Solar irradiance, sunshine duration and daylight illuminance. derived from METEOSAT data at some European sites

Transcription

1 Solar irradiance, sunshine duration and daylight illuminance derived from METEOSAT data at some European sites by Jan Asle Olseth and Arvid Skartveit Geophysical Institute, University of Bergen, AllJgaten 7, N-57 Bergen, NORWAY ( Theoretical and Applied Climatology, in press) Pages 1-2 Figs. 1-9, Tabs. 1-2

2 Summary Radiometric ground truth data from seven Norwegian stations (58-64 N), and from five European stations (38-61 N), are in the present paper compared to satellite-derived data. Hourly global irradiance at ground level is estimated by the Heliosat procedure from the "visible" channel of the geostationary satellite METEOSAT. With increasing latitude this satelllite sees the earth's surface at an increasingly unfavourable angle. The global irradiance estimates nevertheless reproduce our high latitude ground truth data with negligible Mean Bias Deviations (MBD) and only minor deviations regarding frequency distributions. Moreover, the Root Mean Square Deviations (RMSD) are comparable to those typically seen between ground truth stations some 2-3 km apart. By a number of auxiliary models, a multiplicity of solar radiation resource data at ground level is obtained from these satellite-derived global irradiance data, and made available at the SATEL-LIGHT www server. The accuracy of the half-hourly data thus derived from Heliosat global irradiances, by models for diffuse fraction, luminous efficacy and slope/horizontal ratio, is successfully verified against ground truth data in the present paper. 1. Introduction Detailed knowledge about the spatial and temporal distribtuion of solar radiation resources are needed in order to analyse the variation of a number of solar affected processes related to e.g. meteorology, hydrology, agriculture, daylighting, and solar energy utilization. The density of radiometric stations is, however, limited for economical reasons, and in many cases only satellites can realistically offer the required data. Images taken from geostationary satellites like METEOSAT are therefore a valuable source to retrieve solar radiation data with an almost continuous spatial and temporal coverage (Ineichen and Perez, 1999). With increasing latitude, however, the accuracy of such retrievals declines due to the fact that geostationary satellites see the earth's surface at an increasingly unfavourable angle. This limitation is not shared by the sun-synchronous polar orbiting satellites, but the retrieval of solar irradiance from these satellites (Laine et al., 1999) is hampered by their incomplete temporal coverage. Zelenka et al. (1999) report that estimates of hourly global irradiance based on geostationary satellite data, reproduce ground truth values with RMSDs of typically 2-25%, comparable to the RMSD seen between ground stations some 2-3 km apart. Moreover, they estimate that the RMSD traceable to shortcomings of the satellite-to-irradiance conversion models is likely to be of the order of 12%, while the remaining RMSD is attributed to measurement errors of the surface instruments and, more importantly, to the micro-variability of the irradiation field. The www server established by the SATEL-LIGHT project (Fontoynont et al., 1998; offers access to solar radiation data all over Europe, presently for the years 1996 and These data, derived from METEOSAT data and a number of auxiliary models, are highly resolved both spatially and temporally. Moreover, the data are specifically designed to meet the data requirements of the daylighting and solar energy utilization communities. Through comparison with ground truth observations from seven Norwegian stations and from five International Daylight Measurement Program stations (IDMP, the present paper highlights the accuracy of several steps in the following process by which daylight data are produced at the SATEL-LIGHT server: 1. Half hourly global irradiance is derived from METEOSAT data (Cano et al., 1986) and a cloud-free global radiation model (Kasten 1996, Page 1996, Dumortier 1995, 1998). 2

