import json
from pywps import ComplexInput, ComplexOutput
from pywps import FORMATS
from pywps import LiteralInput
from pywps import Process
from pywps.app.Common import Metadata
import spotpy as sp
import xarray as xr
funcs = {f.__name__: f for f in sp.objectivefunctions._all_functions}
[docs]class ObjectiveFunctionProcess(Process):
def __init__(self):
inputs = [ComplexInput('obs', 'Stream flow observation',
abstract='Steam flow observation time series',
supported_formats=(FORMATS.NETCDF,)),
ComplexInput('sim', 'Stream flow simulation',
abstract='Stream flow simulation time series',
supported_formats=(FORMATS.NETCDF,)),
LiteralInput('name', 'Objective function name',
abstract="One or multiple objective function name. If None, defaults to all.",
data_type='string',
allowed_values=tuple(funcs.keys()),
default=None,
min_occurs=0,
max_occurs=17)
]
outputs = [ComplexOutput('metrics', 'Objective function values',
abstract="Returns up to 17 objective function values, depending on the user's "
"requests. By default all 17 are returned. JSON dictionary format.",
supported_formats=(FORMATS.JSON, )),
]
super(ObjectiveFunctionProcess, self).__init__(
self._handler,
identifier="objective-function",
title="Objective-function process based on SpotPy and its 17 objective functions.",
version="1.0",
abstract="This process takes two NETCDF files (one containing variable 'q_sim' and the other 'q_obs') "
"and computes objective-function metrics between them.",
metadata=[Metadata("SPOTPY Documentation", "http://fb09-pasig.umwelt.uni-giessen.de/spotpy/")],
inputs=inputs,
outputs=outputs,
keywords=["objective functions", "hydrological signatures", "optimization"] + list(funcs.keys()),
status_supported=True,
store_supported=True)
def _handler(self, request, response):
obs_fn = request.inputs['obs'][0].file
sim_fn = request.inputs['sim'][0].file
if 'name' in request.inputs:
names = [i.data for i in request.inputs['name']]
else:
names = funcs.keys()
obs = xr.open_dataset(obs_fn)
sim = xr.open_dataset(sim_fn)
# There is no support yet for handling NaN in SpotPy. Here we're starting from the second index to avoid missing
# values in the first time index for obs.
out = {}
for name in names:
da = xr.apply_ufunc(funcs[name],
obs['q_obs'].isel(time=slice(1, None)),
sim['q_sim'].isel(time=slice(1, None)),
input_core_dims=[['time', ], ['time', ]],
vectorize=True)
# For now we're assuming there is just one basin
out[name] = da.data[0]
response.outputs['metrics'].data = json.dumps(out)
return response
