Download numerical weather prediction datasets (HRRR, RAP, GFS, IFS, etc.) from NOMADS, NODD partners (Amazon, Google, Microsoft), ECMWF open data, and the University of Utah Pando Archive System.
On every Wednesday, GEFS reforecast dataset has different ensemble size and forecast length, and URLs are slightly different from other dates.
I made a small change to corresponding functions, as below
class gefs_reforecast:
........
def template(self):
self.DESCRIPTION = "Global Ensemble Forecast System (GEFS)"
self.DETAILS = {
"aws": "https://registry.opendata.aws/noaa-gefs-reforecast/",
}
self.PRODUCTS = {
"GEFSv12/reforecast": "reforecasts for 2000-2019",
}
if self.date.weekday() == 2:
# GEFS reforecast has extended ensemble on every Wednesday
max_member_size = 10
max_fxx = "Days:10-35"
else:
max_member_size = 4
max_fxx = "Days:10-16"
# Adjust "member" argument
# - Member 0 is the control member
# - Members 1-4 are the perturbation members
if self.member == 0:
member = f"c{self.member:02d}"
elif self.member > 0 and self.member < max_member_size:
member = f"p{self.member:02d}"
else:
raise ValueError(f"GEFS 'member' must be within range of [0 - {max_member_size}].")
# Adjust "fxx" argument (given in hours)
# This is used to define the directory to enter rather than the filename.
if self.fxx <= 240:
fxx = "Days:1-10"
else:
fxx = max_fxx
This change at least satisfies this call
H = Herbie("2000-01-05", model="gefs_reforecast", fxx=34 * 24, member=9, variable_level="ugrd_hgt")
However, I didn't test it thoroughly and I am not in a comfortable situation I could send a PR in. Could you kindly review and consider adopting this change?
On every Wednesday, GEFS reforecast dataset has different ensemble size and forecast length, and URLs are slightly different from other dates.
I made a small change to corresponding functions, as below
This change at least satisfies this call
However, I didn't test it thoroughly and I am not in a comfortable situation I could send a PR in. Could you kindly review and consider adopting this change?
Best,