3 2. Horizontal diffuse and beam irradiances are next calculated by a diffuse fraction model (Skartveit et al., 1998). 3. Normal incidence beam irradiance is then readily calculated, and sunshine duration is obtained from the number of half hours with normal incidence beam exceeding 12 Wm Horizontal illuminances are obtained from horizontal diffuse and beam irradiances by a luminous efficacy model (Skartveit and Olseth, 1989). 5. Slope illuminances are finally obtained by a slope / horizontal ratio model (Skartveit and Olseth, 1986). The alternative route, namely to derive the secondary radiation components (2-5) directly from the satellite data and not indirectly from satellite global irradiances, is evaluated by Ineichen and Perez (1999). 2. Theory 2.1 The Heliosat procedure Of the solar irradiance impinging on top of atmosphere (TOA), one highly variable fraction is scattered back to space, a second highly variable fraction is absorbed at the earth s surface, while the remaining smaller and less variable fraction is absorbed in the atmosphere. This fact yields a nearly linear relationship between the former two fractions, which for a fixed surface albedo implies a nearly linear relationship even between atmospheric tramsmittance and TOA planetary albedo. This latter relationship provides the basis of most methodologies to retrieve solar irradiance at ground level from satellite images, since reflected solar radiance observed from satellites is closely related to TOA planetary albedo. The Heliosat procedure, originally proposed by Cano et al. (1986) and recently modified by Beyer et al. (1996) and by the SATEL-LIGHT project (Iehlé et al., 1997; Fontoynont et al., 1998), yields surface global irradiance from pixel counts in the VIS-channel ( :m) of the geostationary satellite METEOSAT. As described by Hammer et al. (1998), these pixel counts are normalised for instrument offset, backscatter from the cloud-free atmosphere, solar zenith angle, and Sun-Earth distance to yield an apparent albedo D for 3 X 5 pixels centred around each of our ground truth stations. These 15 pixels correspond to an area about 5 km in longitude and from 6 km (34 N) to 16 km (64 N) in latitude. A cloud index n is subsequently defined: n = ( r - r ) / ( r c - r ), ( 1 ) providing a measure of cloud cover. The reference values ρ and D c are extracted from a series of images, and refer to the apparent albedo of cloud-free pixels and that of a compact cloud cover, respectively. The overcast reference D c is taken to be the 96 th percentile point of the distribution of normalised counts (Hammer et al., 21). The cloud-free reference ρ has a daily and seasonal trend due to changes in the direction of incident sunlight and surface characteristics, and it is derived from modal values in distributions of normalised counts (Hammer, 2). Note that a snow cover makes D increase by an amount which may vary strongly both in space and time. These cloud indices are therefore a particularly inaccurate measure of cloudiness in case of snow cover. The surface global irradiance is subsequently obtained from the following relation between global clear sky index k g and cloud index n, found empirically from data prior to 1996 by the SATEL-LIGHT team (Fontoynont et al., 1998; 3

4 k g = 1.2, for n -.2, k g = 1 - n, for -.2 n.8, k g = ( n + 25 n 2 ) / 15, for.8 n 1.1, k g =.5, for n 1.1. ( 2a ) ( 2b ) ( 2c ) ( 2d ) The global clear sky index k g is here defined as the ratio between the actual global irradiance H g (observed or satellite-derived) and an average cloud-free global irradiance H : k g = H g / H = H g / ( H b + H d ). ( 3 ) H is the sum of cloud-free beam irradiance H b (Kasten 1996, Page 1996) and cloud-free diffuse irradiance H d (Dumortier 1995), both for Linke turbidity factor appropriate for the station in question (Dumortier 1998). The Linke turbidity factor is a simplified measure of the variable water vapour and aerosol extinction. It is defined as the number of dry and aerosol-free atmospheres which are required to produce the same beam extinction, at relative optical air mass = 2., as does the cloud-free atmosphere in question. 2.2 The diffuse fraction model The hourly diffuse fraction model (Skartveit et al., 1998) requires the following input parameters: solar elevation, regional surface albedo, clearness index (surface global irradiance / extraterrestrial global irradiance), and an index quantifying the hour-to-hour variability of the clear sky index k g (3). Moreover, to prevent excessively high normal incidence beam irradiances at very low solar elevation, the model does not allow a solar elevation dependent maximum beam transmittance to be exceeded. This model was tuned to 32 years of data from Bergen and successfully tested against independent data from the IDMP stations Garston, Gävle, Lisbon, and Lyon (Skartveit et al., 1998), and it is in the present paper applied unmodified even for half hourly and 1 min data. 2.3 The luminous efficacy model Given any irradiance spectrum, the corresponding illuminance is obtained by multiplying this spectrum by the CIE (Commission Internationale de l Éclairage) curve for photopic vision. The luminous efficacy is then defined as the ratio between illuminance and irradiance, both integrated over the entire solar spectrum. The luminous efficacy model of Olseth and Skartveit (1989) is based on spectral irradiances obtained by an interpolation between transmittance models for, respectively, cloud-free sky (Bird and Riordan, 1986) and unbroken cloud cover (Stephens et al, 1984). This interpolation decomposes the diffuse sky irradiance into "blue sky", "dark cloud", and "bright cloud" irradiance. For partly cloudy cases, the model was slightly tuned to hourly global illuminance and irradiance data from Bergen ( ). The parameterized version of the model requires solar elevation, day of year, and diffuse and beam clearness indices as input. In the case of diffuse irradiance, the model is later on tuned to data from Albany, NY (gratefully received from R. Perez) by multiplying the difference between "dark cloud" efficacy and extraterrestrial efficacy by a factor.7 (Skartveit and Olseth, 1994). Moreover, in the case of beam irradiance, the model is slightly modified to explicitly account for variation in column amount of water vapour, under the assumption that water vapour extinction takes place solely at non-visible wavelengths (Olseth and Skartveit, 1997). 2.4 The slope model Given horizontal beam irradiance/illuminance, the beam irradiance/illuminance on any slope is readily computed. To calculate the diffuse slope irradiance/illuminance requires, however, additional information about surface (foreground) reflectance and the horizontal diffuse sky irradiance/illuminance and its angular dis- 4

