lable at ScienceDirect

Renewable Energy 74 (2015) 633e639

Contents lists avai

Renewable Energy

journal homepage: www.elsevier .com/locate/renene

Spatio-temporal characterisation of extended low direct normalirradiance events over Australia using satellite derived solar radiationdata

Ben Elliston a, *, Iain MacGill a, b, Abhnil Prasad c, Merlinde Kay c

a School of Electrical Engineering and Telecommunications, University of New South Wales, Sydney, NSW 2052, Australiab Centre for Energy and Environmental Markets, University of New South Wales, Australiac School of Photovoltaic and Renewable Energy Engineering, University of New South Wales, Australia

a r t i c l e i n f o

Article history:Received 24 December 2013Accepted 21 August 2014Available online

Keywords:Concentrating solar thermalDirect normal irradianceExtreme eventsPower system

* Corresponding author. Tel.: þ61 2 6268 8355; faxE-mail addresses: b.elliston@unsw.edu.au (B. E

(M. Kay).

http://dx.doi.org/10.1016/j.renene.2014.08.0670960-1481/© 2014 Elsevier Ltd. All rights reserved.

a b s t r a c t

As part of an on-going program to develop technological scenarios for 100 per cent renewable generationwithin the Australian National Electricity Market (NEM), we explore the degree to which concentratingsolar thermal (CST) power might reliably contribute to the generation mix. We analyse satellite-derivedhourly direct normal irradiance data provided by the Australian Bureau of Meteorology for Australia overa 13-year period. This large data set covers sufficient time to enable us to characterise the frequency andduration of rare events such as extended periods of heavy cloud cover and hence low solar insolationover regions of Australia. The results highlight those regions with both the highest and lowest occurrenceof extended periods of low DNI. They also identify regions whose correlated climatic characteristicswould reduce overall CST generation variability if the plants were distributed across them. As such, thefindings may assist both project developers, and long-term system planning for reserve generation ca-pacity in future high renewable generation mixes.

© 2014 Elsevier Ltd. All rights reserved.

1. Introduction

The question of what role solar generationmight usefully play inlow carbon future electricity industries is receiving growingattention given recent technological progress and cost reductions.At present, the Australian National Electricity Market (NEM) pro-duces around one third of total Australian greenhouse gas emis-sions, and its predominantly coal-based fuel mix has the highestemissions intensity of any electricity industry in the OECD, and oneof the highest in the world [9]. The current climate science suggeststhat developed countries must aggressively reduce greenhouse gasemissions over the next several decades to a point of near-zeroemissions by 2050 in order to avoid global warming of more than2 �C [12]. If aggressive emissions reduction targets are to be ach-ieved, electricity industries worldwide will need to be transformedover the coming decades to zero carbon electricity sources.

The primary concern for the integration of solar electricity intopower systems at high penetration is the variability of solar

: þ61 2 6268 8443.lliston), m.kay@unsw.edu.au

radiation over various time scales [4]. At short time scales, secondsto minutes, variability creates difficulties for balancing, frequencycontrol, and voltage control. Variability on time scales of hours to aday ahead influences the commitment of other generating units tomaintain adequate supply reserves. On even longer time scales ofdays to weeks and beyond, variability influences decision makingfor power system dispatch and security issues including productionscheduling of hydro, fuel contracting and scheduled generatormaintenance [6].

Solar generators have some particular advantages over othervariable renewable sources such as wind and wave power. Inparticular, the solar resource is more evenly and widely distributedthan these options. Wind and wave power machines must be builtto withstand forces far in excess of what is typically encountered inthe field. However, an unexpected lack of solar energy availabilityoutside normal statistical parameters is of concern for power sys-tem planning and operation. Photovoltaic (PV) and CST plants haverather different characteristics in this regard. In particular, CST re-quires (DNI) to achieve effective solar concentration while PV willprovide some power as long as there is some sunlight. However,CST also has some inherent energy storage through its heatcollection and transfer system, and there are straightforward op-tions to add additional heat storage. Still, if an extended period of

B. Elliston et al. / Renewable Energy 74 (2015) 633e639634

low irradiance occurs, there must be sufficient generating capacity,energy storage or demand-side response available in the powersystem to cover demand without significant solar contribution.

