def test_agg_time():
"""
Tests if the correct number of days are the result
"""
serie = xr.DataArray(
np.arange(11),
dims=('time', ),
coords={'time': pd.date_range('2000-01-01', '2000-01-11')})
blocks = agg_time(serie, ndayagg=3, rolling=False)
assert len(
blocks
) == 11 // 3, "block aggregation with size 3 should result in the exact amount of full blocks that fit within the timeseries"
rolling = agg_time(serie, ndayagg=3, rolling=True)
assert len(
rolling
) == 11 - 2, "rolling aggregation with window size 3 should result in two less left-stamped observation"
blocks_start = agg_time(serie,
ndayagg=3,
rolling=False,
firstday=pd.Timestamp('2000-01-04'))
assert len(blocks_start) == (
11 - 3
) // 3, "block aggregation with size 3 should result in the exact amount of full blocks that fit within the timeseries after the given first day"
clusterfield = xr.open_dataarray(CLUSTERDIR /
't2m-q095.nc').sel(nclusters=15)
if spatial_quantile is None:
reduced = response.groupby(clusterfield).mean(
'stacked_latitude_longitude')
else:
reduced = response.groupby(clusterfield).quantile(
q=spatial_quantile, dim='stacked_latitude_longitude')
reduced = reduced.sel(
clustid=responseclustid) # In this case cluster 9 is western europe.
response.close()
output = []
for responsetimeagg in timeaggs:
responseagg = agg_time(array=reduced,
ndayagg=responsetimeagg,
method='mean',
rolling=True,
firstday=firstday)
logging.debug(
f'aggregated response to {responsetimeagg} day timeseries')
summersubset = responseagg[responseagg.time.dt.season == 'JJA']
if detrend_response:
summersubsetvals = detrend(summersubset.values, axis=0)
else:
summersubsetvals = summersubset.values
summersubset = pd.DataFrame(
summersubsetvals,
index=summersubset.coords['time'].to_index(),
columns=pd.MultiIndex.from_tuples(
[(summersubset.name, responsetimeagg, responseclustid)],
names=['variable', 'timeagg', 'clustid']))
reduced = response.groupby(clusterfield).mean('stacked_latitude_longitude')
reduced = reduced.sel(clustid=9) # In this case cluster 9 is western europe.
response.close()
del response
# Define variable / region combinations for the dummy problem
combinations = {
'sst_nhplus.anom.nc': get_natlantic(),
'z300_nhmin.anom.nc': get_europe()
}
#combinations = {'sst_nhplus.anom.nc':get_natlantic()}
# Define the time scale of the response (to be regressed to)
responsetimeagg = 15
responseagg = agg_time(array=reduced,
ndayagg=responsetimeagg,
method='mean',
rolling=True,
firstday=pd.Timestamp('1981-01-01'))
summersubset = responseagg[responseagg.time.dt.season == 'JJA']
summersubset.values = detrend(summersubset.values) # Detrend here?
summersubset.to_netcdf(
OUTDIR /
'.'.join(['response', str(responsetimeagg), 'nc'])) # Quick and dirty save
# Only rolling aggregation is possible for intercomparing timescales, as those are equally (daily) stamped
timeaggs = [1, 3, 5, 7, 9, 11, 15]
laglist = [-1, -3, -5, -7, -9, -11, -15, -20, -25, -30, -35, -40,
-45] # Eventually will be negative values
# first level loop is variable / block combinations
for inputfile, region in combinations.items():
# Investigate the precursors