diff --git a/marquette/conf/config.yaml b/marquette/conf/config.yaml
index 878d56d..f313ac6 100644
--- a/marquette/conf/config.yaml
+++ b/marquette/conf/config.yaml
@@ -1,12 +1,12 @@
 name: merit_s3
-s3_path: s3://mhpi-spatial/marquette/${name}/
+s3_path: s3://mhpi-spatial/marquette/merit
 data_path: /projects/mhpi/data/MERIT
 zone: 73
 gpu: 6
 create_edges:
   buffer: 0.3334
   dx: 2000
-  edges: ${s3_path}/edges/
+  edges: ${s3_path}/edges
   flowlines: ${data_path}/raw/flowlines/riv_pfaf_${zone}_MERIT_Hydro_v07_Basins_v01_bugfix1.shp
 create_N:
   run_whole_zone: False
diff --git a/marquette/merit_s3/create.py b/marquette/merit_s3/create.py
index e519c19..60a9480 100644
--- a/marquette/merit_s3/create.py
+++ b/marquette/merit_s3/create.py
@@ -6,6 +6,7 @@
 import pandas as pd
 import xarray as xr
 import zarr
+import dask.array
 from dask.dataframe.io.io import from_pandas
 from omegaconf import DictConfig
 from tqdm import tqdm
@@ -48,11 +49,10 @@ def write_streamflow(cfg: DictConfig, edges: zarr.Group) -> None:
         log.info("Streamflow data already exists")
 
 
-def create_edges(cfg: DictConfig) -> zarr.Group:
+def create_edges(cfg: DictConfig) -> xr.Dataset:
     try:
-        root = xr.open_datatree(cfg.create_edges.edges, engine="zarr")
+        edges = xr.open_dataset(f"{cfg.create_edges.edges}/{str(cfg.zone)}", engine="zarr")
         log.info("Edge data already exists on s3")
-        edges = root[str(cfg.zone)]
     except FileNotFoundError:
         log.info("Edge data does not exist. Creating connections")
         polyline_gdf = gpd.read_file(cfg.create_edges.flowlines)
@@ -171,27 +171,25 @@ def create_edges(cfg: DictConfig) -> zarr.Group:
             merged_df[col] = merged_df[col].astype(dtype=np.int32)
         for col in ["len", "len_dir", "slope", "sinuosity", "stream_drop", "uparea"]:
             merged_df[col] = merged_df[col].astype(dtype=np.float32)
-        # xr_dataset = xr.Dataset.from_dataframe(merged_df)
-        ds = merged_df.to_xarray().assign_coords({"merit_basin": merged_df["merit_basin"]})
-        sorted_keys_array = np.array(sorted_keys)
-        sorted_edges = xr.Dataset()
-        for var_name in xr_dataset.data_vars:
-            sorted_edges[var_name] = sort_xarray_dataarray(
-                xr_dataset[var_name],
-                sorted_keys_array,
-                xr_dataset["segment_sorting_index"].values,
-            )
-            shape = sorted_edges[var_name].shape
-            dtype = sorted_edges[var_name].dtype
-            tmp = edges.zeros(var_name, shape=shape, chunks=1000, dtype=dtype)
-            tmp[:] = sorted_edges[var_name].values
-        tmp = edges.zeros(
-            "sorted_keys",
-            shape=sorted_keys_array.shape,
-            chunks=1000,
-            dtype=sorted_keys_array.dtype,
+
+        idx = np.argsort(merged_df["uparea"])
+        sorted_df = merged_df.iloc[idx]
+        merit_basins = sorted_df["merit_basin"]
+        
+        edges = xr.Dataset(
+            {var: (["comid"], sorted_df[var]) for var in sorted_df.columns if var != "crs"},
+            coords={"comid": merit_basins}
+        )
+        edges.attrs['crs'] = sorted_df["crs"].unique()[0]
+        
+        # For faster writes, it's easier to mock the datastore using dask, and then upload the store using to_zarr()
+        # dummies = dask.array.zeros(merit_basins.shape[0], chunks=merit_basins.shape[0])
+        # xr.Dataset(data_vars={"id": ("merit_basins", dummies)}, coords={"merit_basins": merit_basins}).to_zarr(cfg.create_edges.edges, compute=False)
+        edges.to_zarr(
+            store=f"{cfg.create_edges.edges}/{str(cfg.zone)}",
+            mode='w', 
+            consolidated=True
         )
-        tmp[:] = sorted_keys_array
     return edges