def array_juggling(data, month, hour=None):
m = month
if hour != None:
data = data[((data['time.month'] >= 3) & (data['time.month'] <= 5))
& (data['time.hour'] == hour)]
else:
data = data[(data['time.month'] >= 3) & (data['time.month'] <= 5)]
data_years = data.groupby('time.year').mean(axis=0)
# stack lat and lon into a single dimension called allpoints
datastacked = data_years.stack(allpoints=['latitude', 'longitude'])
# apply the function over allpoints to calculate the trend at each point
print('Entering trend calc')
dtrend = datastacked.groupby('allpoints').apply(
u_darrays.linear_trend_mk, alpha=0.05, eps=0.0001)
ddtrend = dtrend['slope']
try:
pthresh = us.fdr_threshold(dtrend['pval'].values[np.isfinite(
dtrend['pval'].values)],
alpha=0.05)
ddtrend.values[(dtrend['pval'].values > pthresh)
| np.isnan(dtrend['pval'].values)] = np.nan
except ValueError:
ddtrend.values = ddtrend.values * np.nan
pthresh = np.nan
print('p value threshold', pthresh)
# unstack back to lat lon coordinates
return ddtrend.unstack('allpoints'), data_years
def calc_trend(data, month, hour=None, method=None, sig=False, wilks=False):
y0 = 1980
if method is None:
'Please provide trend calc method: polyfit or mk (mann kendall)'
if hour is not None:
if len(month)>1:
data = data[((data['time.month'] >= month[0]) & (data['time.month'] <= month[1])) & (data['time.hour'] == hour) & (data['time.year'] >= y0) & (data['time.year'] <= 2018)]
else:
data = data[(data['time.month'] == month[0]) & (data['time.hour'] == hour) & (data['time.year'] >= y0) & (data['time.year'] <= 2018)]
else:
if len(month)>1:
data = data[((data['time.month'] >= month[0]) & (data['time.month'] <= month[1]))& (data['time.year'] >= y0) & (data['time.year'] <= 2018)]
else:
data = data[(data['time.month'] == month[0]) & (data['time.year'] >= y0) & (data['time.year'] <= 2018)]
if len(data.time)==0:
print('Data does not seem to have picked month or hour. Please check input data')
mean_years = data.groupby('time.year').mean(dim='time').squeeze().load()
# stack lat and lon into a single dimension called allpoints
datastacked = mean_years.stack(allpoints=['lat', 'lon'])
# apply the function over allpoints to calculate the trend at each point
print('Entering trend calc')
alpha = 0.05
# NaNs means there is not enough data, slope = 0 means there is no significant trend.
if method=='mk':
dtrend = datastacked.groupby('allpoints').apply(u_darrays.linear_trend_mk, alpha=alpha, eps=0.0001,nb_missing=10)
dtrend = dtrend.unstack('allpoints')
if sig:
(dtrend['slope'].values)[dtrend['ind'].values==0] = 0
# NaNs means there is not enough data, slope = 0 means there is no significant trend.
if method=='polyfit':
dtrend = datastacked.groupby('allpoints').apply(u_darrays.linear_trend_lingress,nb_missing=10)
dtrend = dtrend.unstack('allpoints')
if sig:
(dtrend['slope'].values)[dtrend['pval'].values > alpha] = 0
ddtrend = dtrend['slope']
if wilks and sig:
try:
pthresh = us.fdr_threshold(dtrend['pval'].values[np.isfinite(dtrend['pval'].values)], alpha=alpha)
ddtrend.values[(dtrend['pval'].values > pthresh) | np.isnan(dtrend['pval'].values)] = np.nan
except ValueError:
ddtrend.values = ddtrend.values * np.nan
pthresh = np.nan
print('p value threshold', pthresh)
# unstack back to lat lon coordinates
return ddtrend, mean_years
def calc_trend(data, month, hour=None, method=None, sig=False, wilks=False):
if method is None:
'Please provide trend calc method: polyfit or mk (mann kendall)'
if hour is not None:
if len(month)>1:
data = data[((data['time.month'] >= month[0]) | (data['time.month'] <= month[1])) & (data['time.hour'] == hour) & (data['time.year'] >= 1983) & (data['time.year'] <= 2017)]
else:
data = data[(data['time.month'] == month[0]) & (data['time.hour'] == hour) & (data['time.year'] >= 1983) & (data['time.year'] <= 2017)]
else:
if len(month)>1:
data = data[((data['time.month'] >= month[0]) | (data['time.month'] <= month[1]))& (data['time.year'] >= 1983) & (data['time.year'] <= 2017)]
else:
data = data[(data['time.month'] == month[0]) & (data['time.year'] >= 1983) & (data['time.year'] <= 2017)]
if len(data.time)==0:
print('Data does not seem to have picked month or hour. Please check input data')
mean_years = data.groupby('time.year').mean(axis=0)
# stack lat and lon into a single dimension called allpoints
datastacked = mean_years.stack(allpoints=['latitude', 'longitude'])
# apply the function over allpoints to calculate the trend at each point
print('Entering trend calc')