As part of previous efforts to characterise the solar radiationresource, it has been common to quantify the annual frequency ofruns where solar radiation does not exceed a threshold value. Thishas typically been done using daily insolation measurements andusually only from a single location [7,2]. In this paper, we usesatellite-derived hourly solar irradiance data for Australia over theperiod 1998e2010 inclusive, to characterise the frequency andduration of runs of extended low DNI at a spatial resolution of5 km � 5 km across the entire Australian continent. Although theDNI threshold that is required to begin producing electrical powerfrom a CST plant varies with the plant design, a value of 400 W/m2

has been chosen for this work as broadly representative across arange of CST power plants in existence today [23,8]. The resultspresented will answer the following questions:

� How long are periods of DNI below 400 W/m2 across differentlocations within Australia?

� How frequently do these low DNI events occur, and what are thesynoptic situations causing these events?

� The location and time of year we see these events occurring.

The rest of the paper first provides some context for assessinglikely CST performance. Themethod used in this study to assess CSTperformance is described in Section 3. Section 4 presents thefindings and analysis by examining the spatial relationship of thesehistorical periods of extended low DNI, while Section 5 concludeswith showing how geographic diversity of concentrating solar po-wer stations can help minimise the impact of these periods.

2. Context for concentrating solar power

The frequency and duration of extended periods of low DNI is ofparticular interest for modelling the operating performance of CSTpower plants because these periods will characterise the durationof what is essentially a forced outage. Like the forced outage of aconventional thermal power plant, it is necessary to plan sufficientreserve generating capacity to meet this contingency.

For CST power plants, steam production declines rapidly oncedirect normal irradiance falls below a critical threshold. Dependingon the CST design in use, this threshold can be the level of irradi-ance required to first overcome power losses in the receiver andthen to achieve the minimum turbine load. Fig. 1 shows the non-existent electrical power being produced by a 100 MW parabolictrough plant without storage, as simulated by the System AdvisorModel [14], during a three day period of extended low DNI. Thereason for this extended low irradiance period was a severe dust

Fig. 1. Simulated 100 MW trough plant in Cobar, New South Wales, September 2009.

storm that developed in central South Australia and travelled acrossto Eastern Australia in the following three days. In a CST systemwith thermal storage, a protracted period that is much longer thanthe number of full-load hours covered by the thermal storage willexhaust all stored energy.

A range of CST plant designs are in operation today. Some, suchas the SEGS plants in California, have no added thermal energystorage. Others, such as Andasol-I and Gemasolar in Spain areequipped with between 7.5 and 15 full-load hours of thermal en-ergy storage respectively. This storage enables the plants tocontinue operating through short periods of low irradiance. How-ever, less common, extreme extended low irradiance events lastingseveral days cannot be overcome by the thermal energy storage. Itis clearly uneconomic to incorporate enough thermal storage towithstand such events.

3. Methodology

We have completed an exhaustive analysis of satellite-derivedsurface solar irradiance data provided by the Australian Bureau ofMeteorology. The data span a 13 year period from 1998 to 2010inclusive and represent the most comprehensive historical recordof solar irradiance for the Australian continent. First, the data setwill be described. We then describe how extended periods of lowDNI were located in a computationally efficient way.

3.1. Surface solar irradiance data

The National Climate Centre at the Bureau of Meteorologyproduces two gridded irradiance products: one for global hori-zontal irradiance and one for DNI. The estimates of global hori-zontal irradiance are made using satellite images collected fromthree satellites over the past 13 years: GMS-5 (JapanMeteorologicalAgency), GOES-9 (U.S. National Oceanic and Atmospheric Admin-istration), andMTSAT-1R (JapanMeteorological Agency). A physicalmodel progressively developed by Refs. [21,22] is in operational useat the Bureau of Meteorology based on earlier work by Refs. [5,10].Any anomalous grids within the satellite images are discarded.

Feedback adjustments to the model from a limited number ofground station pyranometer readings are used to eliminate bias dueto sensor calibration errors, biased estimates of water vapour fromthe numerical weather prediction model and aerosol effects [22]. Itshould be noted that positional accuracy can also be a source oferror with some images having errors of several kilometres. How-ever, the corrections made, are on average, less than 1%. With thesecorrections applied, the satellite-derived data is comparable inuncertainty to good quality pyranometers and can be consideredsufficiently accurate to use for finding periods of irradiance below acertain threshold.

