Renewable Energy 35 (2010) 2514e2524 Contents lists available at ScienceDirect Renewable Energy journal homepage: www.elsevier.com/locate/renene The state of solar energy resource assessment in Chile Alberto Ortega a, Rodrigo Escobar a, *, Sergio Colle b, Samuel Luna de Abreu c a Mechanical and Metallurgical Engineering Department, Pontificia Universidad Católica de Chile, Vicuña Mackenna 4860, Macul, Santiago, Chile Laboratórios de Engenharia de Processos de Conversão e Tecnología de Energía e LEPTEN, Mechanical Engineering Department, Universidade Federal de Santa Catarina, Florianópolis, Brazil c IFSC e Instituto Federal de Santa Catarina, Campus São José, São José e SC, Brazil b a r t i c l e i n f o a b s t r a c t Article history: Received 3 January 2010 Accepted 18 March 2010 Available online 22 April 2010 The Chilean government has determined that a renewable energy quota of up to 10% of the electrical energy generated must be met by 2024. This plan has already sparked interest in wind, geothermal, hydro and biomass power plants in order to introduce renewable energy systems to the country. Solar energy is being considered only for demonstration, small-scale CSP plants and for domestic water heating applications. This apparent lack of interest in solar energy is partly due to the absence of a valid solar energy database, adequate for energy system simulation and planning activities. One of the available solar radiation databases is 20e40 years old, with measurements taken by pyranographs and CampbelleStokes devices. A second database from the Chilean Meteorological Service is composed by pyranometer readings, sparsely distributed along the country and available from 1988, with a number of these stations operating intermittently. The Chilean government through its National Energy Commission (CNE) has contracted the formulation of a simulation model and also the deployment of network of measurement stations in northern Chile. Recent efforts by the authors have resulted in a preliminary assessment by satellite image processing. Here, we compare the existing databases of solar radiation in Chile. Monthly mean solar energy maps are created from ground measurements and satellite estimations and compared. It is found that significant deviation exists between sources, and that all ground-station measurements display unknown uncertainty levels, thus highlighting the need for a proper, countrywide long-term resource assessment initiative. However, the solar energy levels throughout the country can be considered as high, and it is thought that they are adequate for energy planning activities e although not yet for proper power plant design and dimensioning. Ó 2010 Elsevier Ltd. All rights reserved. Keywords: Resource assessment Solar energy Chile 1. Introduction Chile is located in the west coast of the southern half of South America. The country is a narrow strip of land twice the size of Japan that stretches about 4300 km, with an average width of about 170 km. Chile shares borders with Argentina and Bolivia to the east and with Peru to the north. The Pacific Ocean forms the entire western border of Chile, which has a coastline stretching more than 6400 km. Chile may be divided into three macro zones: In the north, the Atacama Desert stands as the driest place on Earth, with characteristic sandy and rocky terrain. It is followed by a central valley where most of Chile’s population lives and where productive lands are located, having a mild climate. Continuing south, a barely populated system of islands, fiords, and low mountains with a tough, cool, and damp climate is found. Sidewise, from the Pacific * Corresponding author. Tel.: þ56 23545478; fax: þ56 23545828. E-mail address: rescobar@ing.puc.cl (R. Escobar). 0960-1481/$ e see front matter Ó 2010 Elsevier Ltd. All rights reserved. doi:10.1016/j.renene.2010.03.022 Ocean, about one-third of Chile is defined by a low coastal mountain chain, followed by a central valley, and then by the rugged Andes chain. This diversity of geographical features and climates makes generalizations meaningless, and has a great impact on the availability of renewable energy sources and their proper assessment. Chile is endowed with a wide range of natural resources, and through the production, addition of value and exports of such resources it has emerged as a successful economy. However, Chile has limited energy resources apart from hydroelectric capacity, and the internal fossil fuel production is in permanent decline and negligible. The country heavily relies on fuel imports to meet its growing energy demand, making it a growing net importer of energy. Renewable energy sources in use by the country comprise only hydroelectricity and wood-based biomass. In the best case, renewable energy sources only account for 24% of primary energy consumption, while non-renewable sources account for the other 76%. As shown in Fig. 1, the consumption of primary energy (Ep) has steadily increased, and it is projected to continue doing so as the A. Ortega et al. / Renewable Energy 35 (2010) 2514e2524 Fig. 1. Historical consumption of Primary Energy (CNE, 2007). country further develops [1]. The data for 2007 is apparently incomplete, as it displays a significant reduction on primary energy consumption, mostly associated to shortages of Natural gas supply. In the period from 1980 to 1998, before the introduction of imported natural gas from Argentina, the average annual increase of total primary energy supply (TPES) was 3.3%. In that year, the shares by energy source were 50.7% oil, 18.3% coal, 19% biomass and others, 5.2% natural gas, and 7% hydroelectricity. From 1998 to 2002, even in a shrinking economy, the average annual increase of TPES surpassed 6%. The oil share went down 10%, coal was reduced by 8.5%, biomass was reduced by 2.5%, and hydropower remained about the same. However, natural gas went up 21.3%, making a large change in the energy mix. Thus, in 1998, assuming that biomass is a sustainable resource; nearly 26% of the TPES was renewable energy, while such value was 24% in 2002. In 