solar resource assessment and forecasting: recent ... · solar resource assessment and forecasting:...
TRANSCRIPT
Solar resource assessment and forecasting: Recent achievements, bankability pressures,
and current challenges Christian A. Gueymard, Ph.D. President, Solar Consulting Services
1. What are the most suitable areas? 2. Can we trust old/distant measured data? Can we trust modeled 8me series
based on satellite data? Can we trust TMYs? 3. What technology to select? What is the resource for each technology?
๏ PV (crystalline silicon or thin films) Fixed flat-‐plate? 1-‐axis or 2-‐axis tracking?
๏ Concentra8ng solar power (CSP): thermal process 1-‐axis or 2-‐axis tracking?
๏ Concentra8ng PV (CPV): direct conversion to electricity 4. What are the uncertainty and the interannual variability in the resource? 5. How much energy can be produced the first year? During the worst years? 6. How reliable is the resource assessment prepared by Company X? Is a second
opinion needed? Who are the real experts? 7. Where to install a weather sta8on? How long are such measurements needed
for bankability? What to do if the modeled and measured data disagree?
Key issues in solar resource assessment
1. Obvious increase in number of journal papers and conference presenta8ons (ASES, ISES, SolarPACES, CPV…)
2. Complexity of radia8ve transfer processes in the atmosphere: more science needed, more scien8sts a]racted.
3. Major solar developers realize the solar resource cannot be modeled with perfect accuracy; some really bad experiences occurred (e.g., Abu Dhabi).
4. Uncertain8es in solar resource data have been shown to be the largest source of error in CSP/CPV produc8on es8mates.
5. Banks become more suspicious, some8mes require second opinions. 6. More resource data providers because of new markets. 7. Increasing interest for nowcas8ng and forecas8ng from operators and
u8li8es (electricity grid stability, reserve dispatching, spot market…) 8. Accelerated convergence between different scien8fic fields:
atmospheric sciences, meteorology, climate, GIS, metrology, radiometry, solar engineering…
Renewed interest in solar resource fundamentals
1930–1980s: • Ångström-type empirical correlations: daily GHI vs. daily sunshine. • First world monthly distribution maps of GHI in 1965 (Löf et al.). • Hot topics: Local vs. “universal” correlations; modeled GHI accuracy;
interpolation/extrapolation; sub-daily data; value of sunshine data.
Progress in Solar Resource Assessment (1)
1980–1990s: • Broadband radiation models used to predict hourly irradiances from
meteorological information (cloud cover…) at specific sites (airports). Example: NREL’s NSRDB 1961–1990.
• Development of TMYs for energy simulations (buildings & solar). • Hot topics: How to obtain aerosol data?; spatial interpolation methods;
combination of measured and modeled data; subjectivity of human cloud observations.
Progress in Solar Resource Assessment (2)
1990s–2000s: • Hourly irradiances predicted from spaceborne cloud observations and
other data. • Huge improvement in spatial resolution, time resolution, making
continuous maps and GIS applications possible. • Thermal imbalance issue discovered in many pyranometers;
development of rotating shadowbands, spectral correction methods. • Hot topics: Clouds vs. snow; sources of aerosol data; validation issues;
derivation of DNI; value of stochastic models.
Progress in Solar Resource Assessment (3)
0
2
4
6
8
10
12
1 2 3 4 5 6 7 8 9 10 11 12 Year
2-Axis Tracking Concentrator1961–1990
San Diego, CA (Lat. 32.73°)Daggett, CA (Lat. 34.87°)M
onth
ly-a
vera
ge ir
radi
atio
n (k
Wh/
m2 p
er d
ay)
Month
Ann
ual A
vera
ge
• •
2000s–now: • Many data providers now offering long-term 15-min to
60-min irradiance time series and global datasets. • Nominal spatial resolution typically ≈3 km, can be disaggregated down
to ≈90 m, with shading analysis. • Nowcasting/Forecasting now possible, but still in their infancy. • Hot topics: Surface reflectance issues; daily vs. monthly aerosol data;
DNI modeling; validation of modeled DNI (lack of HQ measurements); variability; bankability; value of TMYs; modeled/ measured data combination; forecasting; which data should be used where or for what purpose?
Progress in Solar Resource Assessment (4)
๏ Incident irradiance: CSP/CPV plants use DNI: Direct Normal Irradiance
Fixed flat-‐plate collectors use GTI: Global Tilted Irradiance 2-‐axis tracking flat-‐plate collectors use GNI: Global Normal Irradiance.
๏ GHI (Global Horizontal Irradiance) is mostly used in intermediate calcula8ons and “first look”, rough solar resource visualiza8on.
๏ DNI is what 2-‐axis tracking CSP/CPV concentrators can u8lize fully; 1-‐axis trackers (e.g., parabolic troughs) get somewhat less.
What is the fuel of solar systems?
• For each type of concentrator, SREF compares the annual or seasonal resource in rela8ve terms.
• 1-‐axis tracking can approach the resource of 2-‐axis tracking only if collectors are oriented N-‐S and 8lted at la8tude. 1-‐axis troughs oriented E-‐W have a much reduced resource.
