Christian A.

Gueymard

Part 2Overview
Interannual and long-term variability in DNI Spatial variability in DNI Daily frequency distributions Typical Meteorological Year (TMY), use and abuse Resource assessment for large projects: local measurements are important! Solar Resource Enhancement Factor (SREF) Circumsolar irradiance Spectral irradiance & SMARTS Conclusions

For more information: http://www.SolarConsultingServices.com

Interannual DNI Variability (1)

There are good years and bad years in everything, particularly in DNI, due to: Climate cycles (El Nio, La Nia), changes in release of natural aerosols, increase or decrease in pollution, volcanic eruptions, climate change For GHI, it might take only 23 years of measurement to be within 5% of the longterm mean. For DNI, it takes much longer, up to 515 years. Short measurement periods (e.g. 1 year) are not sufficient for serious DNI resource assessment! Special techniques must be used to correct long-term modeled data using short-term measured data.

Eugene data: http://solardat.uoregon.edu/

Interannual DNI Variability (2)

Interannual variability in DNI is much higher (at least double) than that in GHI. This variability is higher in cloudier climates (low Kn), but still significant in clearer regions (high Kn), which are targeted by CSP/CPV. Plots and maps provide this variability in terms of Coefficient of Variation (COV): COV = St. Dev. / Mean This is significant at only a 66% probability level. For a bankable 95% probability, double the COV results.

C.A. Gueymard, Fixed or tracking solar collectors? Helping the decision process with the Solar Resource Enhancement Factor. SPIE Conf. #7046, 2008. S. Wilcox and C.A. Gueymard, Spatial and temporal variability in the solar resource in the United States. ASES Conf., 2010.

http://rredc.nrel.gov/solar/new_data/variability

Long-term DNI Variability (1)

Only the past DNI resource can be known with some (relative) degree of certainty. But the goal of CSP/CPV resource assessment is to obtain projections of 2030 years into the future. Q: How can this be done if there are unknown forcings that result in long-term trends? Only a handful of stations in the world have measured radiation for more than 50 years. Long-term trends in GHI and DNI have been detected. Periods of Brightening and Dimming are now documented.
GHI, 19372006 Potsdam, Germany

Early brightening

Dimming

Brightening

Long-term DNI Variability (2)

Long-term trends do not affect the world equally. Current results indicate a brightening in most of the NH, and a dimming in the tropical regions of the NH and SH. India and China are directly affected, most probably because of the current increase in coal burning and pollution (Asian Brown Cloud).

Trends in GHI (% per decade)

Good news in some areas, bad news in others!

M. Wild et al., J. Geophys. Res. 114D, doi:10.1029/2008JD011382, 2009 M. Wild, J. Geophys. Res. 114D, doi:10.1029/2008JD011470, 2009

How Long-term DNI Variability (3)

Most long-term variability results are for GHI. One difficulty is to transform these results into DNI variability. There are regions where DNI varies more than GHI, others where the reverse occurs.
How DNI will vary during the next 2030 years depends on many unknowns: Air quality regulations and Kyoto-type accords Climate change evolution Possible geoengineering (forced dimming) Volcanic eruptions, etc. So nobody has a definite answer!

L.D. Riihimaki et al., J. Geophys. Res. 114D, doi:10.1029/2008JD010970, 2009

Long-term DNI Variability (4) Main Causes

Cloud climatology Emissions of black carbon (BC) and other aerosols Humidity patterns

Consequences

Spatial DNI Variability

Spatial variability is important for two reasons: In regions of low spatial variability, use of low-res resource maps (e.g., 100x100 km) might be OK, at least for preliminary design. Conversely, in regions of high spatial variability, only hi-res maps (10x10 km or better) should be used. If variability is high, measured data from only nearby weather stations should be trusted.
5x5 matrix 10x10 km grid cells

S. Wilcox and C.A. Gueymard, Spatial and temporal variability in the solar resource in the United States. ASES Conf., 2010.