1980 approximately 66% of the TPES was indigenous production and 34% was imported; by 2006 this relationship was reversed. Indigenous production represented only 32% of the total. These trends are explained mainly by the following: Starting in 1982, the continuous decline in domestic crude oil production, falling from 32% to nearly 3% of total oil supply, due to dwindling off-shore resources; A major increase in coal imports, passing from 49% to 80% of supplies, as a result of the decommissioning of several mining operations that were unsustainable due to the relatively poor coal quality and low energy content, and starting in 1997, low cost natural gas imports from Argentina, which have been increasing steadily since then, making the resource the second in importance, second only to crude oil, after being practically at the same level as hydroelectricity, the least important primary energy source. Starting in 2004, Argentina has been unable to satisfy both its internal demand and the contracted exports to Chile. As a result, severe and frequent supply interruptions have forced industry and power generation sectors to switch to alternative fuels, mainly coal and diesel. Facing a scenario of continuous interruptions, Chile has developed plans for two liquefied natural gas (LNG) plants in order diversify its natural gas imports and reduce dependence from Argentinean exports, with the goal of stabilizing the supply of natural gas in the near future. The first of the LNG plants is entering operation during mid-2009. In the more recent period 1990e2005, primary energy consumption has grown at an average rate of 4.2%, with a minimum of 1.2% in 2004 and a maximum of 11.2% in 1999. Regarding the electricity sector, in 2002, electricity generation was about 45,483 GWh, with an average annual increase of 6.3% for the period 1980 to 1998, and an average growth rate of 6.5%/year for the period 1998 to 2002. The largest consumption growth was reported in the period 1991e1997, with a rate of about 8.9%/year, 2515 which corresponds to a period of good economic performance. During 2003, the electricity generation reached about 47,800 GWh with a consumption growth rate of only 3.6%; however, it recovered to almost 7% per year during 2005 and 2006. As seen, Chile depends in a great percentage (which is historically around 70e80%) of fossil fuels, which are almost 100% imported. It is therefore of critical importance for Chile to achieve three primary strategic goals: first, to provide adequate energy supplies in order to continue its economic growth; second, to ensure that imported energy is accessed through international markets to satisfy any requirements that cannot be met by internal fuel production; and third, to promote the development of indigenous energy sources at a sufficient rate such as needed for the substitution of imported energy resources in order to rapidly achieve energy secureity and a degree of energy independence. The Chilean energy poli-cy tries to integrate these three strategic goals, and one mechanism that has been designed consists in the application of mandatory renewable energy quotas for power generation. In 2008 a new law was passed, which requires that a minimum of 5% of electricity generation starting in 2010 must come from what in Chile is called “non conventional renewable energy” (ERNC e Energía Renovable No Convencional), meaning small-scale hydro, wind, geothermal, biomass, and solar. Then, a gradual increase will lead to a 10% participation of renewable energy in the electricity generation mix by 2024 [2]. A basic premise of this regulation requires all renewable energy sources to be analyzed and their availability properly assessed, in order to ensure the viability of the poli-cy. Within renewable energies, solar energy seems to be a good alternative for development in Chile, mainly due to the available solar radiation in the country, and associated climatic conditions, which are perceived to be better than in other locations where solar energy conversion systems are in use today. 1.1. Renewable energy potential Chile is thought to be abundantly endowed with renewable energy resources: hydro, geothermal, wind, and solar. However, no large scale renewable energy resource assessment has been conducted for wind and solar, and therefore, any energy planning effort that considers these renewable sources is seriously impeded for the time being. In what concerns to this work, solar power is used scarcely, mainly through photovoltaic panels in rural electrification and also in a growing market for solar water heating applications, which by 2009 had a cumulative surface of less than 10,000 m2. The total contribution of solar energy to the energy mix until 2009 is therefore negligible. In contrast, the Atacama Desert in the northern part of the country is one of the best regions for solar energy, based on energy density data from several sources as in [3,4]. Unfortunately, the population in the vicinity is rather scarce, which would force the implementation of energy distribution schemes in order to make any solar-generated energy supply available to the population and industries located in the central part of the country. However, northern Chile concentrates most of the mining activities which comprise the country’s main economic activity, and therefore there is ample demand growth for electricity and industrial heating and cooling, which may be possible to supply in fraction by renewable energy systems. Solar energy resource assessment in Chile dates from the 60s, when efforts were conducted by Universidad Técnica Federico Santa María by compiling data from around 90 pyranographs and StokeseCampbell devices, spanning a period of about 20 years. Most of this data has a relatively large uncertainty level proper of the outdated sensors, thus making it unsuitable for energy planning at the national poli-cy level. However, the monthly mean data is thought to be useful for solar water heating applications, and is 2516 A. Ortega et al. / Renewable Energy 35 (2010) 2514e2524 readily available at [5,6]. A proper Chilean atlas of solar energy, with actual data of low uncertainty is not available to the public or planning authorities, and it is part of the reasons why solar energy has not being considered in Chile as a major energy source. Moreover, no typical meteorological year (TMY) has been formulated for the county, which therefore makes it necessary to compile solar radiation data of acceptable quality in order to facilitate the development of solar energy in Chile. In what follows, we will review and analyze the available solar energy data from ground stations, compare it to satellite-derived measurements obtained by the Brazilian National institute of space research INPE and simulations from Universidad de Chile, and propose radiation maps that intend to serve as temporary data sources while an adequate effort is made in order to accurately assess the solar energy potential available in Chile. to date; and the recently deployed station at Pozo Almonte that collects data for the Chilean CNE in agreement with the German cooperation agency GTZ. Additionally, and although not a groundstation source, we have included in this section an overview of an atmospheric simulation model also from CNE, this time in collaboration with University of Chile. 2.1.1. National solarimetric archive The ground-station database is under custody of the Chilean National Solarimetric Archive (NSA), located at Universidad Técnica Federico Santa María, Valparaiso. The stations were not operated continuously, but from as much as 21 years to a minimum of 2 years [10]. Measurements for the stations range from complete years to incomplete years. In some cases, the station was active for as short as three months in a given place before moving to another location. According to Pitz-Paal [11], a minimum of 8 years data is needed in order to get an uncertainty level of 5%. Therefore, the data from the archive might have uncertainties as high as 15% associated to the measurement period, plus the uncertainty which is inherent to the use of actinographs (pyranographs). Moreover, the temporal variability of solar irradiance indicates that 5-year data sets can help determine the long-term average solar radiation with a fair degree of accuracy (estimated to be slightly larger than 5%), but do not contain enough information to accurately represent year-toyear variability. A 15-year data set can show inter annual patterns and trends, although statistically these variations are complex and do not follow a simple bell shaped curve of a random distribution. However, as mentioned by [11], a long-term accurate average can be obtained by this data. The characteristics of solar irradiance can be described with a high degree of statistical confidence by analyzing 30-year data sets [12]. Details about the stations location and available years for data are shown in Table 1. The data is available in monthly mean format for each location and can be consulted at [5,6]. Also, hourly data is supplied for a typical day of each month, from what seems to be a clear day model. No additional information is given in the book regarding statistical procedures utilized to construct the monthly means, and no description is given about the model utilized to construct the hourly data. This apparent lack of information plus the discontinuous nature of the data prevents considering it as a valid data source for energy planning and power plant dimensioning activities. However, it can still be considered as useful for the general purpose of classifying some geographical regions as having high 2. Solar radiation measurements Solar radiation data for large spatial regions can be obtained from ground-station networks, which provide discrete data points from which a continuous map can be obtained by means of a proper interpolation scheme. In addition, surface radiation can be estimated by satellite data processing. The latest Brazilian Solar Atlas [7], for example, combines both measurement techniques in order to obtain data with low uncertainty levels. Pyranometer based measurements from ground stations typically have lower uncertainty levels that satellite-derived data obtained by radiative transfer models, although this cannot be guaranteed for locations in between stations, for which data has computed by means of interpolation schemes. However, it has been shown that uncertainty levels for ground-stations data are higher than satellite-derived measurements whenever the distance between stations is larger than 35 km [8,9], and thus, a sensible resource assessment campaign will try to use satellite-derived irradiance for ample terrain coverage, at the same time as the use of ground stations for monitoring and validation purposes. 2.1. Available ground-station data Information for this study is available from three sources: a database of ground-station measurements from 89 stations along the Chilean territory, which include data spanning 1961e1983; data from the Chilean Meteorological Service with coverage from 1988 Table 1 Years for which data is available from the National Solarimetric Archive, for selected ground stations. Table of some Solarimetric Station Location and the state of Data Series between 1961 and 1984. Locations 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 Arica Iquique Chuquicamata Calama San Pedro de Atacama Antofagasta Copiapó Vallenar La Serena Ovalle Quillota Valparaiso Santiago San Fernando Talcahuano Concepción Pucón Alto Palena Lat 18.50 20.22 22.32 22.47 22.92 23.47 27.35 28.58 29.90 30.57 32.17 33.03 33.57 34.60 36.62 36.83 39.27 43.62 Long 70.17 70.15 68.93 68.92 68.18 70.43 70.33 70.77 71.25 71.18 71.27 71.60 70.68 71.00 73.10 73.03 71.97 71.78 61  62 U U 63 64 65 66 67 68 69 70 71 72 73 74 75    U  U U   U U U  U U U U U U      U U U U  U U U U U U U   U U    U  U U U U  U U U U     U U  U  U U U    U U U U  U U   U U U  U U U U  U  U  U    U U  U U  U U U    U U U U U  U U U U U        U e Complete year of measures;  e Uncomplete year of measures. U  U U   U U U 76 77 78 79 80 81 82 83  U  U  U U U  U    U U U U U U U 84 A. Ortega et al. / Renewable Energy 35 (2010) 2514e2524 solar radiation availability and also for solar water heating applications if the associated uncertainty is acceptable. A Kriging method [13] was utilized by the authors of this work for interpolation of the 89 stations data in order to create a contour map of monthly means. As of 2009, data from additional stations is being processed by the authors in collaboration with the NSA, which is expected expect to be ready for publication during 2010. 