AssessingSolar / unofficial-psm3-userguide

https://assessingsolar.github.io/unofficial-psm3-userguide
BSD 3-Clause "New" or "Revised" License
1 stars 1 forks source link

Add page on cloud detection performance at low sun elevation #19

Open kandersolar opened 11 months ago

kandersolar commented 11 months ago

Courtesy of @williamhobbs:

image

The sharp “horns” that are noticeable during sunrise and sunset in the BON data in Fig. 5 are due to the false negative identification of clouds by the Bayesian cloud classification algorithm (Heidinger et al. 2012) at extreme solar zenith angles. This is a common feature in the NSRDB and can be visualized in animated maps where clouds can be seen to “blink” into existence as the sun rises.

Source: https://doi.org/10.1016/j.solener.2022.01.004

Other examples: the GCR SURFRAD station around these dates: 2021-01-09, 2021-02-01, 2021-02-08

williamhobbs commented 11 months ago

Thanks for starting this issue. Here are plots of those date ranges, with code below. They just happen to be the first dates I noticed. I haven't checked to make sure that timestamp conventions are correct, but the concept is clear either way.

image

image

image

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

lat, lon = 34.2547, -89.8729 # Goodwin Creek SURFRAD
nsrdb_data, _ = pvlib.iotools.get_psm3(latitude=lat, longitude=lon,
                                       api_key='DEMO_KEY', email='your.email@gmail.com',
                                       names='2021', interval=30)

# convert to UTC
nsrdb_data.index = nsrdb_data.index.tz_convert('UTC')

surfrad_data1, _ = pvlib.iotools.read_surfrad(filename='https://gml.noaa.gov/aftp/data/radiation/surfrad/Goodwin_Creek_MS/2021/gwn21008.dat')
surfrad_data2, _ = pvlib.iotools.read_surfrad(filename='https://gml.noaa.gov/aftp/data/radiation/surfrad/Goodwin_Creek_MS/2021/gwn21009.dat')
surfrad_data = pd.concat([surfrad_data1,surfrad_data2])
surfrad_30m = surfrad_data.resample('30T').mean()
mask = (nsrdb_data.index > '2021-01-08 10:00') & (nsrdb_data.index < '2021-01-10 02:00')
plt.plot(nsrdb_data.GHI[mask], drawstyle='steps', label='nsrdb ghi')
plt.plot(surfrad_30m.ghi, drawstyle='steps', label='surfrad ghi')
plt.legend()
plt.show()

surfrad_data3, _ = pvlib.iotools.read_surfrad(filename='https://gml.noaa.gov/aftp/data/radiation/surfrad/Goodwin_Creek_MS/2021/gwn21032.dat')
surfrad_30m = surfrad_data3.resample('30T').mean()
mask = (nsrdb_data.index > '2021-02-01 10:00') & (nsrdb_data.index < '2021-02-02 02:00')
plt.plot(nsrdb_data.GHI[mask], drawstyle='steps', label='nsrdb ghi')
plt.plot(surfrad_30m.ghi, drawstyle='steps', label='surfrad ghi')
plt.legend()
plt.show()

surfrad_data4, _ = pvlib.iotools.read_surfrad(filename='https://gml.noaa.gov/aftp/data/radiation/surfrad/Goodwin_Creek_MS/2021/gwn21038.dat')
surfrad_data5, _ = pvlib.iotools.read_surfrad(filename='https://gml.noaa.gov/aftp/data/radiation/surfrad/Goodwin_Creek_MS/2021/gwn21039.dat')
surfrad_data6, _ = pvlib.iotools.read_surfrad(filename='https://gml.noaa.gov/aftp/data/radiation/surfrad/Goodwin_Creek_MS/2021/gwn21040.dat')
surfrad_data = pd.concat([surfrad_data4,surfrad_data5,surfrad_data6,])
surfrad_30m = surfrad_data.resample('30T').mean()
mask = (nsrdb_data.index > '2021-02-07 10:00') & (nsrdb_data.index < '2021-02-10 02:00')
plt.plot(nsrdb_data.GHI[mask], drawstyle='steps', label='nsrdb ghi')
plt.plot(surfrad_30m.ghi, drawstyle='steps', label='surfrad ghi')
plt.legend()
plt.show()
williamhobbs commented 11 months ago

In several NREL reports that include validation, it was only performed on time intervals with solar zenith angles less than 80 degrees:

https://www.nrel.gov/docs/fy17osti/67722.pdf https://www.nrel.gov/docs/fy20osti/74386.pdf https://www.nrel.gov/docs/fy22osti/82063.pdf

There's also https://doi.org/10.1063/5.0030992, which is referenced in the paper Kevin linked (emphasis mine):

However, under certain nonoptimal conditions (e.g., extreme solar zenith angles or viewing angles), the DCOMP model can fail to retrieve these cloud properties, which forces the PSM to perform a gap-filling procedure that is described in Section 2.2; this is a long-known issue and is well-described by Yang (2021, Section V.B.)

But I can't seem to find a free version of that paper.

AdamRJensen commented 11 months ago

There's also https://doi.org/10.1063/5.0030992, which is referenced in the paper Kevin linked (emphasis mine): But I can't seem to find a free version of that paper.

@williamhobbs It should be available on Research Gate.

williamhobbs commented 11 months ago

Thanks! Apparently I didn’t look very hard…