Daily Frequency Distributions

Most generally, daily frequency distributions are highly skewed. This suggests a log-normal probability distribution, for instance. At high-DNI sites, the most typical days of the year (modal value) provide much more direct energy than the average (mean value) days of the year. This is reversed at cloudy sites. Hence, the mean daily DNI should not be the only indicator to use when evaluating the potential of the solar resource.
14 12 10
Frequency %
Alice Springs, avg 7.36 kWh/m2 Bermuda, avg. 3.76 kWh/m2

Daily frequencies

8 6 4 2 0

9
2

10

11

12

13

14

15

Daily DNI (kWh/m )

Typical Meteorological YearTMY (1)

For decades, TMYs have been used by engineers to simulate solar systems or building energy performance. TMYs conveniently replace 30 years of data with a single typical year. Models of solar system power output prediction (e.g., PVWatts, http://www.nrel.gov/rredc/pvwatts/) or of performance and economic estimates to help decision making (e.g., Solar Advisor Model, https://www.nrel.gov/analysis/sam/) still rely heavily on TMY-type data. To define each of the 12 months of a synthetic year, TMYs use weighting factors to select the most typical year among a long series of available data (including modeled irradiance). In the U.S., three different series of TMY files have been produced. The weight they all used for DNI is relatively small. It should not be construed that TMY3 is more advanced or better than TMY2!
TMY Period GHI weight DNI weight # Stations 19521975 12/24 0 222 TMY2 1961-1990 5/20 5/20 239 TMY3 (i) 19762005 (ii) 19912005 5/20 5/20 (i) 239 (ii) 950

Typical Meteorological YearTMY (2)

Q: Are TMY data appropriate for CSP/CPV applications?
TMYs have some potential drawbacks: DNI in TMY data is 99% modeled. At clear sites, the TMY hourly distributions usually show discrepancies above 500 W/m2, compared to measured data. This is due to the use of climatological monthly values (rather than discrete daily values) for the aerosol data. Hourly values are used. This may not be ideal for non-linear systems with thresholds above 150 W/m2 (see why in Pt. 1 of this webinar). Non-typical low-DNI years are excluded from the data pool. Using TMYs for risk assessment is risky.
20 Golden, CO Sunup hourly frequencies Measured NSRDB TMY3

Hourly frequencies of 19912005 NSRDB data used to obtain TMY3 for Golden, CO. Compared to measurements, note the NSRDB and TMY3 overestimations below 900 W/m2, and underestimations above.

16 Frequency %

12 8

100

200

300

400

500

600
2

700

800

900

1000 1100

DNI bins (W/m )

Typical Meteorological YearTMY (3)

To obtain bankable data, the use of TMYs is inappropriate. The risk of bad years cannot be assessed correctly. TMY may seriously overestimate the P90 exceedance probability. Example: For Boulder, the total annual DNI from TMY2 happens to correspond to P50, but this is far from being a general rule.

Local Measurements
An essential part of CSP/CPV resource assessment! Two types of weather stations, depending on radiometer technology Minimum measurement period recommended: 1 year Performance and prices vary [Ask us for more details and custom solutions!] These short-term measurements should then be used to correct long-term satellite-based modeled data using appropriate statistical methods.

Solar Resource Enhancement Factor (1)

Q: What is the average annual resource of CSP/CPV compared to that for other solar technologies? For each type of concentrator, the Solar Resource Enhancement Factor can help decide

C.A. Gueymard, Fixed or tracking solar collectors? Helping the decision process with the Solar Resource Enhancement Factor. SPIE Conf. #7046, 2008.

Solar Resource Enhancement Factor (2)

Latitude is not a good predictor for the solar resource. Based on the 19611990 NSRDB (excluding Alaska), the minimum U.S. resource (measured by KT or Kn) is found at Quillayute (northern Washington state), whereas the maximum is found at Daggett, California. KT = GHI/ETHI Kn = DNI/ETNI

Solar Resource Enhancement Factor (3)

Know your competition! Flat-plate PV collectors on 2-axis trackers have a sizeable resource advantage over CSP/CPV. With recent smart 2-axis trackers, the annual resource for planar collectors may increase another 515% (depending on cloudiness). This is severe competition

Plots based on the SREF method

Circumsolar Irradiance (1)

Definition Scattering is typically very strong around the sun, so the sky looks bright. This is diffuse radiation that behaves like direct radiation, and can thus be concentrated. Measurement Circumsolar irradiance (CSI) is difficult to measure, but is possible with a specially modified NIP. T.H. Jeys and L.L. Vant-Hull, Solar Energy 18, 343-348, 1976. Routine measurements of DNI actually include CSI within 2.52.9 of the sun center. Such data slightly overestimate the true DNI that can be used by CSP/CPV since their concentration ratio is high and the subtended cone is smaller (usually <1). The CS radiance (intensity) can be measured only with specialized equipment. The only known current instrument to be designed for this is SAM, which scans from the sun center to 8 from it.