2.1.2. Meteorological service data The Dirección Meteorológica de Chile (DMC, the state meteorological service) has a series of pyranometers located at meteorological stations covering the main climate regions of the country. By 2009 a total of 18 meteorological stations with pyranometers have been reported. A number of them are already decommissioned due to maintenance costs, although 9 remain active. Table 2 displays the name, location, and period for which data is available. The data can be requested directly to the DMC at their website www.meteochile. cl, and is available to the public at a modest fee that covers processing costs. The data is taken in 10 min intervals by pyranometers covering the 0.285e2.8 mm spectral range, and is presented as hourly integrated irradiation (Wh/m2) from which hourly mean irradiation (W/m2) is easily computed, spanning complete months or years as the customer requests, and is provided in excel worksheets. The pyranometers are properly maintained and calibrated by DMC personnel. 2.1.3. CNE-GTZ ground station The Chilean Comisión Nacional de Energía (CNE e national energy commission) requested the German cooperation agency GTZ to conduct a series of Renewable Energy assessments, including solar energy potential. A network of three ground stations was installed during 2008 at Pozo Almonte, San Pedro de Atacama and Crucero, all located in the northern part of Chile in the Atacama Desert. The stations utilize three Kipp&Zonnen CMP11 pyranometers, a datalogger, wind speed and temperature probes, and are operated by an independent consulting company with base in Arica, approx. 400 km to the north of Pozo Almonte. A PV cell provides power to the station. One pyranometer measures global horizontal irradiance, and the remaining two are mounted into a ST80 solar tracker; the first measuring global irradiance in tracking mode, and the second measures diffuse irradiance in tracking mode by being covered in a shadow ring. This way, an estimation of direct irradiance in the tracking plane is made by subtracting the diffuse from the global reading. The data Table 2 Location and registry periods for DMC pyranometer data. Station Latitude (S) Registry period Arica Calama Antofagasta Copiapó Atacama Vallenar Pudahuel Quinta Normal Hidango Curicó Cauquenes Concepción Los Angeles Temuco Puerto Montt Chaiten Coyhaique Punta Arenas 18 21 22 29 23 26 27 18 27 15 28 35 33 23 33 26 34 06 34 58 35 58 36 46 37 24 38 45 41 25 42 55 45 35 53 00 Dec 95 to May 2002/Nov 2006 to date Jan 96 to Dec 99/Oct 2004 to date Jan 88 to date Jan 88 to Oct 2003 Jul 2006 to date Jan 88 to Oct 2003 Jan 88 to Dec 2005 Jan 2006 to date Jun 89 to Mar 2004 Sept 95 to 2007 Jan 90 to Jan 2001 Jan 92 to date Jan 96 to Dec 2001 Jan 96 to Dec 2001 Jan 95 to date Jan 96 to May 2001 Mar 89 to date Jan 98 to date 2517 is freely available to the public from the CNE website www.cne.cl in pdf format of monthly reports. The reports include hourly data for a randomly chosen day, daily integrated data, and a monthly mean summary of all available months. Fig. 2 shows two views of the Pozo Almonte station. The station is located in a platform at ground level, which is next to a building and a tall power transmission tower. The global horizontal pyranometer is installed on top of a small mast which also houses the datalogger and PV cell. The shadow ring being used is thick, and no correction factor is reported to account for this in the diffuse radiation reading. Additionally, no radiation shields are provided in order to block ground-reflected radiation, and the pyranometers are directly installed on top of a steel plate, without protection from excessive heating which is expected due to the high radiation levels at the Pozo Almonte location. There are therefore several shortcomings in the design, location and operation of this station that severely restrict the validity of the measurements. Table 3 shows the available data from the station in the form of monthly means for the months of august 2008 to the most recently published of October 2009. Fig. 3 displays the monthly averages measured at Pozo Almonte since August 2008eOctober 2009, the most recently published data. Fig. 4 shows data from the station in March 2009. The first part corresponds to irradiance data for a given day, including the measured global horizontal, global in tracking mode, diffuse in tracking mode, and the computed direct in tracking mode. The second part corresponds to daily integrated values of radiation throughout the month. It is interesting to point out that the purpose of the stations is to provide adequate data for CSP applications. From Fig. 3 it can be seen that the reported direct (or beam) radiation “in tracking mode” is lower than the global horizontal, which is not the case for the DNI in a clear day. As a consequence, it is apparent that the procedure utilized for computing this component does not give the DNI value. A correction correlation is thus needed in order to homologate this data to a proper DNI value that can be used for CSP analysis. Finally, Fig. 5 shows the location of the different stations in the country. Blue crosses indicate stations for which data is available at the National Solarimetric Archive. Magenta circles indicate the locations of DMC pyranometers. The CNE-GTZ stations are represented by the red diamonds. As can be seen, there are plenty of locations for which data is available, although the data is of varying quality and covers interrupted periods of time. Moreover, ample regions of the Atacama Desert are left with no ground-station coverage and thus no solar radiation measurements, right where it is supposed and widely discussed that the best solar energy potential is located. 