Merge pull request #89 from NatLabRockies/bnb/gk2a

Bespoke data model for GK2A data. NASA and GK2A data models now subclass base model. Eventually this should be refactored to rely more on nsrdb_vars.csv for variable definitions.

Author: bnb32

nsrdb/data_model/variable_factory.py

@@ -24,6 +24,7 @@ class VarFactory:
     """Factory pattern to retrieve ancillary variable helper objects."""
 
     # mapping of NSRDB variable names to helper objects
+    # TODO: split this up into psm and mlcloud variables
     MAPPING: ClassVar = {
         'asymmetry': AsymVar,
         'air_temperature': MerraVar,

nsrdb/preprocessing/base_data_model.py

@@ -0,0 +1,380 @@
+"""Shared utilities for converting satellite data to UWISC format."""
+
+import logging
+import os
+import re
+from concurrent.futures import ThreadPoolExecutor, as_completed
+from functools import cached_property
+from glob import glob
+
+import numpy as np
+import pandas as pd
+from rex.utilities.solar_position import SolarPosition
+
+DROP_VARS = ['relative_time']
+
+UWISC_CLOUD_TYPE = {
+    'N/A': -15,
+    'Clear': 0,
+    'Probably Clear': 1,
+    'Fog': 2,
+    'Water': 3,
+    'Super-Cooled Water': 4,
+    'Mixed': 5,
+    'Opaque Ice': 6,
+    'Cirrus': 7,
+    'Overlapping': 8,
+    'Overshooting': 9,
+    'Unknown': 10,
+    'Dust': 11,
+    'Smoke': 12,
+}
+
+
+def expand_input_patterns(input_pattern):
+    """Expand one or more glob patterns into a de-duplicated file list."""
+    patterns = (
+        [input_pattern]
+        if isinstance(input_pattern, str)
+        else list(input_pattern)
+    )
+    files = []
+    for pattern in patterns:
+        files.extend(glob(pattern, recursive=True))
+    return list(dict.fromkeys(files))
+
+
+def run_data_model_jobs(
+    data_model_class,
+    input_pattern,
+    output_pattern,
+    *,
+    max_workers=None,
+    group_inputs=None,
+    logger=None,
+):
+    """Run a data-model conversion job over expanded input patterns."""
+    logger = logger or logging.getLogger(data_model_class.__module__)
+    files = expand_input_patterns(input_pattern)
+    job_inputs = group_inputs(files) if group_inputs is not None else files
+
+    if max_workers == 1:
+        for input_data in job_inputs:
+            data_model_class.run(input_data, output_pattern)
+    else:
+        with ThreadPoolExecutor(max_workers=max_workers) as executor:
+            futures = {}
+            for input_data in job_inputs:
+                future = executor.submit(
+                    data_model_class.run,
+                    input_data,
+                    output_pattern,
+                )
+                futures[future] = input_data
+
+            for future in as_completed(futures):
+                try:
+                    future.result()
+                except Exception as error:
+                    logger.error(
+                        'Error processing file(s): %s',
+                        futures[future],
+                    )
+                    logger.exception(error)
+
+    logger.info('Finished converting %s files.', len(files))
+
+
+class BaseUwiscDataModel:
+    """Shared preprocessing pipeline for UWISC-format conversion."""
+
+    NAME_MAP = {}
+    CLOUD_TYPE_MAP = {}
+    CLOUD_TYPE_SOURCE_VAR = None
+    TIMESTAMP_PATTERN = r'.*_([0-9]+).([0-9]+).([0-9]+).\w+'
+
+    def __init__(self, input_data, output_pattern):
+        self.input_data = input_data
+        self.output_pattern = output_pattern
+
+    @staticmethod
+    def _as_list(input_data):
+        """Normalize a single input or grouped inputs into a list."""
+        if isinstance(input_data, (list, tuple)):
+            return list(input_data)
+        return [input_data]
+
+    @cached_property
+    def input_files(self):
+        """Get the normalized list of input files."""
+        return self._as_list(self.input_data)
+
+    @classmethod
+    def get_primary_input_file(cls, input_files):
+        """Get the primary input file used for naming outputs."""
+        return input_files[0]
+
+    @cached_property
+    def primary_input_file(self):
+        """Get the primary input file used for naming outputs."""
+        return self.get_primary_input_file(self.input_files)
+
+    @classmethod
+    def parse_timestamp(cls, input_file):
+        """Parse a timestamp tuple from an input file path."""
+        ts = re.match(cls.TIMESTAMP_PATTERN, input_file).groups()
+        year, doy, hour = ts
+        minute = hour[2:] if len(hour) > 2 else '00'
+        hour = hour[:2]
+        secs = '000'
+        return year, doy, hour, minute, secs
+