Circumsolar Irradiance (2)

Modeling The clear-sky CSI (up to 10) can be modeled with SMARTS, if the atmospheric input data is available. Below 3, the CS effect is found negligible under very clear conditions, but can represent up to 5% of DNI under very hazy conditions.
C.A. Gueymard, Spectral circumsolar radiation contribution to CPV. Proc. CPV-6 Conf., Freiburg, 2010. C.A. Gueymard, Solar Energy 71, 325-346, 2001.

http://www.solarconsultingservices.com/smarts.php

Under thin cirrus clouds, the CS effect becomes important, but its modeling is then difficult. A large collection of SAM measurements would be needed to develop simple empirical models. We are now trying to make such a research project possible, in collaboration with SAMs manufacturer, as well as U.S. and European institutions.

Circumsolar Irradiance (3)

Sun and Sky Radiance The radiance of the suns disc is not constant (limb darkening effect). The circumsolar sky radiance decreases exponentially with radial distance The slope of this decrease increases with optical depth (clear hazy thin cloud).
Monument Valley analogy

Linear scale

Logarithmic scale

Logarithmic scale

Circumsolar Irradiance (4)

Characteristics of CS irradiance The CS effect is more pronounced at shorter wavelengths, since it is caused by scattering The CS/DNI fraction increases linearly with the opening angle It is also a function of air mass and optical depth (aerosol or cloud) More results to be presented at the CPV-7 conference (2011).

Spectral Irradiance (1)

The direct spectrum red shifts when air mass (AM) increases or when aerosol turbidity (AOD) increases Below 700 nm, atmospheric extinction is dominated by scattering Above 700 nm, it is dominated by absorption (water vapor, CO2) Reference AM1.5 spectra have been standardized by ASTM: E891 (1987) and G173 (2003). The latter was specially designed for CPV.
C.A. Gueymard et al., Solar Energy 73, 443-467, 2002.

Spectral Irradiance (2)

Routine spectral measurements are difficult and costly Spectral modeling is possible with various existing codes, e.g., SMARTS SMARTS was used to develop ASTM G173 and other standards (IEC) SMARTS is commonly used tool to evaluate spectral effects in PV and CPV, and offers compatibility with current standards All PV cells have a strong spectral selectivity. SMARTS can be used to evaluate spectral mismatch correction factors, or the output of multijunction (MJ) cells under variable spectral conditions.
MJ: 41% eff., for HCPV c-Si: 22% eff., for LCPV 4 kW, 3 suns

JX Crystals

Spectral Irradiance (3)

t2 t2 t1

Daily-average direct spectrum: Daily Spectral Enhancement Factor:

Edn = En ( t )En ( t ) / En ( t )
t1

DSEF = [ Edn

4000 280

dn

S d ]/[ Esn

4000 280

sn

S d ]

C.A. Gueymard, Daily spectral effects on concentrating PV solar cells as affected by realistic aerosol optical depth and other atmospheric conditions. SPIE Conf. #7410, 2009. A.L. Dobbin et al., How important is the resolution of atmospheric data in calculations of spectral irradiance Conf., 2010. and energy yield for (III-V) triple-junction cells? CPV-6

Spectral Irradiance (4)

It is found that, for any type of solar cell, the spectral effect is a strong function of AOD. One goal of the current R&D is to fine tune MJ cells by optimizing their bandgap combinations as a function of the regional average spectrum. This might result in significant increases in the annual energy output. Cirrus clouds appear to affect the performance of CPV modules, but it is unclear if its because of spectral or circumsolar effect (or both).
G. Peharz et al., Evaluation of satellite cirrus data for performance models of CPV modules. CPV-6 Conf., 2010.

Conclusions (2)
The DNI solar resource is highly variable and difficult to model using past data. Projecting it 2030 years into the future is even more difficult. Local radiation measurements are still the best source of data, and are necessary to derive the bankable data needed for big projects. However, the type of radiometer should be selected properly, its limitations known, and appropriate maintenance provided. If local DNI measurements are available for only a short period (less than 5 years), they should be used in conjunction with long-term modeled data to obtain locally adjusted time series spanning at least 10 years. The use of TMY data is not recommended, particularly for a non-linear operation (startup threshold). In that case, sub-hourly time series are ideal. Circumsolar and spectral effects have second-order importance, but should still be studied for better simulation, and possible fine tuning of CPV cells. The benefit of a larger circumsolar contribution to LCPV systems cannot be evaluated yet. Because of the lack of high-quality measured DNI data in the public domain, the science of resource assessment progresses only slowly.

Thank you!