TMY versus Single-Year

TMYs are built from monthly segments pulled from actual (single-year) data. As such, one might expect each month in a TMY dataset to be identical to the same month in the corresponding single-year data. This notebook demonstrates that this expectation does not exactly hold for PSM3.

import pvlib
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt

First, let’s fetch a normal TMY file. I specify names='tmy-2020' so that this notebook fetches the same TMY data in the future.

kwargs = dict(latitude=40, longitude=-80, api_key='DEMO_KEY', email='assessingsolar@gmail.com',
              map_variables=True, leap_day=True)
real_tmy = pvlib.iotools.get_psm3(names='tmy-2020', **kwargs)[0].drop(columns=['Year', 'Month', 'Day', 'Hour', 'Minute'])

real_tmy.head(10)

	pressure	albedo	wind_speed	temp_air	Dew Point	dni	ghi	wind_direction	dhi
1999-01-01 00:30:00-05:00	970.0	0.87	0.6	-6.0	-10.0	0.0	0.0	256.0	0.0
1999-01-01 01:30:00-05:00	970.0	0.87	0.6	-7.0	-10.0	0.0	0.0	258.0	0.0
1999-01-01 02:30:00-05:00	970.0	0.87	0.6	-7.0	-11.0	0.0	0.0	262.0	0.0
1999-01-01 03:30:00-05:00	970.0	0.87	0.6	-7.0	-12.0	0.0	0.0	265.0	0.0
1999-01-01 04:30:00-05:00	970.0	0.87	0.6	-7.0	-12.0	0.0	0.0	270.0	0.0
1999-01-01 05:30:00-05:00	970.0	0.87	0.6	-8.0	-13.0	0.0	0.0	275.0	0.0
1999-01-01 06:30:00-05:00	970.0	0.87	0.6	-8.0	-13.0	0.0	0.0	280.0	0.0
1999-01-01 07:30:00-05:00	970.0	0.87	0.6	-8.0	-14.0	0.0	0.0	280.0	0.0
1999-01-01 08:30:00-05:00	980.0	0.87	0.6	-8.0	-14.0	497.0	98.0	279.0	39.0
1999-01-01 09:30:00-05:00	980.0	0.87	0.6	-7.0	-14.0	712.0	258.0	283.0	72.0

Now, let’s build the same TMY using the corresponding single-year data:

dfs = []
for year in np.unique(real_tmy.index.year):
    dfs.append(pvlib.iotools.get_psm3(names=year, **kwargs)[0].drop(columns=['Year', 'Month', 'Day', 'Hour', 'Minute']))

df = pd.concat(dfs)
fake_tmy = df.reindex(real_tmy.index)

fake_tmy.head(10)

	pressure	albedo	wind_speed	temp_air	Dew Point	dni	ghi	wind_direction	dhi
1999-01-01 00:30:00-05:00	970.0	0.866	0.6	-6.0	-10.0	0.0	0.0	256.3	0.0
1999-01-01 01:30:00-05:00	970.0	0.866	0.6	-7.0	-10.0	0.0	0.0	258.3	0.0
1999-01-01 02:30:00-05:00	970.0	0.866	0.6	-7.0	-11.0	0.0	0.0	261.5	0.0
1999-01-01 03:30:00-05:00	970.0	0.866	0.6	-7.0	-12.0	0.0	0.0	265.1	0.0
1999-01-01 04:30:00-05:00	970.0	0.866	0.6	-7.0	-12.0	0.0	0.0	269.7	0.0
1999-01-01 05:30:00-05:00	970.0	0.866	0.6	-8.0	-13.0	0.0	0.0	275.0	0.0
1999-01-01 06:30:00-05:00	970.0	0.866	0.6	-8.0	-13.0	0.0	0.0	279.6	0.0
1999-01-01 07:30:00-05:00	970.0	0.866	0.6	-8.0	-14.0	0.0	0.0	280.4	0.0
1999-01-01 08:30:00-05:00	980.0	0.866	0.6	-8.0	-14.0	497.0	98.0	279.4	39.0
1999-01-01 09:30:00-05:00	980.0	0.866	0.6	-7.0	-14.0	712.0	258.0	283.2	72.0

Right away we can see that wind_direction is somewhat different. Note also that albedo is different, although it’s a precision difference rather than truly different values. Let’s take a broader look:

columns = ['ghi', 'dni', 'dhi', 'temp_air', 'wind_speed',
           'wind_direction', 'pressure', 'albedo', 'Dew Point']

fig, axes = plt.subplots(3, 3, figsize=(6, 6))
axes = axes.ravel()

for ax, col in zip(axes, columns):
    ax.scatter(real_tmy[col], fake_tmy[col], s=4)
    ax.set_title(col)
    ax.set_xlabel("PSM3 TMY")
    ax.set_ylabel("PSM3 Single-Year")
    x = real_tmy[col].mean()
    ax.axline((x, x), slope=1)

fig.tight_layout()

How often do the values not match up?

(real_tmy != fake_tmy)[columns].mean()

ghi               0.000228
dni               0.000228
dhi               0.000228
temp_air          0.005594
wind_speed        0.003995
wind_direction    0.499543
pressure          0.003653
albedo            0.792922
Dew Point         0.006279
dtype: float64

And what about just during the daytime?

mask = real_tmy['ghi'] > 0
(real_tmy.loc[mask] != fake_tmy.loc[mask])[columns].mean()

ghi               0.000460
dni               0.000460
dhi               0.000460
temp_air          0.000230
wind_speed        0.000230
wind_direction    0.412319
pressure          0.000230
albedo            0.774764
Dew Point         0.000460
dtype: float64

%load_ext watermark
%watermark --iversions -u -d -t

Last updated: 2022-10-16 17:29:59

pandas    : 1.5.0
matplotlib: 3.5.2
numpy     : 1.22.3
pvlib     : 0.9.3