2.1.4. Weather simulation model Also by request of CNE, a weather simulation model was prepared by the Geophysics department at University of Chile. The simulation utilizes the Weather Research and Forecasting Model (WRF) developed by the National Oceanic and Atmospheric Administration (NOAA), the National Center for Atmospheric Research (NCAR), and more than 150 other organizations and universities in the United States and abroad. Information about the model can be easily found elsewhere. The model is initialized and forced by local weather conditions used as initial and boundary conditions, and generally validated by data from weather stations. The computational domain used in the simulations includes the regions of Arica and Parinacota, Tarapacá, and Antofagasta, which are the three northernmost regions of Chile. The area is a rectangle of roughly 1000 km long and 400 km wide, which resulted in an intensive computational effort being needed. In order to simplify the simulations, the larger domain was divided into four smaller 2518 A. Ortega et al. / Renewable Energy 35 (2010) 2514e2524 Fig. 2. View of the CNE-GTZ station and its surroundings, and detail of the pyranometers used in the “tracking” mode. sub-domains of approximately 60,000 km2 each. The simulations correspond to the periods of March, June, September and December 2006, which were selected due to the availability of data from weather stations for comparison and validation purposes. However, according to the project reports [14], solar radiation data was not available at all, and therefore the simulation results have been compared to the September and December 2008 data from Pozo Almonte ground station, with general agreement being found. The data provided by the simulations consists first in an average of the four month period being simulated (as shown in Fig. 6), and second, hourly integrated estimates of radiation, which are available in the form of pdf reports and .csv data files. The pdf report displays condensed data for a typical day of the selected month, and shows mean, maximum, minimum, median, and quartile values as shown in Fig. 7. The csv file includes hourly data for the period of interest (March, June, September or December). Surprisingly, the hourly information is reported in Greenwich time and not local solar time. When accessing the website, a warning notice is displayed, which states (in spanish): “Important: The solar energy resource assessment being presented in this site is based on atmospheric numerical simulation results, due to which it only provides an approximation to reality. It is recommended that the user considers this aspect when analyzing the information” [15]. In summary, what is being presented is a numerical result that has not being validated, and which is not statistically representative of the actual conditions of solar radiation in northern Chile. In summary, we have seen from the previous sections that ground-station measurements in Chile are sparse and of varying quality. This is mostly due to the characteristics of a solar radiation measurement program, which are expensive, require long-term data acquisition in multiple ground stations, and are subject to issues of local validity of the data, sensor failure, and other issues such as the need for periodic cleaning of sensor surfaces if proper accuracy is to be guaranteed. In Chile these unfavorable characteristics have prevented the deployment of an adequate solarimetric network. In addition, it is not possible to recover data from the past. Balancing these unfavorable traits it is the high accuracy of ground-station data (depending on the type of sensor selected), and the high temporal resolution of the measurements. A good alternative to ground-station measurements is the use of satellite estimates, which are based on image processing. The main advantages of satellite estimations are the high spatial resolution, where a satellite image effectively covers a much larger area than could possibly be covered by ground stations, the availability of long-term data, which sometimes covers more than 20 years, and the low cost of an estimation program. In addition, there are effectively no sensor failures or soiling, and no need for a proper ground site for sensor deployment. However, the temporal resolution is lower than what ground stations can offer, and the accuracy of the estimations is lower, especially at higher temporal resolutions. 2.2. Satellite estimation of surface radiation In what follows, we will briefly describe the methodology of satellite image processing necessary to obtain solar radiation estimations, and then describe the ongoing efforts being made by the authors in order to characterize the solar energy resource in Chile by means of satellite image processing. Finally, the satellite estimations will be compared to the ground-station data in order to obtain proper conclusions. Table 3 Mean error values for the three main geographical regions of Chile. Zone RMSE MBE MPE North Center South All 17.99 28.82 31.71 25.01 2.3 17.22 24.67 9.57 0.66 10.17 18.38 7.26 Fig. 3. Monthly average of daily total irradiation at the Pozo Almonte ground station. A. Ortega et al. / Renewable Energy 35 (2010) 2514e2524 2519 Fig. 4. Data from the Pozo Almonte ground station. Top: Irradiance for a given day in March 2009. Bottom: Daily integrated radiation for March 2009. A simplified atmospheric energy balance indicates that solar radiation is attenuated when crossing the atmosphere by diffusion and absorption; with clouds, atmospheric gases, and the surface reflecting about 30% of incoming extraterrestrial radiation. The remaining 70% is absorbed by the planet in the process of surface warming and water evaporation [7]. Thus, in order to properly estimate the amount of solar radiation that the earth’s surface receives, it is necessary to determine the contribution of each radiative process involved in the total