+    @classmethod
+    def parse_timestamp_string(cls, input_file):
+        """Parse the output timestamp string from an input file path."""
+        return f's{"".join(cls.parse_timestamp(input_file))}'
+
+    @cached_property
+    def timestamp(self):
+        """Get the parsed timestamp tuple for the primary input file."""
+        return self.parse_timestamp(self.primary_input_file)
+
+    @cached_property
+    def timestamp_string(self):
+        """Get the parsed output timestamp string for the primary input."""
+        return self.parse_timestamp_string(self.primary_input_file)
+
+    @cached_property
+    def time_index(self):
+        """Get a single-step time index for the primary input file."""
+        year, doy, hour, minute, _ = self.timestamp
+        timestamp = pd.to_datetime(
+            f'{year}{doy}{hour}{minute}00', format='%Y%j%H%M%S'
+        )
+        return pd.DatetimeIndex([timestamp])
+
+    @cached_property
+    def output_file(self):
+        """Get output file name for the configured output pattern."""
+        year, doy, *_ = self.timestamp
+        return self.output_pattern.format(
+            year=year,
+            doy=doy,
+            timestamp=self.timestamp_string,
+        )
+
+    @classmethod
+    def open_dataset(cls, input_data):
+        """Open the raw input data as an xarray dataset."""
+        raise NotImplementedError
+
+    @cached_property
+    def ds(self):
+        """Get xarray dataset for raw input data."""
+        return self.open_dataset(self.input_data)
+
+    @classmethod
+    def transform_raw_data(cls, ds):
+        """Apply any subclass-specific preprocessing to the raw dataset."""
+        return ds
+
+    def get_solar_zenith(self, ds):
+        """Derive the solar zenith angle for the dataset."""
+        lats = ds['latitude'].values
+        lons = ds['longitude'].values
+        solar_pos = SolarPosition(
+            self.time_index,
+            np.column_stack((lats.ravel(), lons.ravel())),
+        )
+        return np.asarray(solar_pos.zenith).reshape(lats.shape)
+
+    def get_solar_azimuth(self, ds):
+        """Derive the solar azimuth angle for the dataset."""
+        lats = ds['latitude'].values
+        lons = ds['longitude'].values
+        solar_pos = SolarPosition(
+            self.time_index,
+            np.column_stack((lats.ravel(), lons.ravel())),
+        )
+        return np.asarray(solar_pos.azimuth).reshape(lats.shape)
+
+    @classmethod
+    def rename_vars(cls, ds):
+        """Rename variables to uwisc conventions."""
+        for current_name, target_name in cls.NAME_MAP.items():
+            if current_name in ds.data_vars:
+                ds = ds.rename({current_name: target_name})
+        return ds
+
+    @classmethod
+    def drop_vars(cls, ds):
+        """Drop variables that are not part of the UWISC output schema."""
+        for var_name in DROP_VARS:
+            if var_name in ds.data_vars:
+                ds = ds.drop_vars(var_name)
+        return ds
+
+    @staticmethod
+    def _rename_spatial_dims(ds):
+        """Rename alternative spatial dimensions to the shared convention."""
+        rename_map = {}
+        if 'Lines' in ds.dims:
+            rename_map['Lines'] = 'south_north'
+        if 'Pixels' in ds.dims:
+            rename_map['Pixels'] = 'west_east'
+        if 'dim_y' in ds.dims:
+            rename_map['dim_y'] = 'south_north'
+        if 'dim_x' in ds.dims:
+            rename_map['dim_x'] = 'west_east'
+        if rename_map:
+            ds = ds.rename(rename_map)
+        return ds
+
+    @staticmethod
+    def _promote_lat_lon_coords(ds):
+        """Ensure latitude and longitude are stored as coordinates."""
+        sdims = ('south_north', 'west_east')
+        if ('lat' in ds.coords or 'lat' in ds.data_vars) and (
+            'latitude' not in ds.coords and 'latitude' not in ds.data_vars
+        ):
+            ds = ds.rename({'lat': 'latitude', 'lon': 'longitude'})
+
+        if ds['latitude'].ndim == 1 and ds['longitude'].ndim == 1:
+            ds['south_north'] = ds['latitude']
+            ds['west_east'] = ds['longitude']
+
+            lons, lats = np.meshgrid(ds['longitude'], ds['latitude'])
+            ds = ds.assign_coords({
+                'latitude': (sdims, lats),
+                'longitude': (sdims, lons),
+            })
+
+        coord_names = [
+            name for name in ('latitude', 'longitude') if name in ds.data_vars
+        ]
+        if coord_names:
+            ds = ds.set_coords(coord_names)
+        return ds
+
+    def remap_dims(self, ds):
