from enstools.feature.identification.pv_streamer.processing import *
from enstools.feature.identification import IdentificationTechnique
import xarray as xr
import numpy as np
import copy
from enstools.feature.identification.pv_streamer.object_desc import get_object_data
from enstools.feature.identification.pv_streamer.projection_cdo import project_latlon_to_stereo
from enstools.feature.util.pb2_properties_api import ObjectProperties
class PVIdentification(IdentificationTechnique):
def __init__(self, unit='pv', mode_2d_layer=None, theta_range=None, extract_containing=None, centroid_lat_thr=None,
**kwargs):
"""
Initialize the PV Identification.
Parameters (experimental)
----------
unit
pv data unit: 'pv' or 'pvu'
mode_2d_layer
2d identification for given layer in dataset
theta_range
slice this range out of dataset to process
extract_containing
extract objects contains the selected layer (K) or overlaps with selected layer-range
centroid_lat_thr
only keep if centroid is north of this threshold
kwargs
"""
self.config = None # config
self.distance_map = None # precomputed distance map, reduced from 8 directions to one due to conformal projection
self.area_map = None # precomputed area map (km^2 per stereographic pixel)
# put custom args
self.pv_data_unit = unit
self.theta_range = theta_range
self.sel_layer = extract_containing
self.centroid_lat_thr = centroid_lat_thr # centroid < thr degree north -> discard
self.layer_2d_id = mode_2d_layer
pass
def precompute(self, dataset: xr.Dataset, **kwargs):
print("Precompute for PV identification...")
# init filter dataset, remove stuff we do not need, makes processing faster
from .data_util import compute_global_fields, filter_dataset
dataset = filter_dataset(dataset, self.theta_range, self.layer_2d_id)
# create configuration
from .configuration import DetectionConfig
self.config = DetectionConfig.config_from_dataset(dataset)
self.config.pv_data_unit = self.pv_data_unit
self.config.sel_layer = self.sel_layer
self.config.centroid_lat_thr = self.centroid_lat_thr
pv = dataset[self.config.pv_dim]
pv_tmp = xr.where(xr.ufuncs.logical_and(pv[self.config.latitude_dim] == 90, pv < -999), 10, pv)
dataset[self.config.pv_dim].values = pv_tmp.transpose(*pv.dims).values
# transform dataset to stereographic
stereo_ds = project_latlon_to_stereo(dataset)
# computes pvu, pvu2
stereo_ds = compute_global_fields(stereo_ds, self.config)
from .projection_cdo import precompute_distance_map_simple, precompute_area_map_simple
self.distance_map = precompute_distance_map_simple(stereo_ds)
self.area_map = precompute_area_map_simple(self.distance_map)
# if only 2D dataset, create third dimension for uniform handling
from enstools.feature.util.enstools_utils import get_vertical_dim, add_vertical_dim
if self.config.dims == 2:
add_vertical_dim(stereo_ds, inplace=True)
self.config.vertical_dim = get_vertical_dim(stereo_ds)
# internal fields (resp. output fields)
stereo_ds['streamer'] = xr.zeros_like(stereo_ds[self.config.pv_dim], dtype=int)
# stereo_ds['inner_PV'] = xr.zeros_like(stereo_ds[self.config.pv_dim], dtype=float) a debug mode
# TODO drop for this algorithm unneccessary fields. speedup later (drop pv?)
return stereo_ds
# main identify, called on spatial 2d/3d subsets from enstools-feature.
def identify(self, spatial_stereo_ds: xr.Dataset, object_block, **kwargs):
print("PV identify")
levels_list = spatial_stereo_ds.coords[self.config.vertical_dim].values
# preprocess binary fields: remove disturbances etc. in 2d, in 3d more flexible
if spatial_stereo_ds['pvu2'].data.ndim == 3 and spatial_stereo_ds['pvu2'].data.shape[0] > 1:
preprocessed_pvu2 = spatial_stereo_ds['pvu2'].data
else:
preprocessed_pvu2 = preprocess_pvu2(spatial_stereo_ds['pvu2'], self.config) # spatial_stereo_ds['pvu2'].data
# create distance map from outer contour using precomputed distance functions
dist_from_outer = create_dist_map_outer(preprocessed_pvu2, self.config, self.distance_map)
core_reservoir = (dist_from_outer > self.config.alg.w_bar)
# get data pole: maximum distance from boundary (center of reservoir)
pv_pole = get_data_pole(dist_from_outer)