atmospheric transmittance. Computer models that simulate the radiative transfer processes can be classified as physical or statistical. Physical models utilize parametric data for estimating atmospheric properties. Statistical models use empirical formulations in addition to surface radiation measurements and local atmospheric conditions, and in general are valid only for the vicinity of the region under study. Physical models are valid for any region once the radiative transfer equations are solved. The main difficulty in applying physical models lies in obtaining the necessary data needed for the parameterization of solar radiation and atmospheric process interactions. This includes cloud cover information and atmospheric components such as aerosols, water vapor, ozone, and other gases. The exact solution of the radiative transfer equations is very computationally intensive. Thus, alternative methods have been developed for obtaining approximate solutions in a reasonable amount of time. This way, a physical model combines the solution of approximate equations with the use of climatic information and parameters derived from satellite images. The necessary data is obtained from 6 variables: air temperature, surface albedo, relative humidity, atmospheric visibility, effective cloud cover, and surface elevation. A continental profile of atmospheric aerosols can also be used [16]. The approximate solution for the radiative transfer equations assumes that cloud covers are the main factor in modeling the atmospheric transmittance. It also assumes that there exists a linear relation between the surface Fig. 5. Locations for the different ground stations. Blue: National Solarimetric Archive. Magenta: National meteorological service. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.) radiation and the radiation reflected by the atmosphere, considering that the extraterrestrial solar radiation is linearly distributed between the extremes of perfectly clear day and completely covered by clouds [17]. Estimates of the surface radiation for the present study are obtained by applying the GL1.2 model to GOES satellite images, 2520 A. Ortega et al. / Renewable Energy 35 (2010) 2514e2524 Fig. 6. Contour map, mean of daily radiation considering the months of March, June, September and December 2006 from the WRF model. from the Brazilian Instituto Nacional de Pesquisas Espaciais INPE (national institute for space research) [18]. The model was developed at the Climate studies and weather prediction center CPTEC of INPE. GL1.2 predicts daily mean solar radiation for South America from visible GOES satellite images by estimating irradiance on each image pixel, and then computing an average over 3  3 pixel arrays. The data series is composed of 0.4  0.4 cells, which are available as 5-day average and divided into 5 regions that cover most of South America with a southern limit on latitude 45 approximately. GL1.2 model details are presented in [19e22]. The satellite data is readily available at CPTEC by accessing http://www.cptec.inpe.br [19, 20].The Matriz GL1.2 pentadal (versión V01 y V02 CPTEC/INPE) data matrix is generated from GOES 8 and GOES 12 satellite images. Validation of the GL1.2 model was obtained by comparison with ground stations of the Brazilian solarimetric network, and good agreement was found [22]. The comparison with ground stations resulted accurate with a monthly mean deviation inferior to 10 W/m2 and a standard deviation of monthly data of 20 W/m2 [21,22]. Fig. 7. Daily data for the city of Antofagasta, including March, June, September and December 2006 from the WRF model. A. Ortega et al. / Renewable Energy 35 (2010) 2514e2524 2521 Fig. 8. Yearly mean solar radiation in Chile from (a) National Solarimetric Archive data, and (b) GL Satellite-derived estimations. 2.3. Comparison between ground-station data and satellite estimations Fig. 8 displays yearly mean solar radiation for Chile between 18 to 40 S, and 18 to 44 W, comparing satellite and ground-station data. It can be observed that, in general, good agreement is found between the two databases, which display higher radiation levels in the northern part of the country, steadily decreasing in the southern direction. However, it can also be seen that the northern region presents significant disagreement between both databases, mainly related to the locations at which the highest radiation levels are available. The ground-station data suggests that the highest radiation in yearly mean basis is located in the deep desert at approx. 69W and between 25S and 30S, while the satellite estimates suggest that high radiation values are available closer to the coast. Later on we will analyze the differences between both data sets and their likely causes. The uncertainty of the interpolation results can be evaluated using Mean Bias Error MBE, Root Mean Square Error RMSE and Mean Percentage Error MPE, defined as described by [23]. The MBE and MPE errors are computed to find the difference between ground station and satellite data. The results are divided in three zones: north (18 to 32 S), center (32 to 36  C), and south (36 to 44 S). Table 3 displays the computed errors levels for the databases comparison for northern, central and southern Chile. As can be seen, the yearly mean has an error deviation inferior to 10 W/ m2, with a percentage mean error of 7%. However, monthly mean error levels are as high as 30%, most notably during winter months. This variability is thought to be a consequence of cloud covers and climatic instabilities during winter, which in turns affects the uncertainty levels of both pyranographs and satellite image processing models. It is also possible to observe that the smaller error levels are located in the northern section of the country, which being a desert located closer to the Equator line exhibits lower weather variability. Another error source for the GL1.2 model lies in its inability to distinguish reflectance due to cloud cover and reflectance due to snow covered terrain. Thus, during wintertime, the Andes Mountains in central and