+        """Rename dims and coords to standards and build 2D lat/lon grids."""
+        for var_name in ds.data_vars:
+            single_ts = (
+                'time' in ds[var_name].dims
+                and ds[var_name].transpose('time', ...).shape[0] == 1
+            )
+            if single_ts and var_name != 'reference_time':
+                ds[var_name] = (
+                    ('south_north', 'west_east'),
+                    ds[var_name].isel(time=0).data,
+                )
+
+        ref_time = ds.attrs.get('reference_time', None)
+        if ref_time is not None:
+            time_index = pd.DatetimeIndex([ref_time]).values
+        else:
+            time_index = self.time_index.values
+        ds = self._rename_spatial_dims(ds)
+        ds = self._promote_lat_lon_coords(ds)
+        ds = ds.assign_coords({'time': ('time', time_index)})
+        return ds
+
+    @classmethod
+    def fill_missing_vars(cls, ds):
+        """Fill any missing variables with NaN arrays."""
+        for var_name in cls.NAME_MAP:
+            if var_name not in ds.data_vars:
+                ds[var_name] = (
+                    ('south_north', 'west_east'),
+                    np.full(
+                        (ds.sizes['south_north'], ds.sizes['west_east']),
+                        np.nan,
+                    ),
+                )
+        return ds
+
+    def derive_solar_angles(self, ds):
+        """Derive solar angles if not already present in the dataset."""
+        if 'solar_zenith_angle' not in ds.data_vars:
+            ds['solar_zenith_angle'] = (
+                ('south_north', 'west_east'),
+                self.get_solar_zenith(ds),
+            )
+        if 'solar_azimuth_angle' not in ds.data_vars:
+            ds['solar_azimuth_angle'] = (
+                ('south_north', 'west_east'),
+                self.get_solar_azimuth(ds),
+            )
+        return ds
+
+    @classmethod
+    def remap_cloud_phase(cls, ds):
+        """Map source cloud phase flags to UWISC cloud types."""
+        if cls.CLOUD_TYPE_SOURCE_VAR is None:
+            return ds
+
+        cloud_type_name = cls.NAME_MAP[cls.CLOUD_TYPE_SOURCE_VAR]
+        cloud_type = ds[cloud_type_name].values.copy()
+        for value, cloud_source in cls.CLOUD_TYPE_MAP.items():
+            cloud_type = np.where(
+                ds[cloud_type_name].values.astype(int) == int(value),
+                UWISC_CLOUD_TYPE[cloud_source],
+                cloud_type,
+            )
+        ds[cloud_type_name] = (ds[cloud_type_name].dims, cloud_type)
+        return ds
+
+    @classmethod
+    def derive_stdevs(cls, ds):
+        """Derive standard deviations used as training features."""
+        for var_name in ('refl_0_65um_nom', 'temp_11_0um_nom'):
+            stddev = (
+                ds[var_name]
+                .rolling(
+                    south_north=3,
+                    west_east=3,
+                    center=True,
+                    min_periods=1,
+                )
+                .std()
+            )
+            ds[f'{var_name}_stddev_3x3'] = stddev
+        return ds
+
+    def process_dataset(self, ds):
+        """Run the shared UWISC preprocessing pipeline on a dataset."""
+        ds = self.remap_dims(ds)
+        ds = self.fill_missing_vars(ds)
+        ds = self.transform_raw_data(ds)
+        ds = self.rename_vars(ds)
+        ds = self.drop_vars(ds)
+        ds = self.remap_cloud_phase(ds)
+        ds = self.derive_stdevs(ds)
+        ds = self.derive_solar_angles(ds)
+        return ds
+
+    @classmethod
+    def write_output(cls, ds, output_file):
+        """Write converted dataset to the final output file."""
+        os.makedirs(os.path.dirname(output_file), exist_ok=True)
+        ds = ds.transpose('south_north', 'west_east', ...)
+        ds.load().to_netcdf(output_file, format='NETCDF4', engine='h5netcdf')
+
+    @classmethod
+    def run(cls, input_data, output_pattern):
+        """Run the conversion routine and write the converted dataset."""
+        logger = logging.getLogger(cls.__module__)
+        data_model = cls(input_data, output_pattern)
+
+        if os.path.exists(data_model.output_file):
+            logger.info(
+                '%s already exists. Skipping conversion.',
+                data_model.output_file,
+            )
+            return
+
+        logger.info('Getting xarray dataset for %s', input_data)
+        ds = data_model.process_dataset(data_model.ds)
+
+        logger.info('Writing converted file to %s', data_model.output_file)
+        cls.write_output(ds, data_model.output_file)