UTM coordinate
     reference system.
@@ -382,8 +382,10 @@ def xr_collate_fn(image_and_mask: tuple) -> xr.Dataset:
     """
     # Turn 2 xr.DataArray objects into 1 xr.Dataset with multiple data vars
     image, mask = image_and_mask
-    dataset: xr.Dataset = image.isel(band=0).to_dataset(name="image")
-    dataset["mask"] = mask
+    dataset: xr.Dataset = xr.merge(
+        objects=[image.isel(band=0).rename("image"), mask.rename("mask")],
+        join="override",
+    )
 
     # Clip dataset to bounding box extent of where labels are
     mask_extent: tuple = mask.where(cond=mask == 1, drop=True).rio.bounds()