The DNI data provided by the Bureau of Meteorology is derivedfrom global horizontal irradiance using an adapted version of adiffuse fraction model by Ref. [20]. The data set will be brieflydescribed here, but full details are available in a metadata docu-ment that accompanies the data sets [16].

The spatial extent of both data sets includes the entire Austra-lian continent (10.05 �S to 43.95 �S, 112.05 �E to 153.95 �E). Thespatial resolution is 0.05� � 0.05�, or approximately 5 km � 5 km.

The temporal resolution of the data is 1 h and the range is shownin Table 1. No data are available from July 1, 2001 to June 30, 2003due to a long transition from the GMS-5 satellite imagery to GOES-9imagery. Up to date grids are being produced as images becomeavailable. The data includes up to 18 hourly grids per day. Hours 12to 17 (UTC) are excluded for brevity, as these grids are dark on everyday of the year. Additional grids may be missing due to absent orpoor quality images.

Fig. 2. Length of longest run of DNI below 400 W/m2.

Table 1Satellite sources used (Source: Bureau of Meteorology).

Start date End date Satellite Instrument

1998-01-01 2001-06-30 GMS-5 VISSRa

2003-07-01 2005-10-31 GOES-9 GOES I-M2005-11-01 2010-12-31 MTSAT-1R JAMIb

a Visible and infrared spin scan radiometer.b Japanese advanced meteorological imager.

B. Elliston et al. / Renewable Energy 74 (2015) 633e639 635

This data set represents the most comprehensive estimate DNIavailable for the Australian continent. Compared with a network ofsurface observation stations, satellite-derived data has exceptionalspatial coverage. Although data derived from high availabilityweather satellites should be much more reliable than a network ofground stations that require regular cleaning and calibration [21],the data set is missing numerous hourly grids for a range of reasons.In some cases, basic interpolation is required to handle these.

3.2. Computing environment

This work was carried out using access to a computer systemwith eight CPUs and 48 GB of memory. The machine runs the GNU/Linux operating system. The software for the present research waswritten in the Python programming language and makes use of anumber of Python extension packages:

1. Numpy, providing high-performance N-dimensional arrays [13];2. PyTables, for storing very large tables [1]; and3. Matplotlib, for visualising the data [11].

The PyTables package is important for work involving very largedata sets. In many environments, arrays are limited in size by theamount of virtual memory in themachine. PyTables overcomes thislimitation by allowing very large tables to be stored on disk andaccessed in segments as the array elements are referred to.

The Bureau of Meteorology gridded irradiance data are suppliedas a series of human-readable ASCII text files. Each hourly grid ofthe 13 year period is supplied in a single file. Within each file, theindividual hourly observed irradiance values (W/m2) for theAustralian continent (679� 839 pixels) are represented as 679 linesof 839 columns of plain text. The data were loaded into aPyTables compressed 3-dimensional array of dimensions113,952 � 679 � 839. PyTables efficiently manages decompressionof the array as it is accessed.

Unlike the original set of files, the table does not have anymissing hours. If an hourly grid was missing from the original data,the table is populated with a grid of “no data” values. This makes itpossible to traverse the entire period, hour by hour, without specialhandling of missing hourly grids.

3.3. Data processing

To find the longest period of insolation below a given threshold,the program iterates through the array, hour by hour, and recordsthe longest sequence of hours where the irradiance is below athreshold in each grid location. To speed up the search, the inde-pendence of the gridded data is exploited. The DNI data set is storedas 113,952 hourly grids of dimensions 679� 839. Searching for runsof low irradiance necessarily introduces a dependency betweengrid n and grid n þ 1, but within an hourly grid, each cell can beexamined in parallel.

At each time step, three 679-by-839 matrices are updated: Bbeing amatrix of binary values indicating whether irradiance at thecurrent time step is below the threshold (1 ¼ below the threshold),

C being the current count of consecutive hours below the threshold,and M being a matrix containing the maximum number ofconsecutive hours encountered thus far. As the program iteratesthrough the hourly grids, the current count of hours below thethreshold is computed using an entrywise product (Eq. (1)). As Bcontains only binary values, this has the effect of either incre-menting the count at a grid location, or resetting it to zero (Eq. (2)).At the end of each time step, M is updated (Eq. (3)). This techniquepermits the use of fast matrix operations and avoids the need toloop over each element of the matrices, giving high performanceand acceptable running time.

