def _build_norms(vals, days, yr_days):
mth_days = get_md_array(days[DATE])
# Build 15-day norms for every day in days
day_nums = np.arange(yr_days.size)
# Column 0: mean, Column 1: std
norms = np.ones([yr_days.size, 2]) * np.NAN
fin_mask = np.isfinite(vals)
for x in day_nums:
date = yr_days[x]
srt_date = date - A_WEEK
end_date = date + A_WEEK
date_range = get_date_array(srt_date, end_date)
mth_days_range = get_md_array(date_range)
date_mask = np.in1d(mth_days, mth_days_range)
vals_rng = vals[np.logical_and(fin_mask, date_mask)]
if vals_rng.size >= MIN_NORM_VALUES:
norms[x, 0], norms[x, 1] = _biweight_mean_std(vals_rng)
return norms
def _build_percentiles(vals, days, yr_days):
mth_days = get_md_array(days[DATE])
# Build 29-day percentiles for every day in yr_days
day_nums = np.arange(yr_days.size)
# Columns: 30th,50th,70th,90th,95th
percentiles = np.ones([yr_days.size, 5]) * np.NAN
for x in day_nums:
date = yr_days[x]
srt_date = date - TWO_WEEKS
end_date = date + TWO_WEEKS
date_range = get_date_array(srt_date, end_date)
mth_days_range = get_md_array(date_range)
date_mask = np.in1d(mth_days, mth_days_range)
vals_rng = vals[np.logical_and(
np.logical_and(_not_nan(vals), vals > 0), date_mask)]
if vals_rng.size >= MIN_PERCENTILE_VALUES:
percentiles[x, :] = [
np.percentile(vals_rng, 30),
np.percentile(vals_rng, 50),
np.percentile(vals_rng, 70),
np.percentile(vals_rng, 90),
np.percentile(vals_rng, 95)
]
return percentiles
def _get_pctiles_md_masks(days, yr_days):
mth_days = get_md_array(days[DATE])
day_nums = np.arange(yr_days.size)
pctile_masks = np.zeros((yr_days.size, mth_days.size), dtype=np.bool)
for x in day_nums:
mth_day = yr_days[x]
srt_mth_day = mth_day - TWO_WEEKS
end_mth_day = mth_day + TWO_WEEKS
date_range = get_date_array(srt_mth_day, end_mth_day)
mth_days_range = get_md_array(date_range)
pctile_masks[x, :] = np.in1d(mth_days, mth_days_range)
return pctile_masks
def _get_norms_md_masks(days, yr_days):
mth_days = get_md_array(days[DATE])
# For non-leap years
day_nums = np.arange(yr_days.size)
norm_masks = np.zeros((yr_days.size, mth_days.size), dtype=np.bool)
for x in day_nums:
mth_day = yr_days[x]
srt_mth_day = mth_day - A_WEEK
end_mth_day = mth_day + A_WEEK
date_range = get_date_array(srt_mth_day, end_mth_day)
mth_days_range = get_md_array(date_range)
norm_masks[x, :] = np.in1d(mth_days, mth_days_range)
return norm_masks
NGH_CORR = 0.8
NGH_RESID_CUTOFF = 8.0
NGH_RESID_STD_CUTOFF = 4.0
ANOMALY_CUTOFF = 10.0
MIN_DAYS_MTH_WINDOW = 40
MIN_NGHS = 3
MAX_NGHS = 7
NGH_STNS_ID = "NGH_STNS_ID"
NGH_STNS_MASK_OVERLAP = "NGH_STNS_MASK_OVERLAP"
NGH_STNS_WGHTS = "NGH_STNS_WGHTS"
NGH_STNS_MODEL = "NGH_STNS_MODEL"
NGH_STNS_OBS = "NGH_STNS_OBS"
# Constants for building Tmin/Tmax normals
MONTHS = np.arange(1, 13)
DATES_366 = get_date_array(datetime(2004, 1, 1), datetime(2004, 12, 31))
DATES_365 = get_date_array(datetime(2003, 1, 1), datetime(2003, 12, 31))
MIN_NORM_VALUES = 100
TWX_TO_GHCN_FLAGS_MAP = {
QA_OK: "",
QA_MISSING: "",
DUP: "D",
QA_DUP_YEAR: "D",
QA_DUP_MONTH: "D",
QA_DUP_YEAR_MONTH: "D",
QA_DUP_WITHIN_MONTH: "D",
QA_GAP: "G",
QA_INTERNAL_INCONSIST: "I",
QA_STREAK: "K",
QA_MEGA_INCONSIST: "M",