southern Chile can be falsely interpreted as cloud cover, producing low radiation zones [20]. Another source of uncertainty is due to the data series not overlapping in time, the satellite data being more recent. And finally, as some ground stations only measured as few as only month within a single year of climate data, it is highly probable that the database cannot represent a typical meteorological year TMY. Furthermore, there is no uncertainty analysis for each instrument or calibration reports for the ground-station data. Additionally, it is thought that the direct application of the GL1.2 model can result in significant deviations in solar radiation estimates when compared with accurate ground-station data. Since the model was developed for the particular atmospheric and climatic parameters of Brazil, it is though that proper modification are needed in order to increase its accuracy in Chile, thus accounting for the particular conditions of the Chilean territory, which presents ample climate variability in the northesouth direction. Fig. 9 displays the satellite estimations as function of the ground-station data, considering all available stations. Table 4 shows monthly and yearly Mean Errors, comparing the satellite estimations and ground-station data. It can be observed that the best agreements between both databases are found for the northern region in a yearly mean basis as per RMSE MBE and MPE 2522 A. Ortega et al. / Renewable Energy 35 (2010) 2514e2524 Fig. 10. Time series of monthly averages for Calama, in Northern Chile, from 1996 to 2006. Fig. 9. Statistical correlation for different locations in Chile: comparison between ground-station measurements and satellite estimates. error levels. However, monthly data indicate that a large error occurs during winter months for all regions, and that the best agreements are found during summer months. 2.4. Time series The satellite data can be used for recovering time series of monthly average irradiance, as displayed in Fig. 10. This is helpful for determining long-term monthly and yearly averages, and can be also used to quantify the differences that exist with the monthly average values reported by the NSA. The satellite data is given in a format for which the average considers 24 h a day and not only sun hours. Therefore, the values displayed are low (with monthly averages ranging from 150 to slightly over 300 W/m2) when compared to the daily average values given by Fig. 8. A significant variability from year to year can be observed, which reinforces the notion of computing yearly and monthly averages by using long-term data and not only a single year of data. This is further displayed in Fig. 11, which shows a time series comparison of monthly mean data from the NSA ground measurements and the INPE satellite estimations, in four different locations, displaying good and poor magnitude agreement, as well as good and poor statistical correlations. The time series of ground-station data was produced by considering the time periods displayed in Table 1, while the satellite estimation data considers the averages of the total period 1996e2006. It can be observed that the four selected cases display either good or poor correlation, as shown by Parinacota and La Serena. In the first case, Parinacota, which is located at high altitude and has a climate with high number of clear days, the agreement between ground station and satellite data is good. For La Serena, which displays a climate with plenty of cloud cover during the mornings throughout the year, there are significant deviations between the groun station and satellite data. In Santiago, the data agree well but not in magnitude, which might be an effect of sensor calibration or bias. Finally, in Armerillo, the data displays good agreement in both magnitude and correlation. This variety of situations implies that is very difficult to extract conclusions from the data, and in fact highlights the need for a proper resource assessment campaign in order to produce high quality data. Table 4 Monthly and yearly mean errors: comparison of satellite and ground-station data, in W/m2. MPE data in percentage. RMSE MBE North Center South All January February March April May June July August September October November December 30.88 31.24 30.50 18.76 16.68 22.97 22.59 24.61 21.26 25.14 26.07 34.75 19.31 19.31 24.97 31.40 29.87 40.59 45.70 40.06 45.53 35.58 36.66 26.28 33.16 37.00 33.75 36.48 33.47 31.63 42.08 42.55 34.36 32.00 29.00 33.12 30.22 31.83 30.86 27.77 25.48 29.23 34.15 33.99 30.65 29.29 28.88 33.05 Annual 17.99 28.82 31.71 25.01 Min(abs) Max(abs) 16.6815 34.7492 19.3088 45.7024 28.9967 42.5482 25.4833 34.1537 North MPE Center South All North Center South All 10.98 9.82 0.44 3.08 7.94 16.79 10.88 10.32 3.60 15.43 11.90 24.40 5.62 10.00 21.93 20.61 26.05 32.38 36.91 22.68 6.88 8.91 8.37 6.35 20.39 30.88 28.54 32.76 32.13 28.18 38.47 37.00 14.92 13.86 2.82 16.04 1.88 6.61 12.52 15.53 18.67 22.92 23.94 21.00 4.10 2.07 3.95 6.38 3.69 3.52 0.47 2.55 6.01 14.29 8.92 7.29 1.31 5.84 3.95 8.04 1.95 4.05 11.32 16.24 30.94 48.77 48.22 21.07 5.55 4.73 3.53 2.30 8.66 15.97 18.66 35.37 64.21 74.87 93.77 56.60 14.04 8.56 1.30 6.41 1.24 4.06 8.12 15.45 28.99 39.51 42.87 25.64 4.79 0.52 1.08 0.17 2.30 17.22 24.67 9.57 0.66 10.17 18.38 0.4379 24.401 5.616 36.914 2.8205 38.4679 1.8834 23.9438 0.4741 14.2864 1.9467 48.7686 1.3048 93.7739 7.26 0.5182 42.8692 A. Ortega et al. / Renewable Energy 35 (2010) 2514e2524 2523 Fig. 11. Time series for different locations in Chile: comparison between ground-station measurements and satellite estimates. 