# core_reservoir might have multiple areas: which one is main reservoir? flood biggest one (pole!)!
# also compute mask containing cutoffs (isolated areas)
# overwrite core_reservoir with only biggest one
core_reservoir, cutoff_mask = flood_reservoir(core_reservoir, preprocessed_pvu2, pv_pole)
if core_reservoir.ndim == 3 and core_reservoir.shape[0] > 1: # 3d keep top layer where growing can start
core_reservoir[-1] = preprocessed_pvu2[-1]
# ceate distance map from outer boundary of inner reservoir growing outwards, only in pvu>2 regions.
dist_expand = create_dist_map_inner(core_reservoir, preprocessed_pvu2, self.config, self.distance_map)
# set isolated areas as infinite distance from main reservoir
dist_expand[cutoff_mask] = np.inf
streamer_areas = (dist_expand > self.config.alg.w_bar * 1.05) # bit delta for subpixel inacc.
del core_reservoir
"""
from .plotting import plot_pvu2_binary, plot_erosion, plot_inner_binary, plot_dilation, plot_streamer_areas, \
plot_pvu2_pp, plot_erosion_animate, plot_dilation_animate
# plot_pvu2_binary(spatial_stereo_ds, self.config)
# plot_pvu2_pp(spatial_stereo_ds, preprocessed_pvu2, self.config)
# plot_erosion(spatial_stereo_ds, dist_from_outer, self.config)
# plot_erosion_animate(spatial_stereo_ds, dist_from_outer, self.config)
# plot_inner_binary(spatial_stereo_ds, core_reservoir, self.config)
# plot_dilation(spatial_stereo_ds, dist_expand, self.config)
# plot_dilation_animate(spatial_stereo_ds, dist_expand, self.config)
# plot_streamer_areas(spatial_stereo_ds, dist_expand, streamer_areas_exact, self.config)
# exit()
"""
# filter area based. here also 3d filtering strategy
streamer_areas = filter_area_based(streamer_areas, preprocessed_pvu2, self.config) # data_filtering
# label continuous 3d areas
labeled_areas = label_areas(streamer_areas)
del preprocessed_pvu2
del streamer_areas
# labeled_areas[~spatial_stereo_ds['pvu2']] = 0 # mask areas from holes closed before
spatial_stereo_ds['streamer'].values = labeled_areas
# generate object descriptions
object_properties_with_corres_indices = get_object_data(spatial_stereo_ds, levels_list, dist_expand, self.area_map, self.config)
# filter object descriptions
spatial_stereo_ds, object_properties_with_corres_indices = filter_object_based(spatial_stereo_ds, object_properties_with_corres_indices, self.config)
# squish filtered IDs.
spatial_stereo_ds, object_properties_with_corres_indices = squish(spatial_stereo_ds, object_properties_with_corres_indices)
for id_, obj in object_properties_with_corres_indices:
ObjectProperties.add_to_objects(object_block, obj, id=id_)
del dist_from_outer
del dist_expand
del cutoff_mask
"""
# TODO one debug mode: reservoir is reextended core. so new_PV v streamers = PV
spatial_stereo_ds['inner_PV'].values = copy.deepcopy(spatial_stereo_ds[
self.config.pv_dim].values)
spatial_stereo_ds['inner_PV'].values[spatial_stereo_ds['streamer'].values.astype(dtype=bool)] = 0
"""
del labeled_areas
return spatial_stereo_ds
def postprocess(self, dataset: xr.Dataset, pb2_desc, **kwargs):
dropped = dataset.drop_vars(['pvu', 'pvu2'])
# TODO remove vertical dim if dataset was 2D.
return dropped, pb2_desc