• 2-‐axis tracking concentrators have a be]er resource than la8tude-‐8lt flat-‐plates only in very sunny areas.
Solar Resource Enhancement Factor
or ???
FAQ: What type of collector should be used in any given area?
or
0.7
0.8
0.9
1
10 15 20 25 30 35 40 45 50
1-Axis E-W (0°) 1-Axis N-S (0°) 1-Axis N-S (L)
Sola
r R
esourc
e E
nhancem
ent F
acto
r
North Latitude (°)
Annual SREF222 U.S. Sites
Concentrating Collectors
0.6
0.7
0.8
0.9
1
1.1
1.2
0.1 0.2 0.3 0.4 0.5 0.6 0.7
Sola
r Res
ourc
e En
hanc
emen
t Fac
tor
Annual K
Annual Ratio222 U.S. Sites
2-Axis Concentrating vs Latitude-Tilt Flat-Plate Collectors
1. Improve DNI modeling: from empirical to physical deriva8on 2. Improve radia8ve transfer modeling through clouds: 3D, shading effects… 3. Improve aerosol data 4. Improve refresh frequency of surface reflectance data: Snow, ice… 5. Understand why solar resource datasets differ so much! 6. Evaluate future trends in solar resource, in rela8on with climate change
(dimming vs. brightening) 7. Increase availability, spa8al distribu8on, and quality of public-‐domain
ground-‐truth measurements of aerosols and solar irradiance 8. Harmonize valida8on methods; standardize uncertainty repor8ng, TMY
deriva8on 9. Validate, validate, validate! Aerosol data, modeled irradiance 8me series,
TMYs, forecasts, … 10. Standardize bankability requirements and methods 11. Improve awareness about desirable exper8se and creden8als
Developers and banks tend to trust “big names” (engineering firms or consultants in solar technology), who may not know much about solar resource issues.
12. Accuracy is not always welcome: Beware of special interests!
Challenges
• Inaccurate empirical deriva8on of DNI from GHI
• Explains large random errors in hourly or sub-‐hourly DNI/GTI from current satellite-‐derived datasets
• Future solu8on: Use more sophis8cated physical modeling
DNI Modeling
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
NREL, Golden, COHourly data 2006–2008
PSP pyranometers
ErbsMeasured (PSPs)
Diff
use
ratio
, K
Clearness ratio, KT
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
NREL, Golden, COHourly data 2006–2008
CM22 pyranometers
ErbsMeasured (CM22s)
Diff
use
ratio
, K
Clearness ratio, KT
Physical Radiative Modeling
MODIS view over Kamchatka GHI
Cloud
GOES 11
• Requires cloud retrieval information (layer cloud cover, ice/water cloud optical depths…) and surface optical properties (vegetation, temperature…) rather than visible radiance data: more complexity.
• Example: GSIP, based on GOES real-time data. Encouraging initial validation…
• Reanalysis data can provide historical time series of various atmospheric constituents.
• NWP and mesoscale models now used for cloud and solar radiation forecasts
Convergence with meteorology/ climate/atmospheric sciences
Large differences between solar resource maps (par8cularly DNI), and between modeled and measured data, over many areas.
Kenya NREL vs. DLR:
up to 50% differences
NASA-SSE NREL
-4
-3
-2
-1
0
1
2
3
4
5
6
7
8
9
-30
-20
-10
0
10
20
30
40
50
60
70
80
90
100
0 1 2 3 4 5 6 7 8 9 10 11 12 13 14
Long-term mean DNITamanrasset
DNI-Meas3TierISISGeoModelMeteonormSSESWERA
DN
I (k
Wh
/m2)
% D
iffere
nce
Month
Yr
Sahara Large monthly and annual errors in all datasets: Inaccurate aerosol data, etc.
Differences in Resource Maps (1)
0
1
2
3
4
5
6
7
Lodwar Kitale Maralal Wamba Baringo Eldoret Meru Nakuru
Predicted Annual DNI Resource in Kenya
NRELDLR
Dai
ly-m
ean
Annu
al D
NI (
kWh/
m2 )
City (North to South gradient)
Equa
tor
• Large differences in DNI resource maps in Asia • Obvious effect of spa8al resolu8on; interpola8on methods too risky.
Differences in Resource Maps (2)
Eastern India Low- vs. high-
resolution data
• No public-‐domain HQ radia8on measurements there: Serious valida8on nearly impossible. • Radia8on climate changes rapidly: urbaniza8on, pollu8on, dust, etc. • Resource for CSP may not be as high as most developers expect (monsoon, dust, etc.). • DNI’s nega8ve trend: 1–10% per decade (dimming).
• Resource oren evaluated as simple average or interpola8on of low-‐ resolu8on (free) data.
• Gross performance mis-‐ predic8ons can be expected in many cases!
