new gfs loader

ocf_datapipes/load/nwp/providers/gfs.py

@@ -1,50 +1,33 @@
 """Open GFS Forecast data"""
 
 import logging
-from pathlib import Path
-from typing import Union
 
-import pandas as pd
 import xarray as xr
 
+from ocf_datapipes.load.nwp.providers.utils import open_zarr_paths
+
 _log = logging.getLogger(__name__)
 
 
-def open_gfs(zarr_path: Union[Path, str]) -> xr.Dataset:
+def open_gfs(zarr_path) -> xr.DataArray:
     """
-    Opens GFS dataset
+    Opens the GFS data
 
     Args:
-        zarr_path: Path to Zarr(s) to open
+        zarr_path: Path to the zarr to open
 
     Returns:
-        Xarray dataset of GFS Forecasts
+        Xarray DataArray of the NWP data
     """
-
     _log.info("Loading NWP GFS data")
 
-    if "*" in zarr_path:
-        nwp = xr.open_mfdataset(zarr_path, engine="zarr", combine="time", chunks="auto")
-    else:
-        nwp = xr.load_dataset(zarr_path, engine="zarr", mode="r", chunks="auto")
-
-    variables = list(nwp.keys())
+    # Open data
+    gfs: xr.Dataset = open_zarr_paths(zarr_path, time_dim="init_time_utc")
+    nwp: xr.DataArray = gfs.to_array()
 
-    nwp = xr.concat([nwp[v] for v in variables], "channel")
-    nwp = nwp.assign_coords(channel=variables)
+    del gfs
 
-    nwp = nwp.transpose("time", "step", "channel", "latitude", "longitude")
-    nwp = nwp.rename({"time": "init_time_utc"})
+    nwp = nwp.rename({"variable": "channel"})
     nwp = nwp.transpose("init_time_utc", "step", "channel", "latitude", "longitude")
-    if "valid_time" in nwp.coords.keys():
-        nwp = nwp.drop("valid_time")
-
-    _log.debug("Interpolating hour 0 to NWP data")
-    nwp_step0 = nwp.interp(step=[pd.Timedelta(hours=0)])
-    nwp = xr.concat([nwp_step0, nwp], dim="step")
-    nwp = nwp.resample(init_time_utc="60min").pad()
-    nwp = nwp.resample(step="60min").pad()
-
-    _log.debug(nwp)
 
     return nwp

ocf_datapipes/select/filter_channels.py

@@ -3,6 +3,7 @@
 import logging
 from typing import List, Union
 
+import numpy as np
 import xarray as xr
 from torch.utils.data import IterDataPipe, functional_datapipe
 
@@ -18,6 +19,7 @@ def __init__(
         source_datapipe: IterDataPipe,
         channels: List[str],
         dim_name: str = "channel",
+        provider: str = None,
     ):
         """
         Filter channels
@@ -26,10 +28,22 @@ def __init__(
             source_datapipe: Datapipe of Xarray objects
             channels: Channel names to keep
             dim_name: Dimension name for channels
+            provider: Name of NWP source, if available
         """
         self.source_datapipe = source_datapipe
         self.channels = channels
         self.dim_name = dim_name
+        self.provider = provider
+
+        if self.provider == "gfs":
+            flux_vars = np.intersect1d(self.channels, ["dswrf", "dlwrf"])
+
+            if len(flux_vars) > 0:
+                logger.warning(
+                    f"You have requested channels that have no step 0: {flux_vars}. "
+                    f"Step 0 will be set to NaN. "
+                    f"For more info see https://github.com/openclimatefix/ocf_datapipes/issues/253"
+                )
 
     def __iter__(self) -> Union[xr.DataArray, xr.Dataset]:
         for xr_data in self.source_datapipe:

ocf_datapipes/training/common.py

@@ -159,7 +159,10 @@ def open_and_return_datapipes(
                     nwp_conf.nwp_zarr_path,
                     provider=nwp_conf.nwp_provider,
                 )
-                .filter_channels(nwp_conf.nwp_channels)
+                .filter_channels(
+                    nwp_conf.nwp_channels,
+                    provider=nwp_conf.nwp_provider,
+                )
                 .add_t0_idx_and_sample_period_duration(
                     sample_period_duration=minutes(nwp_conf.time_resolution_minutes),
                     history_duration=minutes(nwp_conf.history_minutes),

ocf_datapipes/utils/consts.py

@@ -134,22 +134,45 @@ def __getitem__(self, key):
 
 # ------ GFS
 GFS_STD = {
-    "t": 5.017000766747606,
-    "dswrf": 233.1834250473355,
-    "prate": 0.00021690701537950742,
-    "dlwrf": 46.571,
-    "u": 4.165,
-    "v": 4.123,
+    "dlwrf": 96.305916,
+    "dswrf": 246.18533,
+    "hcc": 42.525383,
+    "lcc": 44.3732,
+    "mcc": 43.150745,
+    "prate": 0.00010159573,
+    "r": 25.440672,
+    "sde": 0.43345627,
+    "t": 22.825893,
+    "tcc": 41.030598,
+    "u10": 5.470838,
+    "u100": 6.8899174,
+    "v10": 4.7401133,
+    "v100": 6.076132,
+    "vis": 8294.022,
+    "u": 10.614556,
+    "v": 7.176398,
 }
 GFS_MEAN = {
-    "t": 285.7799539185846,
-    "dswrf": 294.6696933986283,
-    "prate": 3.6078121378638696e-05,
-    "dlwrf": 319,
-    "u": 0.552,
-    "v": -0.477,
+    "dlwrf": 298.342,
+    "dswrf": 168.12321,
+    "hcc": 35.272,
+    "lcc": 43.578342,
+    "mcc": 33.738823,
+    "prate": 2.8190969e-05,
+    "r": 18.359747,
+    "sde": 0.36937004,
+    "t": 278.5223,
+    "tcc": 66.841606,
+    "u10": -0.0022310058,
+    "u100": 0.0823025,
+    "v10": 0.06219831,
+    "v100": 0.0797807,
+    "vis": 19628.32,
+    "u": 11.645444,
+    "v": 0.12330122,
 }
 
+
 GFS_VARIABLE_NAMES = tuple(GFS_MEAN.keys())
 GFS_STD = _to_data_array(GFS_STD)
 GFS_MEAN = _to_data_array(GFS_MEAN)

tests/data/gfs.zarr/.zattrs

@@ -0,0 +1,14 @@
+{
+    "Conventions": "CF-1.7",
+    "GRIB_centre": "kwbc",
+    "GRIB_centreDescription": "US National Weather Service - NCEP",
+    "GRIB_edition": 2,
+    "GRIB_subCentre": 0,
+    "description": "Global Forecast System",
+    "institution": "US National Weather Service - NCEP",
+    "local_grib": "/root/data/gfs/20230624/subset_56ef2a91__gfs.t00z.pgrb2.1p00.f000",
+    "model": "gfs",
+    "product": "pgrb2.1p00",
+    "remote_grib": "https://noaa-gfs-bdp-pds.s3.amazonaws.com/gfs.20230624/00/atmos/gfs.t00z.pgrb2.1p00.f000",
+    "search": ":[TLMH]CDC:|:[UV]GRD:(?:10|100) m|:RH:sigma|:(?:TMP|VIS|PRATE|SNOD|D[SL]WRF):surface|:RH:100 mb"
+}

tests/data/gfs.zarr/.zgroup

@@ -0,0 +1,3 @@
+{
+    "zarr_format": 2
+}