alpha = 0.05
# NaNs means there is not enough data, slope = 0 means there is no significant trend.
if method=='mk':
dtrend = datastacked.groupby('allpoints').apply(u_darrays.linear_trend_mk, alpha=alpha, eps=0.01,nb_missing=5)
dtrend = dtrend.unstack('allpoints')
if sig:
(dtrend['slope'].values)[dtrend['ind'].values==0] = 0
# NaNs means there is not enough data, slope = 0 means there is no significant trend.
if method=='polyfit':
dtrend = datastacked.groupby('allpoints').apply(u_darrays.linear_trend_lingress,nb_missing=10)
dtrend = dtrend.unstack('allpoints')
if sig:
(dtrend['slope'].values)[dtrend['pval'].values > alpha] = 0
ddtrend = dtrend['slope']
if wilks and sig:
try:
pthresh = us.fdr_threshold(dtrend['pval'].values[np.isfinite(dtrend['pval'].values)], alpha=alpha)
ddtrend.values[(dtrend['pval'].values > pthresh) | np.isnan(dtrend['pval'].values)] = np.nan
except ValueError:
ddtrend.values = ddtrend.values * np.nan
pthresh = np.nan
print('p value threshold', pthresh)
# unstack back to lat lon coordinates
return ddtrend, mean_years
def calc_trend(data, month, hour=None, method=None, sig=False, wilks=False):
if method is None:
'Please provide trend calc method: polyfit or mk (mann kendall)'
if hour is not None:
if len(month) > 1:
data = data[((data['time.month'] >= month[0])
& (data['time.month'] <= month[1]))
& (data['time.hour'] == hour) &
(data['time.year'] >= 1983) &
(data['time.year'] <= 2017)]
else:
data = data[(data['time.month'] == month[0])
& (data['time.hour'] == hour) &
(data['time.year'] >= 1983) &
(data['time.year'] <= 2017)]
else:
if len(month) > 1:
data = data[((data['time.month'] >= month[0])
& (data['time.month'] <= month[1]))
& (data['time.year'] >= 1983) &
(data['time.year'] <= 2017)]
else:
data = data[(data['time.month'] == month[0])
& (data['time.year'] >= 1983) &
(data['time.year'] <= 2017)]
if len(data.time) == 0:
print(
'Data does not seem to have picked month or hour. Please check input data'
)
#ipdb.set_trace()
mean_years = data.groupby('time.year').mean('time')
highpos = 0
if mean_years.name == 'z':
highpos = []
three = mean_years.coarsen(year=11, boundary='trim').mean()
for my in three:
pos = np.unravel_index(np.argmax(my.values), my.values.shape)
#ipdb.set_trace()
mlat = my.latitude.values[pos[0]]
mlon = my.longitude.values[pos[1]]
highpos.append((mlon, mlat))
if mean_years.name == 'u200':
three = mean_years.coarsen(year=11, boundary='trim').mean()
three.values = np.abs(three.values)
# ipdb.set_trace()
highpos = three.argmin('latitude')
# stack lat and lon into a single dimension called allpoints
datastacked = mean_years.stack(allpoints=['latitude', 'longitude'])
# apply the function over allpoints to calculate the trend at each point
print('Entering trend calc')
alpha = 0.05
# NaNs means there is not enough data, slope = 0 means there is no significant trend.
if method == 'mk':
dtrend = datastacked.groupby('allpoints').apply(
u_darrays.linear_trend_mk, alpha=alpha, eps=0.01, nb_missing=10)
dtrend = dtrend.unstack('allpoints')
if sig:
(dtrend['slope'].values)[dtrend['ind'].values == 0] = 0
# NaNs means there is not enough data, slope = 0 means there is no significant trend.
if method == 'polyfit':
dtrend = datastacked.groupby('allpoints').apply(
u_darrays.linear_trend_lingress, nb_missing=10)
dtrend = dtrend.unstack('allpoints')
if sig:
(dtrend['slope'].values)[dtrend['pval'].values > alpha] = 0
ddtrend = dtrend['slope']
if wilks and sig:
try:
pthresh = us.fdr_threshold(dtrend['pval'].values[np.isfinite(
dtrend['pval'].values)],
alpha=alpha)
ddtrend.values[(dtrend['pval'].values > pthresh)
| np.isnan(dtrend['pval'].values)] = np.nan
except ValueError:
ddtrend.values = ddtrend.values * np.nan
pthresh = np.nan
print('p value threshold', pthresh)
# unstack back to lat lon coordinates
return ddtrend, mean_years, highpos