• Typical bias achievable globally with the best satellite-‐based data sources: < ±3% for GHI (±5% or more over difficult areas) < ±7% for DNI (±15% or more over difficult areas)
• Typical hourly RMSD achievable in semi-‐arid regions with the best data sources: < 14% for GHI, < 25% for DNI
• Only experts can tell which would be the best data source over a given area! • Sources of uncertainty: 1. Systema8c features in simplified/empirical parts of the radia8ve model (lack of physics) 2. Oversimplifica8on of the cloud index method for some types of clouds, 3D effects, … 3. Local issues: complex terrain, costal zones, mountains, urban/industrial polluted areas 4. High-‐albedo surfaces misinterpreted as clouds 5. Satellite imagery: Spa8al resolu8on, misloca8on 6. Ground (pinpoint) observa8ons vs. area-‐averaged values from satellite cloud data 7. Poor 8me and/or spa8al descrip8on of local variability in atmospheric or cloud data 8. Errors in measured data used for valida8on: miscalibra8on, no regular cleaning, inadvertent shading, instrument malfunc8on, data gaps, lack of QC…
• Highest uncertainty zones: temperate/cloudy climates, complex terrain, high-‐AOD (hazy/turbid) areas (par8cularly for DNI).
Uncertainty in Modeled Irradiance
• Hourly DNI
• Annual DNI
Uncertainty in Modeled DNI
Variable Humid tropics
Arid & semi-‐arid
Temperate climate
Steep terrain
Snow & ice
Costal zones
Polluted areas
Eleva8on, shading
very low very low very low low very low very low
Clear-‐sky Model
very low very low very low very low low very low low
Aerosols low high low medium low high
Water vapor
very low very low very low low very low very low
Cloud index
medium/high
low medium low medium low low
Variable Clear sky Sca@ered clouds Cloudy/overcast
Eleva8on, shading very low very low very low
Clear-‐sky model low very low very low
Aerosols high low very low
Water vapor low very low very low
Cloud index low moderate very low
• Over arid areas, AOD largest source of uncertainty in DNI and GTI. • Not enough HQ ground observa8ons • Satellite observa8ons and chemical transport models s8ll have significant biases • Difficult quan8fica8on of AOD-‐to-‐DNI error propaga8on • Need for improved data quality over many regions (sun belt) • Need for higher spa8al resolu8on • Need for HQ daily AOD data, 1980–now • Need for good AOD forecasts.
Aerosol Issues
SOLSUN AOD Database
AOD550, North America, 2000–2011 Avg.
• Calibrated aerosol data (monthly, 2000–2011) for the world, 0.5x0.5° resolution • Huge improvement over existing
satellite data in many areas (SW USA…)
• Commercial product for solar resource data providers, climate research, etc.
• Available soon (contact us)
0.00
0.10
0.20
0.30
0.40
0.50
0.60
0.70
2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012
Desert Rock, Nevada2001–2011
Gridded Data vs. Ground Truth
ObservationsSOLSUNMODIS-TerraMISR
Mon
thly
AO
D @
550
nm
MonthAOD550, Europe, Aug. 2003 (heat wave, fires)
• Monthly variability can change from low (summer) to high (winter) • Seasonal compensa8ons over the year • Interannual variability: DNI >> GTI >> GHI.
Interannual Variability
TMYs are not a panacea! TMYs are usually based on ≈100% modeled data. At clear sites, TMY2/TMY3 hourly distribu8ons show significant discrepancies above 500 W/m2.
Hourly values are used: Not ideal for non-‐linear systems with thresholds above 200 W/m2 (CSP). 1-‐min to 15-‐min data desirable, but not accurately modeled.
TMYs are made to represent average condi8ons (≈P50); extreme months are excluded by design.
Financial ins8tu8ons are most interested in worst-‐case scenarios (P90, P95, P99), not in P50; bankable reports cannot be obtained from TMY data!
Lack of standard TMY methodology; specialized TMYs may be needed (buildings, solar/PV, CSP/CPV).
TMY Issues
50
60
70
80
90
100
0 200 400 600 800 1000
Denver/Golden, CONSRDB, 1991–2005
Hourly DNI
TMY3
Modeled, overall
Measured
Modeled, year 1–15
Cu
mu
lative
Fre
qu
en
cy (
%)
Irradiance (W/m2)
0
10
20
30
40
50
60
70
80
90
100
4.2 4.4 4.6 4.8 5 5.2 5.4 5.6 5.8 6
Boulder, CONSRDB, 1961–1990
Annual DNI
Pro
babili
ty o
f E
xceedance (
%)
Annual DNI (kWh/m2)
TMY2
Median
0
4
8
12
16
20
0 100 200 300 400 500 600 700 800 900 1000 1100
Golden, COSunup hourly frequencies
MeasuredNSRDBTMY3
Freq
uenc
y %
DNI bins (W/m2)
• Cri8cal part of solar resource assessment, necessary to sort out local variability effects at different 8me scales.
• Performance and prices vary…
• Adverse condi8ons frequently exist (dust, snow, frost, birds…): regular maintenance is essen8al!
• Minimum measurement period for bankability: 9–12 months.
Short-‐term observa8ons should be used to correct long-‐term satellite-‐ based modeled data with appropriate NWP and sta8s8cal methods.
Only way to guarantee ±5% accuracy and bankability for CSP/CPV or in “difficult” areas!
For expert advice: h]p://solarconsul8ngservices.com
Local Measurements