iceds-data / ensemble_data.py
charlieoneill's picture
Upload folder using huggingface_hub
8d9d827 verified
from tqdm import tqdm
import requests
import os
import xarray as xr
import pandas as pd
from glob import glob
import cfgrib
# ### GLOSEA5 ###
# path = "raw/glosea.grib"
# print("Processing Glosea5...")
# # Open the GRIB file as an xarray dataset
# ds_1 = cfgrib.open_datasets("raw/glosea_1.grib")
# ds_2 = cfgrib.open_datasets("raw/glosea_2.grib")
# ds_3 = cfgrib.open_datasets("raw/glosea_3.grib")
# glosea_xarray_1, glosea_xarray_2, glosea_xarray_3 = ds_1[0], ds_2[0], ds_3[0]
# # Convert to pandas DataFrame
# df_1 = glosea_xarray_1.to_dataframe().reset_index()
# df_2 = glosea_xarray_2.to_dataframe().reset_index()
# df_3 = glosea_xarray_3.to_dataframe().reset_index()
# # Concatenate the two DataFrames
# glosea_df = pd.concat([df_1, df_2, df_3], ignore_index=True)
# # Convert tprate NaN to 0.0
# glosea_df['tprate'] = glosea_df['tprate'].fillna(0.0)
# # Aggregate across steps
# glosea_df = glosea_df.groupby(['valid_time', 'latitude', 'longitude'])['tprate'].mean().reset_index()
# # Keep between 1981 and 2019
# glosea_df = glosea_df[glosea_df['valid_time'].between('1981-01-01', '2018-12-31')]
# # Rename columns
# glosea_df.rename(columns={'tprate': 'pr', 'latitude': 'lat', 'longitude': 'lon', 'valid_time': 'time'}, inplace=True)
# glosea_df.reset_index(inplace=True, drop=True)
# print(f'Saving Glosea5 data to parquet (length of dataframe = {len(glosea_df)})...')
# glosea_df.to_parquet('processed/glosea5.parquet')
# # Delete any files ending in .idx from the raw folder
# for file in os.listdir("raw"):
# if file.endswith(".idx"):
# os.remove(os.path.join("raw", file))
# ### ECMWF ###
# path = "raw/ecmwf.grib"
# print("Processing ECMWF...")
# # Open the GRIB file as an xarray dataset
# ds_1 = cfgrib.open_datasets("raw/ecmwf_1.grib")
# ds_2 = cfgrib.open_datasets("raw/ecmwf_2.grib")
# ds_3 = cfgrib.open_datasets("raw/ecmwf_3.grib")
# ecmwf_xarray_1, ecmwf_xarray_2, ecmwf_xarray_3 = ds_1[0], ds_2[0], ds_3[0]
# # Convert to pandas DataFrame
# df_1 = ecmwf_xarray_1.to_dataframe().reset_index()
# df_2 = ecmwf_xarray_2.to_dataframe().reset_index()
# df_3 = ecmwf_xarray_3.to_dataframe().reset_index()
# # Concatenate the two DataFrames
# ecmwf_df = pd.concat([df_1, df_2, df_3], ignore_index=True)
# # Convert tprate NaN to 0.0
# ecmwf_df['tprate'] = ecmwf_df['tprate'].fillna(0.0)
# # Aggregate across steps
# ecmwf_df = ecmwf_df.groupby(['valid_time', 'latitude', 'longitude'])['tprate'].mean().reset_index()
# # Keep between 1981 and 2019
# ecmwf_df = ecmwf_df[ecmwf_df['valid_time'].between('1981-01-01', '2018-12-31')]
# # Rename columns
# ecmwf_df.rename(columns={'tprate': 'pr', 'latitude': 'lat', 'longitude': 'lon', 'valid_time': 'time'}, inplace=True)
# ecmwf_df.reset_index(inplace=True, drop=True)
# print(f'Saving ECMWF data to parquet (length of dataframe = {len(ecmwf_df)})...')
# ecmwf_df.to_parquet('processed/ecmwf.parquet')
# # Delete any files ending in .idx from the raw folder
# for file in os.listdir("raw"):
# if file.endswith(".idx"):
# os.remove(os.path.join("raw", file))
# ### ACCESS-S2 ###
# # Define the output file path
# output_file = 'processed/access.parquet'
# # Define the path and file pattern
# path = "/g/data/ux62/access-s2/hindcast/calibrated/atmos/pr/daily/e09/"
# # Check if the output file already exists and read it if it does
# if os.path.exists(output_file):
# master_df = pd.read_parquet(output_file)
# # Extract already processed years
# processed_years = master_df['time'].dt.year.unique()
# else:
# # Initialise an empty DataFrame if the file does not exist
# master_df = pd.DataFrame()
# processed_years = []
# # Generate file patterns for each year from 1983 to 2018 and get matching files
# files = []
# for year in range(1983, 2018):
# if year not in processed_years:
# pattern = f"*pr_{year}*.nc"
# files.extend(glob(os.path.join(path, pattern)))
# print(f"Processing data for years: {set(range(1983, 2018)) - set(processed_years)}")
# # Loop through the list of files and load each one
# for file in tqdm(files):
# # Load the xarray dataset
# ds = xr.open_dataset(file)
# # Slice the dataset for three specific lat/lon grids
# ds_sliced1 = ds.sel(lon=slice(142, 145), lat=slice(-25, -22))
# ds_sliced2 = ds.sel(lon=slice(150, 153), lat=slice(-29, -26))
# ds_sliced3 = ds.sel(lon=slice(143, 146), lat=slice(-20, -17))
# # Flatten the sliced data for 'pr' variable for both slices
# df1 = ds_sliced1['pr'].to_dataframe().reset_index()
# df2 = ds_sliced2['pr'].to_dataframe().reset_index()
# df3 = ds_sliced3['pr'].to_dataframe().reset_index()
# # Concatenate the two DataFrames
# combined_df = pd.concat([df1, df2, df3], ignore_index=True)
# # Filter rows where latitude and longitude are integers
# combined_df = combined_df[combined_df['lat'].apply(lambda x: x.is_integer())]
# combined_df = combined_df[combined_df['lon'].apply(lambda x: x.is_integer())]
# # Append the DataFrame to the master DataFrame
# master_df = pd.concat([master_df, combined_df], ignore_index=True)
# # Close the xarray dataset
# ds.close()
# # Save the updated master_df to the Parquet file
# os.makedirs(os.path.dirname(output_file), exist_ok=True)
# master_df.to_parquet(output_file)
# # Print the processed year for tracking
# processed_year = pd.to_datetime(master_df['time'].max()).year
# print(f"Year {processed_year} processed and saved.")
# # After the loop, perform any final processing needed on master_df
# if os.path.exists(output_file):
# master_df = pd.read_parquet(output_file)
# master_df['time'] = pd.to_datetime(master_df['time'])
# # Group by time, lat, and lon, then sum the pr values
# deduped_df = master_df.groupby(['time', 'lat', 'lon']).agg({'pr': 'sum'}).reset_index()
# # Save the final processed DataFrame to a Parquet file
# deduped_df.to_parquet(output_file)
# print(f"Final file saved to {output_file}")
# else:
# print(f"No data processed. {output_file} does not exist.")
# def create_master_parquet():
# files = ['access', 'ecmwf', 'glosea5', 'silo']
# frames = []
# for file in files:
# df = pd.read_parquet(f'processed/{file}.parquet')
# df['model'] = file
# frames.append(df)
# access = frames[0]
# access.reset_index(inplace=True, drop=True)
# columns = access.columns
# # Convert time to string
# access['time'] = access['time'].astype(str)
# ecmwf = frames[1]
# ecmwf.rename(columns={'date': 'time', 'precip': 'pr', 'latitude': 'lat', 'longitude': 'lon'}, inplace=True)
# ecmwf = ecmwf[columns]
# ecmwf.reset_index(inplace=True, drop=True)
# # Convert time to string
# ecmwf['time'] = ecmwf['time'].astype(str)
# glosea = frames[2]
# glosea.rename(columns={'date': 'time', 'tprate': 'pr', 'latitude': 'lat', 'longitude': 'lon'}, inplace=True)
# glosea = glosea[columns]
# glosea.reset_index(inplace=True, drop=True)
# # Convert time to string
# glosea['time'] = glosea['time'].astype(str)
# silo = frames[3]
# silo.rename(columns={'daily_rain': 'pr'}, inplace=True)
# silo = silo[columns]
# # Convert lat and lon to float32
# silo['lat'] = silo['lat'].astype('float32')
# silo['lon'] = silo['lon'].astype('float32')
# silo.reset_index(inplace=True, drop=True)
# # Convert time to string
# silo['time'] = silo['time'].astype(str)
# dfs = [access, ecmwf, glosea, silo]
# master_df = pd.concat(dfs)
# master_df.reset_index(inplace=True, drop=True)
# master_df.to_parquet('processed/master.parquet')
# print(f"Final file saved to processed/master.parquet")
import pandas as pd
def standardize_df(df, rename_dict, default_columns):
"""Standardize the DataFrame structure."""
df = df.rename(columns=rename_dict)
df = df[default_columns]
df.reset_index(inplace=True, drop=True)
df['time'] = df['time'].astype(str)
return df
def create_master_parquet():
files = ['access', 'ecmwf', 'glosea5', 'silo']
rename_dicts = [
{},
{'date': 'time', 'precip': 'pr', 'latitude': 'lat', 'longitude': 'lon'},
{'date': 'time', 'tprate': 'pr', 'latitude': 'lat', 'longitude': 'lon'},
{'daily_rain': 'pr'}
]
# Read and append the 'model' column to each DataFrame
frames = []
for file, rename_dict in zip(files, rename_dicts):
df = pd.read_parquet(f'processed/{file}.parquet')
df['model'] = file
df = standardize_df(df, rename_dict, default_columns=['time', 'lat', 'lon', 'pr', 'model'])
frames.append(df)
# Use the first DataFrame (access) as a template for column names
columns = frames[0].columns
# Standardize each DataFrame
for i in range(1, len(frames)):
frames[i] = standardize_df(frames[i], rename_dicts[i], columns)
master_df = pd.concat(frames)
master_df.reset_index(inplace=True, drop=True)
master_df.to_parquet('processed/master.parquet')
print("Final file saved to processed/master.parquet")
create_master_parquet()