C ¼ ðCþ BÞ+B (1)

ci;j ¼��

ci;j þ 1�$1 ¼ ci;j þ 1; if bi;j is 1�

ci;j þ 0�$0 ¼ 0; if bi;j is 0

(2)

M ¼ maxðM;CÞ (3)

4. Results and analysis

The first part of the analysis concentrated on investigating thelongest runs of low DNI for each grid location across Australia, andthe implications for resource planning and forecasting. The algo-rithm in Section 3.3 is run, such that at the end of the run, theprogram returns the M matrix and a second matrix giving thestarting hour number for each of the longest runs of low DNI givenin M.

Fig. 2 shows the number of consecutive days where the DNI wasless than 400 W/m2 for every hour covering the 13 year period ofdata. From here on these events will be referred to as lulls. Theregion of Australia that experiences the longest period of lulls oc-curs in the North of the continent, particularly far North Queens-land. Interestingly, this region has excellent annual solar insolation.The longest lull was found to extend to a maximum of 20 daysduring a record breaking rainfall event. In many locations along theeastern coastline which can experience greater rainfall events thanother parts of the continent, these lulls can extend to over sevendays. This information is useful for resource planning, in knowinghow long and how often a CST plant might not be able to generate

B. Elliston et al. / Renewable Energy 74 (2015) 633e639636

power over its lifetime. Such information has value to both po-tential project developers but also power system operators andcentral planners required to assess future generation capacity re-quirements. Long outages of this type are similar in nature to forcedoutages of a conventional power station, with the exception thatoutages due to lulls are somewhat easier to forecast. Anotherimportant aspect of Fig. 2 is that it illustrates geographical regionsthat experience similar weather conditions, which tells us if wegeographically spread our solar power systems, which areas weshould avoid for experiencing prolonged lulls at the same time.

Both resource analysis and weather/power forecasting wouldrequire one to know if these prolonged lulls occur often or areextreme weather events. To pursue this line of investigation wethen ran a version of the algorithm in Section 3.3 to find out whattime of year these prolonged lulls occur. Once a seasonality of theseevents has been identified, we canmatch these lull periods to MSLPcharts and satellite imagery to narrow down the weather events.Fig. 3 shows the seasonal timing of the longest lulls of DNI below400 W/m2 for the 13 year time period. This figure shows thatprolonged lulls occur in the warmer months (Dec, Jan, Feb) alongthe eastern coastline and North Queensland, and during the wintermonths (June, July, Aug) over the southern part of the continent.This presents a potentially valuable opportunity to reduce aggre-gate variability of CST generation through the dispersion of plantsacross reasonably sunny locations from Northern to SouthernAustralia. To identify the weather patterns associated with theseprolonged lulls we looked at the monthly mean MSLP analysesderived from the Bureau of Meteorology's GASP and ACCESSmodels [17]. The broadscale atmospheric circulation patterns thattypically occur in the warmer seasons over the North of thecontinent are bands of low pressure, troughs and fromNovember toApril the monsoon trough, which can last from seven days toweeksat its peak. This brings extended periods of cloud cover and rain,which in turnwould cease operation of concentrated solar plants. Itis noted there will be some variability in the circulation patternsfromyear to year, but on average the location of the systems is whatis observed.

To identify the specific time of year and hence system thatcontributed to the longest lull in irradiance, the algorithm wasagain run to extract the specific period of time and location. Thespecific system that contributed to the prolonged lull that was

Fig. 3. Seasonal timing of longest runs of DNI below 400 W/m2. The longest lulls occurin Winter in the Southern part of the continent and over the Northern part in Summer.

observed in Fig. 2 over North Queensland is attributed to recordhigh rainfall over January/February 2009 in this region, as shown inFig. 4, which shows a decile rainfall map for January 1 to February28, 2009. The decile map indicates whether rainfall is aboveaverage, average or below average for the time period chosen. Forthis case, there were times where the rainfall was the highest onrecord, which is what contributed to this extreme event. Fig. 5shows the average irradiance over Australia for February 2009,highlighting the impact of this event for severely reduced averageirradiance in far north Queensland. In terms of both resourceanalysis and forecasting, determining how often extreme eventslike this occur is vital for both, and detailed below are two casestudy sites where this was analysed.

