def single_location_comparison(location=[31.65, 77.34],
station='Banjar',
min_year=2000,
max_year=2011):
"""Plot model outputs for given coordinates over time."""
aphro_ds = aphrodite.collect_APHRO(location,
minyear=min_year,
maxyear=max_year)
cru_ds = cru.collect_CRU(location, minyear=min_year, maxyear=max_year)
era5_ds = era5.collect_ERA5(location, minyear=min_year, maxyear=max_year)
gpm_ds = gpm.collect_GPM(location, minyear=min_year, maxyear=max_year)
wrf_ds = beas_sutlej_wrf.collect_BC_WRF(location,
minyear=min_year,
maxyear=max_year)
gauge_ds = beas_sutlej_gauges.gauge_download(station,
minyear=min_year,
maxyear=max_year)
# cmip_ds = cmip5.collect_CMIP5()
# cordex_ds = cordex.collect_CORDEX()
# model_ts = model_prep([lat, lon], data_filepath='single_loc_test.csv', \
# model_filepath=model_filepath)
timeseries = [gauge_ds, gpm_ds, era5_ds, wrf_ds, aphro_ds, cru_ds]
tims.benchmarking_subplots(timeseries, reference_dataset=gauge_ds)
dataset_stats(timeseries, ref_ds=gauge_ds)
def gauge_stats():
"""Print mean, standard deviations and slope for datasets."""
bs_station_df = pd.read_csv('_Data/bs_only_gauge_info.csv')
'''
mlm_val_stations = ['Bhakra', 'Suni' 'Pandoh', 'Janjehl', 'Bhuntar',
'Rampur']
val_stations = ['Banjar', 'Larji', 'Bhuntar', 'Sainj',
'Bhakra', 'Kasol', 'Suni', 'Pandoh', 'Janjehl', 'Rampur']
'''
r2_list = []
rmse_list = []
for s in tqdm(bs_station_df):
gauge_ds = beas_sutlej_gauges.gauge_download(s,
minyear=2000,
maxyear=2011)
gauge_maxy = gauge_ds.time.max().values
gauge_miny = gauge_ds.time.min().values
miny = gauge_miny - 0.0001
maxy = gauge_maxy + 0.0001
location = bs_station_df[s].values
aphro_ds = aphrodite.collect_APHRO(location,
minyear=miny,
maxyear=maxy)
cru_ds = cru.collect_CRU(location, minyear=miny, maxyear=maxy)
era5_ds = era5.collect_ERA5(location, minyear=miny, maxyear=maxy)
gpm_ds = gpm.collect_GPM(location, minyear=miny, maxyear=maxy)
wrf_ds = beas_sutlej_wrf.collect_BC_WRF(location,
minyear=miny,
maxyear=maxy)
timeseries = [era5_ds, gpm_ds, aphro_ds, cru_ds, wrf_ds]
r2s, rmses = dataset_stats(timeseries, ref_ds=gauge_ds, ret=True)
r2_list.append(r2s)
rmse_list.append(rmses)
avg_r2 = np.array(r2_list).mean(axis=0)
avg_rmse = np.array(rmse_list).mean(axis=0)
return avg_r2, avg_rmse
(station_df['lat'] > 31)
& (station_df['lat'] < 31.23)]
hf_train_df5 = station_df[(station_df['lon'] > 78.2)]
# + list(hf_train_df3['index'].values) + list(hf_train_df4['index'].values)
# + list(hf_train_df5['index'].values)
hf_train_stations = list(hf_train_df5['index'].values) + list(
hf_train_df2['index'].values)
# + (['Banjar', 'Larji', 'Bhuntar', 'Sainj', 'Bhakra', 'Kasol', 'Suni',
# 'Pandoh', 'Janjehl', 'Rampur'])
lf_train_stations = hf_train_stations
hf_val_stations = hf_train_stations[0]
hf_train_list = []
for station in hf_train_stations:
station_ds = beas_sutlej_gauges.gauge_download(station,
minyear=1980,
maxyear=2010)
station_ds['z'] = station_df[station].values[2]
station_ds['slope'] = srtm.find_slope(station).slope.values
station_ds = station_ds.set_coords('z')
station_ds = station_ds.set_coords('slope')
station_ds = station_ds.expand_dims(dim={
'lat': 1,
'lon': 1,
'z': 1,
'slope': 1
})
hf_train_list.append(station_ds)
hf_train_ds = xr.merge(hf_train_list)
# ERA5 data
hf_train_df3 = station_df[(station_df['lon'] > 77.0)
& (station_df['lat'] < 31)]
hf_train_df4 = station_df[(station_df['lon'] < 78.0)
& (station_df['lon'] > 77.0) &
(station_df['lat'] > 31)
& (station_df['lat'] < 31.23)]
hf_train_df5 = station_df[(station_df['lon'] > 78.2)]
# + list(hf_train_df2['index'].values) + list(hf_train_df3['index'].values)
# + list(hf_train_df4['index'].values) + list(hf_train_df5['index'].values)
hf_train_stations = list(hf_train_df5['index'].values)
hf_train_list = []
for station in hf_train_stations:
station_ds = beas_sutlej_gauges.gauge_download(station,
minyear=2000,
maxyear=2005)
station_ds['z'] = station_df[station].values[2]
station_ds['slope'] = srtm.find_slope(station).slope.values
station_ds = station_ds.set_coords('z')
station_ds = station_ds.set_coords('slope')
station_ds = station_ds.expand_dims(dim={
'lat': 1,
'lon': 1,
'z': 1,
'slope': 1
})
hf_train_list.append(station_ds)
hf_train_ds = xr.merge(hf_train_list)
hf_train_df = hf_train_ds.to_dataframe().dropna().reset_index()
model.mean_function, model.Y_metadata)
return test_log_likelihood
hf_heldout_ll = []
hf_train_ll = []
R2_hf_list = []
R2_lf_list = []
# Calculate values
for station in station_list:
# Prepare data
hf_ds = beas_sutlej_gauges.gauge_download(station,
minyear=1980,
maxyear=2011)
lf_ds = era5.gauge_download(station, minyear=1980, maxyear=2011)
hf_ds = hf_ds.assign_coords({"time": lf_ds.time[:len(hf_ds.time.values)]})
hf_ds = hf_ds.dropna(dim='time')
lf_ds = lf_ds.dropna(dim='time')
# Transform data
hf_ds['tp_tr'] = dp.log_transform(hf_ds['tp'].values)
lf_ds['tp_tr'] = dp.log_transform(lf_ds['tp'].values)
# High fidelity data needs to be smaller in length then low fidelity data
x_train_l = lf_ds.time[:330].values.reshape(-1, 1)
x_train_h = hf_ds.time[:240].values.reshape(-1, 1)