5 tribution. Assuming (at least qualitatively correct) that the angular distribution of the diffuse illuminance is the same as in the case of diffuse solar irradiance, we apply our slope model (Skartveit and Olseth, 1986) for diffuse irradiance even for diffuse illuminance. This model assumes Lambertian ground reflectance, while sky radiance anisotropy is parameterized as follows: One fraction, equal to the beam transmittance, of the horizontal diffuse irradiance is treated as circumsolar radiation (Hay, 1979). Another fraction, decreasing from.3 at overcast to zero at beam transmittance =.15, is treated as collimated radiation from zenith. The remaining horizontal diffuse irradiance is treated as isotropic sky radiance. 3. Data Sources 3.1 METEOSAT/Heliosat data The Heliosat data for the years were downloaded from the SATEL-LIGHT www server ( where even the derivation of these data is outlined. Heliosat data prior to 1996 were gratefully received from A. Hammer at Carl von Ossietzky University, Oldenburg, Germany. 3.2 Ground truth data Ground truth data were kindly made available (see Acknowledgements) from five European IDMP stations and from seven Norwegian stations (Fig. 1, Tab. 1). The time resolution of these data, in the following referred to as individual observations, differ somewhat between the 12 stations. Thus, 1 min averages are used for Geneva, half-hourly averages are used for Lyon and Garston, while hourly averages are used for the remaining 9 stations. Kipp&Zonen CM11 pyranometers were used for global and diffuse irradiance, except for Geneva (CM1), and for Lisbon and Lyon (CM6). Diffuse irradiance at Bergen was measured with a 3cm radius circular disk at 3 cm distance. Eppley NIP (Normal Incidence Pyrheliometer) was here used to measure normal incidence beam irradiance each 2s, and ground truth sunshine duration is found from the number of NIP readings > 12 Wm -2. At the six remaining Norwegian stations, sunshine duration was recorded by Solar 111 sunshine recorders made by Haenni, Switzerland. At Apelsvoll, however, the recorder was incorrectly orientated in the azimuth direction, and sunshine duration data from this station are therefore discarded. The sensors that measured illuminances on vertical surfaces at Geneva, were screened off from ground reflected radiation by a black screen (honeycomb). Further details about shading devices and illuminance sensors at the IDMP stations are found at 4. Results/Verification 4.1 The luminous efficacy model Since our original luminous efficacy model (Skartveit and Olseth, 1989) has been sligthly modified, it is in the present paper tested against independent data from two Research class (Garston, Geneva) and three General class (Gävle, Lisbon, Lyon) European IDMP stations. For these stations, the luminous efficacy model is run with climatological average monthly water vapour amounts, derived from surface dew-point temperatures (WMO, 1982). Fig. 2 shows distributions of observed hourly illuminances and luminous efficacies at the two Research class stations, along with corresponding distributions of deviations (modelled - observed). These individual deviations cover a quite narrow range in the case of illuminances, while the corresponding range is more significant in the case of luminous efficacy. Fig. 3 shows, for all the five 5