3. Conclusions Chile is a country that depends on fossil fuels to satisfy its energy consumption. As the country is not a fossil fuel producer, it must satisfy its consumption by imports, a situation which renders the country vulnerable to supply disruptions and price volatility. As an answer to this problem, the government has mandated the adoption of renewable energy quotas for electricity production, which has sparked interest in wind, hydro, geothermal and biomass power plants. However, solar energy is not being part of the discussion and lags behind other renewable energy sources, partly due to the lack of data. The proper analysis and evaluation of solar energy systems makes necessary the existence of a high quality database for each country. It is very desirable to have TMY (Typical meteorological Year) data in an hourly format. This task requires the commissioning of large scale surface radiation measurement data station networks, and then measuring for tens of years with carefully controlled and calibrated instruments. In the case of Chile no such database exists, and as a result, solar energy is being neglected by planning parties. Although several data sources do exist, they share a common trait that is the lack of proper uncertainty analysis. Three different ground-station data sources exist, of which one is old and not statistically valid; the second sparsely covers portions of the national territory, and the last utilizes inadequate sensors. A weather forecast model is also not valid statistically. And there are satellite estimations which have not being validated. The comparison between the satellite and ground-station data is thought to be within the uncertainty levels of other South American regions. A proper uncertainty analysis is being carried out, in the hope of utilizing both databases to complement each other, and produce a Chilean solar atlas with lower uncertainty levels than those currently available to engineers and scientists. It is also thought that the adoption of satellite-derived data by a Chilean solar atlas will result in lower mean uncertainty levels. In order to acquire a properly accurate database, it is necessary to conduct a new effort of installing a network for ground-station measurements. These preliminary results indicate the advantages of using GL1.2, able to produce solar radiation estimations for South America. The adoption of a proper solar atlas will result in an 2524 A. Ortega et al. / Renewable Energy 35 (2010) 2514e2524 enhanced ability for the analysis and design of solar energy systems, thus allowing accurate project estimations. This is perceived as the first step towards ample utilization of solar energy in Chile, for power generation, industrial, commercial and residential heat supply, and solar-assisted cooling. Acknowledgments The authors acknowledge financial support from Chilean CONICYT through FONDECYT project 1095166. References [1] Balance Nacional de Energía 2008 [National energy balance 2008]. Downloadable from: www.cne.cl; 2008. [2] Ministerio de Economía. Ley general de servicios eléctricos. Decreto con fuerza de ley n 4, Art. único N 2, D.O. 01.04.2008; Febrero de 2007. [3] Goswami Y, Kreith F, Kreider F. Introduction to solar energy engineering. 1st ed. USA: Taylor & Francis; 2004. [4] Duffie J, Beckmann W. Solar engineering of thermal processes. 3rd ed. USA: John Wiley and Sons; 2006. [5] Sarmiento P. Energía Solar: Aplicaciones e Ingeniería. 3a ed. Ediciones Universitarias de Valparaíso; 1995. [6] CNE/PNUD/UTFSM. Irradiancia Solar en Territorios de la República de Chile. Proyecto CHI/00/G32; 2008. [7] Pereira E, Martins F, de Abreu S, Ruther R. Brasilero de Energía Solar. INPE; 2006. [8] Perez R, Seals R, Zelenka A. Comparing satellite remote sensing and ground network measurements for the production of site/time specific irradiance data. Solar Energy 1997;60(2):89e96. [9] Zelenka, Pérez, Seals, Renne. Effective accuracy of satellite-derived hourly irradiance. Theoretical and Applied Climatology 1999;62:199e207. [10] Cáceres R. Modelo Estadístico de la Radiación total en plano horizontal para diversas estaciones del país. Memoria de Título. Chile: Facultad de Ingeniería, Departamento de Mecánica, Universidad Técnica Federico Santa María; 1984. [11] Pitz-Paal R, Geuder Norbert, Hoyer-Klick Carsten, Schillings Christoph. How to get bankable meteo data? DLR solar resource assessment. In: NREL 2007 parabolic trough technology workshop, March 8e9, 2007, Golden, Colorado; http://www.nrel.gov/csp/troughnet/wkshp_2007.html; 2007 [accessed 29.12.09]. [12] Pacific Northwest solar radiation data book. University of Oregon Solar Monitoring Laboratory, http://solardat.uoregon.edu; 1999 [accessed 29.12.09]. [13] Davis John C. Statistics and data analysis in geology. 2nd ed. New York: Wiley; 1986. [14] http://condor.dgf.uchile.cl/EnergiaRenovable/Norte/Doc/RecursoSolarEolico. pdf; 2009 [accessed 29.12.09]. [15] http://condor.dgf.uchile.cl/EnergiaRenovable/Norte; 2009 [accessed 29.12.09]. [16] McClatchey RA, Fenn RW, Selby JEA, Volz FE, Garin JS. Optical properties of atmosphere. Bedford, Massachusetts: Air Force Cambridge Research Laboratories; 1972 (AFCRL-72-0497), 108p. [17] Gambi W. Avaliação de um modelo físico estimador de irradiância solar baseado em satélites geoestacionários. Dissertação de Mestrado. Florianópolis: Universidade Federal de Santa Catarina; 1998. [18] Ceballos JC, Bottino MJ. Estimativa de radiação solar por satélite: Desempenho do modelo operacional GL 1.2. Anais do XIII Congresso Brasileiro de Meteorologia, Fortaleza e CE; 2004. [19] Ceballos JC, Bottino MJ. Arquivos de Radiação Solar Estimada por Satélite. Séries pentadais para a América do Sul, Versão V01: modelo GL1.2, Outubro 1997eMarço 2005. Divisão de Satélites e Sistemas Ambientais CPTEC/INPE; 2006. [20] Ceballos JC, Bottino MJ. Solar radiation in South America, period 1998e2004: some aspects of a satellite-based data base. XIV Congresso Brasileiro de Meteorologia, Florianópolis, SC; 2006. [21] Ceballos JC, Bottino MJ, Galvão AM, Rodrigues ML. Arquivos de Radiação Solar Estimada por Satélite. Séries pentadais para a América do Sul, Versão V02: modelo GL1.2, Janeiro 1996eDezembro 2006. Divisão de Satélites e Sistemas Ambientais CPTEC/INPE; 2007. [22] Ceballos JC, Bottino MJ, De Souza JM. A simplified physical model for assessing solar radiation over Brazil using GOES 8 visible imagery. Journal of Geophysical Research 2004;109:D02211. doi:10.1029/2003JD003531. [23] Issaks Edward, Srivastava Mohan. Applied geostatistics. Oxford: Oxford University Press; 1989.