def spi_gamma(precip_monthly_values,
month_scale,
valid_min,
valid_max):
return distribution_fitter.fit_to_gamma(precip_monthly_values,
month_scale,
valid_min,
valid_max)
def compute_indicator_by_lons_pooled(input_dataset,
output_dataset,
input_var_name,
output_var_name,
month_scale,
valid_min,
valid_max):
while True:
# get the longitude index we'll use from the worker queue
dim1_index = worker_queue.get()
# lock the thread when doing I/O
lock.acquire()
# slice out the period of record for the x/y point
data = input_dataset.variables[input_var_name][:, dim1_index, :]
# release the lock since we'll not share anything else until doing the I/O to the output dataset
lock.release()
# keep the original data shape, we'll use this to reshape later
original_shape = input_dataset.variables[input_var_name].shape
for dim2_index in range(input_dataset.variables[input_var_name].shape[2]):
# only process non-empty grid cells, i.e. data array contains at least some non-NaN values
if isinstance(data[:, dim2_index], np.ma.MaskedArray) and data[:, dim2_index].mask.all():
pass
else: # we have some valid values to work with
logger.info('Processing x/y {}/{}'.format(dim1_index, dim2_index))
# perform a fitting to gamma
data[:, dim2_index] = distribution_fitter.fit_to_gamma(data[:, dim2_index],
month_scale,
valid_min,
valid_max)
# reacquire the thread lock for doing NetCDF I/O
lock.acquire()
# slice out the period of record for the x/y point
output_dataset.variables[output_var_name][:, dim1_index, :] = np.reshape(data, (original_shape[0], 1, original_shape[2]))
# release the lock since we'll not share anything else until doing the I/O to the output dataset
lock.release()
# communicate to the queue that this task has been completed and the item can be removed
worker_queue.task_done()
def compute_indicator(input_dataset,
output_dataset,
input_var_name,
output_var_name,
month_scale,
valid_min,
valid_max,
dim1_index, # typically lon, for example with gridded datasets
dim2_index): # typically lat, for example with gridded datasets
# lock the thread when doing I/O
lock.acquire()
# slice out the period of record for the x/y point
data = input_dataset.variables[input_var_name][:, dim1_index, dim2_index]
# release the lock since we'll not share anything else until doing the I/O to the output dataset
lock.release()
# only process non-empty grid cells, i.e. data array contains at least some non-NaN values
if (isinstance(data, np.ma.MaskedArray)) and data.mask.all():
pass
# # create a filler data array for the time series which is all NaN values
# data = np.full((data.size(),), np.NaN)
else: # we have some valid values to work with
# logger.info('Processing x/y {}/{}'.format(dim1_index, dim2_index))
# perform a fitting to gamma
data = distribution_fitter.fit_to_gamma(data,
month_scale,
valid_min,
valid_max)
# reacquire the thread lock for doing NetCDF I/O
lock.acquire()
# slice out the period of record for the x/y point
output_dataset.variables[output_var_name][:, dim1_index, dim2_index] = data
# release the lock since we'll not share anything else until doing the I/O to the output dataset
lock.release()
def spei_gamma(precip_monthly_values,
temp_monthly_values,
data_start_year,
latitude,
month_scale,
valid_min,
valid_max):
# compute the PET values using Thornthwaite's equation
pet_monthly_values = thornthwaite(temp_monthly_values, latitude, data_start_year, np.nan)
# offset we add to the (P - PE values) in order to bring all values into the positive range
p_minus_pe_offset = 1000.0
# get the difference of P - PE, assign the values into the temperature array (in order to reuse the memory)
p_minus_pe = (precip_monthly_values - pet_monthly_values) + p_minus_pe_offset
# perform the SPEI computation (fit to the Gamma distribution) and assign the values into the dataset
return distribution_fitter.fit_to_gamma(p_minus_pe,
month_scale,
valid_min,
valid_max)