990 remove manual adjustment of start index (#991)

* replaced start_index workaround by new xugrid setting

* updated minimal xugrid version

* updated whatsnew

dfm_tools/meshkernel_helpers.py

@@ -185,17 +185,13 @@ def meshkernel_to_UgridDataset(mk:meshkernel.MeshKernel, crs:(int,str) = None) -
         raise ValueError(f"crs has is_geographic={crs_is_geographic} and grid has is_geographic={grid_is_geographic}. This is not allowed.")
 
     # TODO: below is not correctly handled by xugrid yet, projected=False does not give is_geographic=True
-    # related issue is https://github.com/Deltares/dfm_tools/issues/686
+    # related issue is https://github.com/Deltares/xugrid/issues/187
     xu_grid = xu.Ugrid2d.from_meshkernel(mesh2d_grid, projected= not crs_is_geographic, crs=crs)
 
-    #convert 0-based to 1-based indices for connectivity variables like face_node_connectivity
+    # convert 0-based to 1-based indices for connectivity variables like face_node_connectivity
+    # this is required by delft3dfm
+    xu_grid.start_index = 1
     xu_grid_ds = xu_grid.to_dataset()
-    xu_grid_ds = xr.decode_cf(xu_grid_ds) #decode_cf is essential since it replaces fillvalues with nans
-    ds_idx = xu_grid_ds.filter_by_attrs(start_index=0)
-    for varn_conn in ds_idx.data_vars:
-        xu_grid_ds[varn_conn] += 1 #from startindex 0 to 1 (fillvalues are now nans)
-        xu_grid_ds[varn_conn].attrs["start_index"] += 1
-        xu_grid_ds[varn_conn].encoding["_FillValue"] = -1 #can be any value <=0, but not 0 is currently the most convenient for proper xugrid plots.
 
     # convert to uds and add attrs and crs
     xu_grid_uds = xu.UgridDataset(xu_grid_ds)

docs/whats-new.md

@@ -9,6 +9,7 @@
 - deprecated `dfmt.open_dataset_extra()` (partly replaced by `dfmt.open_prepare_dataset()`) in [#974](https://github.com/Deltares/dfm_tools/pull/974)
 - improved nan-conversion in `dfmt.forcinglike_to_Dataset()` in [#982](https://github.com/Deltares/dfm_tools/pull/982)
 - improved performance of `dfmt.open_partitioned_dataset()` for datasets with many variables in [#984](https://github.com/Deltares/dfm_tools/pull/984)
+- simplified `dfmt.meshkernel_to_UgridDataset()` by using new xugrid version in [#991](https://github.com/Deltares/dfm_tools/pull/991)
 
 
 ## 0.25.0 (2024-08-16)

pyproject.toml

@@ -36,8 +36,8 @@ dependencies = [
 	"netcdf4>=1.5.4",
 	#bottleneck<1.3.3 pip install fails in py39
 	"bottleneck>=1.3.3",
-	#xugrid<0.12.0 has sub-optimal performance because of accessing dataarrays of variables
-	"xugrid>=0.12.0",
+	#xugrid<0.12.1 has issue with fill_value and no flexible start_index
+	"xugrid>=0.12.1",
 	#cdsapi<0.7.2 has different error upon dummy dataset
 	"cdsapi>=0.7.2",
 	#pydap<3.4.0 is from May 2017 and does not support newer python versions