To obtain some insight into the frequency of such low irradianceevents, the inland location of Roma, Queensland (26.572 S,148.790 E) was chosen for closer examination. Roma was chosen asit has previously been identified as a suitable site for CST plants dueto its high average annual direct normal insolation and access to thetransmission network. The programwas again run to determine themaximum number of hours below the 400 W/m2 threshold forRoma, but we were interested in the ten longest events thatoccurred at this location and also the time of year these occurred.

Table 2 shows the results for the ten longest lull events. Thereare four instances of events of six days duration (1998, 2000, 2008,2010) in the 13 year period, again during the warmer months. Ofthe 10 longest events, four occur in one year, February, April andNovember 2000. The months in question in 2000 also experiencedabove average cloudiness during that period. The period from 1998to 2000 has also been described as one long la Ni~na phase, with theyear 2000 as the second wettest on record [18]. The main lull eventthat lasted six days in February 2000 is associated with a north-west cloud band. Fig. 6a shows the infrared satellite GMS-5 imagecontaining coverage over Australia, illustrating the cloud band.These cloud bands are extensive regions of cloud that span acrossAustralia from roughly north-west to south-east. Extensive rainfallis often associated with them. Fig. 6b is the satellite derived DNIshowing the effect of the cloud band in Fig. 6a. Irradiance values are150 W/m2 or lower across the path of the cloud band which in-cludes Roma. An important point to consider when studying Fig. 6b,is that a spatial correlation of potential plants along the east tosouth east coast of Australia would show that outages would haveoccurred to all plants during that period. Most of the lull events thatwe see in the warmer months in far north Queensland are associ-ated with the monsoonal trough or trough systems. Seven of theten longest events occurred during thewarmermonths in Australia.

Another important question we can answer using the extensivesatellite data is the length of positive runs of DNI over the thresholdof 400 W/m2. The same analysis was carried out that producedFig. 2, however we now look for the longest consecutive periodswhere the DNI is greater than 400 W/m2. Fig. 7 shows theconsecutive hours where DNI was greater than 400 W/m2. We seethat the longest consecutive periods occur in the north to north-west of the continent. Theoretically, this would identify the loca-tion as ideal for CST, however there is no transmission infrastruc-ture in the region, which rules it out as an ideal site for CST. Thesouthern coastline and all along the east coast, display lowconsecutive days compared to the rest of the continent. We how-ever can identify the regions where there are moderate consecutivedays above 400 W/m2, and from Fig. 2, the region where thenumber of prolonged lull days is low, to determine the best regionfor siting CST. Both figures show that south-west Queensland fitsthe criteria for CST. Analysis was also run to find the location thathas the shortest lull in the 13 year period compared with everyother location in Australia.We are interested in finding outwhethera site that has on average lower irradiance values but less times of

Fig. 4. Rainfall decile map for the period January 1 - February 28, 2009, showing well above average rainfall in the northern part of the continent and below average in the southernpart of the continent. Map analysis produced by the BoM.[15].

B. Elliston et al. / Renewable Energy 74 (2015) 633e639 637

outages or variability would be a better trade off when choosingpotential solar sites. The algorithm that was run to determine thelongest lulls was again used, however this time adapted to searchfor the shortest lull and the location this occurs. The minimumvalue in the result matrix M is 89 h and occurs at Curnamona, SouthAustralia (31.65 S, 139.55 E). The ten longest events obtained atCurnamona are listed in Table 3. The ten longest low irradianceevents occur in eight of the 13 years. In 2004 and 2007, two events

Fig. 5. Daily Average DNI over February 2009 from the satellite derived data.

occurred in the one year. The duration of these events (in hours)quickly tails at around 64 h. The reason for this being that if youhave a daywhere every daylight hour until sunset has DNI less than400 W/m2, then the lull cannot be broken until sunrise the nextday. A majority of the lulls occur in autumn and winter, whichcorresponds with the seasonality seen in Fig. 3.

The final piece of analysis carried out was to investigate theintra-annual variability of DNI over the 13 year period with respectto the climatology. Fig. 8 illustrates the variability in DNI, using astandard deviation of anomalies, to show how the DNI has deviatedfrom the climatology. It is interesting to see that the variability inDNI is a lot lower around the middle of the continent, and we seethe largest deviations at the top north end of the continent. Thiscould be due to changes in rainfall amounts and cloud cover overthe region. Two small anomalous bands of high variability occurover Western Australia and South East Australia. The band oversouth east Australia could be due to changes in aerosol amounts,and the cause of these anomalies will be explored in future work.

Geographical diversity of sites is an important consideration fora technology that is highly weather dependant. An initial investi-gation into the spatial distribution of CSP plants to reduce

Table 210 longest events in Roma, Queensland.

Start date End date Hours Days

2000-02-12 2000-02-18 139 62008-01-14 2008-01-20 1382010-02-27 2010-03-05 1361998-12-29 1999-01-03 1342010-09-16 2010-09-21 115 51998-04-13 1998-04-17 1132000-04-22 2000-04-26 1132000-11-13 2000-11-17 1112010-01-03 2010-01-07 1102000-11-07 2000-11-11 109

Fig. 6. a.GMS-5 infrared satellite image showing coverage over Australia and thenorth-west cloud band [3], b. Satellite derived DNI for the period 12 February 2000 at0600 UTC, showing the reduction in DNI due to the cloud band.

Fig. 8. Interannual variability of DNI over 13 years of data.

Fig. 7. Length of longest run of DNI above 400 W/m2.

B. Elliston et al. / Renewable Energy 74 (2015) 633e639638

variability was carried out based on Fig. 8, and Tables 2 and 3 Wechose to correlate DNI anomalies at two reference sites (Roma andCurnamona) with all other possible regions. Figs. 9 and 10 show theDNI correlation plots with distance with respect to each reference

Table 310 longest events in Curnamona, South Australia.

Start date End date Hours Days

1998-07-31 1998-08-04 89 42007-01-17 2007-01-20 831999-03-24 1999-03-27 71 32004-05-31 2004-06-30 662000-03-18 2000-03-20 652006-07-13 2006-07-15 652007-03-18 2007-03-21 652004-07-22 2004-07-24 642005-07-06 2005-07-08 642010-02-01 2010-02-03 64

site respectively. Clearly, regions closer to the reference sites showgreater correlations, which decreases gradually both zonally andmeridionally. Roma shows correlations above 0.9 within a greaterradius (<280 km) when compared to Curnamona (<120 km).

Fig. 9. Correlation of the variability of DNI with respect to distance from Roma.

Fig. 10. Correlation of the variability of DNI with respect to distance from Curnamona.

B. Elliston et al. / Renewable Energy 74 (2015) 633e639 639

However, characterizing variability with the probabilities ofshortest and longest lulls at each site is also crucial for solar siteselection. Using these results, further work could entail closer ex-amination of the spatial distribution of CST plants to reduce thevariability of generation using a similar approach to [19].

Understanding the broadscale weather that influences how wewould operate a solar power plant or site solar locations is vital toenabling the technology to become a player in our energy industry.The way we have investigated the occurrence of lull events is onlyone consideration in determining a suitable solar location, or howto geographically diversify sites. This analysis however is notlimited to choosing suitable solar sites, as previously mentionedanalysis of this type is also useful for forecasting and operation ofsolar plants, both large-scale and distributed.

5. Conclusion

The data analysis work described in this paper has quantifiedthe extent of long periods of low DNI that will detrimentally impacton the operation of CST power plants. The analysis finds that theseperiods are much longer than the capacity of thermal storageincorporated into current CST plants. During such an event, a CSTpower plant would require an auxiliary heating source to continueoperating (eg, from biomass or natural gas co-firing).

We also show that these events are far from uncommon,occurring almost every year and sometimes more than once in agiven year. By examining the starting day of the year for theseevents across the continent, we show that these events can occur atany time of the year, but there is a trend for them to occur inwinterin the south-eastern part of the continent and during summer inthe northern, tropical region of Australia. An unexpected result isthat the single location in Australia with the shortest low DNI eventin the 13 year period is Curnamona, South Australia, near theeastern border with New South Wales. This highlights that a sitefavoured for shorter periods of low DNI may not be a site initiallychosen for its high annual insolation. The occurrence of extendedperiods of low DNI in the summer months in Northern Australiaand winter months in Southern Australia presents a potentiallyuseful correlation for reducing the variability of aggregated CSTplants across the Australian National Electricity Market that runsfrom Northern Queensland to South Australia and Victoria. This

would address questions such as the number, siting, and spatialseparation between CST plants to minimise the impact of lulls.

Acknowledgements

This work was supported by the Australian Renewable EnergyAgency (2-A004) and the NCI National Facility at the AustralianNational University.

Solar radiation data derived from satellite imagery processed bythe Bureau of Meteorology from the Geostationary MeteorologicalSatellite and MTSAT series operated by Japan MeteorologicalAgency and from GOES-9 operated by the National Oceanographicand Atmospheric Administration (NOAA) for the Japan Meteoro-logical Agency.

References

[1] Alted F, Vilata I, Prater S, Mas V, Hedley T, Valentino A, et al. PyTables: Hi-erarchical datasets in Python. 2002.

[2] Baker DG, Enz JW. Climate of Minnesota: part XI - the availability depend-ability of solar radiation at St. Paul, Minnesota. Technical Report 316. Agri-culture Experiment Station, University of Minnesota; 1979.

[3] National Climatic Data Centre. Noaa satellite information service. World WideWeb electronic publication; 2000 [accessed on 03.09.2013].

[4] Denholm P, Hand M. Grid flexibility and storage required to achieve very highpenetration of variable renewable electricity. Energy Policy 2011;39:1817e30.

[5] Diak GR, Gautier C. Improvements to a simple physical model for estimatinginsolation from GOES data. J Clim Appl Meteorol 1983;22:505e8.

[6] Elliston B, MacGill I. The potential role of forecasting for integrating solargeneration into the Australian national electricity market. In: Solar 2010:proceedings of the annual conference of the Australian solar energy society;2010.

[7] Exell R. The fluctuation of solar radiation in Thailand. Sol Energy 1976;18:549e54.

[8] Feldhoff JF, Benitez D, Eck M, Riffelmann KJ. Economic potential of solarthermal power plants with direct steam generation compared with HTFplants. J Sol Energy Eng 2010;132:41001e9.

[9] Garnaut R. The garnaut review 2011: Australia in the global response toclimate change. Cambridge University Press; 2011.

[10] Gautier C, Diak G, Masse S. A simple physical model to estimate incident solarradiation at the surface from GOES satellite data. J Appl Meteorol 1980;19:1005e12.

[11] Hunter JD. Matplotlib: a 2d graphics environment. Comput Sci Eng 2007;9:90e5.

[12] IPCC. Contribution of working group iii to the fourth assessment report of theintergovernmental panel on climate change. Cambridge University Press;2007.

[13] Jones E, Oliphant T, Peterson P, et al. SciPy: open source scientific tools forPython. 2001.

[14] National Renewable Energy Laboratory. System advisor model (SAM). 2012.[15] Bureau of Meteorology. Climate maps. World Wide Web electronic publica-

tion; 2009 [accessed on 03.09 2013].[16] Bureau of Meteorology. Gridded hourly solar direct normal irradiance meta-

data. World Wide Web electronic publication; 2010a [accessed on 15.012011].

[17] Bureau of Meteorology. Australian climate maps. World Wide Web electronicpublication; 2010b [accessed on 03.09 2013].

[18] Bureau of Meteorology. La Ni~na - detailed australian analysis. World WideWeb electronic publication; 2013 [accessed on 12.12 2013].

[19] Perez R, Kivalov S, Schlemmer J, Hemker Jr K, Hoff TE. Short-term irradiancevariability: preliminary estimation of station pair correlation as a function ofdistance. Sol Energy 2012;86:2170e6.

[20] Ridley B, Boland J, Lauret P. Modelling of diffuse solar fraction with multiplepredictors. Renew Energy 2010;35:478e83.

[21] Weymouth G, Le Marshall J. An operational system to estimate global solarexposure over the Australian region from satellite observations. Aust Mete-orol Mag 1999;48:181e95.

[22] Weymouth G, Le Marshall J. Estimation of daily surface solar exposure usingGMS-5 stretched-VISSR observations: the system and basic results. AustMeteorol Mag 2001;50:263e78.

[23] Zhang Y, Smith SJ, Kyle GP, Stackhouse Jr PW. Modeling the potential forthermal concentrating solar power technologies. Energy Policy 2010;38:7884e97.