6 IDMP stations, median values of observed illuminances and luminous efficacies. The median illuminance deviations are small compared both to the differences between the five stations and to the differences between global, diffuse, and beam illuminances (Fig. 3, left). The same applies to a somewhat lesser extent to luminous efficacies, except for some differences between stations with respect to beam luminous efficacy (Fig. 3, right). For both illuminances and luminous efficacies, the smallest deviations are seen at the two Research class stations Garston and Geneva. It is also seen that the deviation in global luminous efficacy varies less than does the deviation in both diffuse and beam luminous efficacies, both at each station separately (Fig. 2) and between the five stations (Fig. 3, right). 4.3 The slope model Since our slope model (Skartveit and Olseth, 1986) has not been tested against illuminance data, it is in the present paper tested, in concert with the above luminous efficacy model, against independent data from the Research class IDMP station Geneva. Model calculations, with ground truth global and horizontal diffuse irradiances as input and vertical illuminances as output, are performed with two different values of the Lambertian albedo of the horizontal foreground, viz. A =. and.1. It is seen (Fig. 4) that the increase of A from. to.1 slightly narrows the range of deviation (modelled - observed), and yields median deviations close to zero on all the four verticals. Fig. 4 thus strongly indicates that the horizontal foreground (honeycomb) of the vertical sensors is closer to having A =.1 than to being completely black (A =.). In fact, this is the same conclusion as that drawn from one year of data from the Swedish General class IDMP station Gävle (Skartveit and Olseth, 1996). We therefore apply foreground Lambertian albedo A =.1 for slope calculations in the following. 4.4 Hourly global and diffuse irradiance Observed hourly global and diffuse irradiances at Bergen ( ) are plotted against their METEOSAT (Heliosat) counterparts in Fig. 5, where the METEOSAT hourly averages are obtained by adequate weighting of half-hourly values. For global irradiance, the METEOSAT versus ground truth MBDs are only some 2-3 Wm -2, which both overall and for solar elevation > 15 amount to less than 1 % of the average global irradiance (Table 2). This overall percentage MBD is somewhat lower than the -3.6 to 2.5% reported by Ineichen and Perez (1999) for Albany, New York and for Geneva and Lausanne, Switzerland. The RMSDs at Bergen are, however, 22% of average global irradiance for solar elevation >3, 27% for solar elevation 15-3, and 38% for solar elevation <15, making an overall percentage RMSD of 25%. This overall percentage RMSD is somewhat lower than the 26-33% reported by Ineichen and Perez (1999), and it conforms reasonably well with overall RMSDs of 2-25% claimed by Zelenka et al. (1999) to be typical for hourly global irradiance estimates derived from satellite "visible"-channel data. Zelenka et al. compared these typical RMSDs to RMSDs between neighbouring ground measurements, and found that the 2-25% RMSD level is reached within only 2-3 km distance, and that RMSD reached some 17% already at 1 km distance. That is, for any application requiring time/site specific hourly data, the user should rely on satellite rather than on a «neighbouring» ground station if the latter operates further away than 2-3 km from the site. The application of our diffuse fraction model on METEOSAT global irradiances, yields diffuse irradiances with MBDs equal to some 8-2 Wm -2, relative to their ground truth counterparts (Table 2). Within the three solar elevation intervals, these MBDs amount to some 5-1% of average global irradiance. This makes an overall percentage MBD of 6%, somewhat higher than the 1-2% reported for Geneva and Albany by Ineichen and Perez (1999). The corresponding RMSDs at Bergen are some 15-27% of average global irradiance, making an overall percentage RMSD of 17% which is slightly lower than the 17-19% reported by Ineichen and Perez. Thus, the present calculation of diffuse irradiance from satellite-derived global irradiance at Bergen (overall diffuse fraction =.48) yields somewhat higher MBD and slightly lower RMSD than does the derivation of diffuse irradiance directly from satellite counts at Geneva and Albany (overall diffuse fractions = ). Note also that, in spite of a remarkably low MBD in global irradiance at Bergen, the major part of the MBD in diffuse irradiance nevertheless derives from 6

7 deviations in METEOSAT global irradiance. This is inferred from the fact that the diffuse irradiance MBDs are reduced from 5-1% to 1-2% of average global irradiance by running the diffuse fraction model directly on ground truth global irradiance (Table 2, Fig. 6c). The above hourly data are even compared by plotting accumulated duration curves. These curves show only moderate differences between the frequency distributions of, respectively, ground truth and METEOSAT derived hourly global irradiance (Fig. 6a). This frequency distribution conformity, between ground truth and satellite-derived data, makes the satellite-derived data fulfill the accuracy required by many data users. However, the curves also quantify the tendency, already seen from Fig. 5, that Heliosat slightly overestimates ground truth global irradiance at Bergen for low irradiance and slightly underestimates it for high irradiance. Corresponding curves for diffuse irradiance reveal the same tendency (Fig. 6b). 4.5 Spatial variation of average global irradiance For the years 1996 and 1997, annual global irradiations derived by the Heliosat procedure are compared to ground truth data from the 7 Norwegian stations (stations 1-7, Fig. 1). The Linke turbidity factor, used in the Heliosat procedure, was allowed to vary with calendar month, elevation, and geographical position according to the SATEL-LIGHT model of turbidity variations in Europe (Dumortier, 1998). Reasonably well in accordance with the findings of Ineichen and Perez (1999), we find that the annual MBD, at all individual stations, is less than 3% of the overall ground truth average, or less than 1% of the range of individual ground truth averages (Fig. 7). This applies in fact both to stations east of the Scandinavian mountain chain (Zone 2, range of sea level monthly SATEL-LIGHT turbidities ) and to stations west of this mountain chain (Zone 1, range of sea level monthly SATEL-LIGHT turbidities ). 4.6 Sunshine duration In Fig. 8a-b), ground truth values of monthly relative sunshine duration (observed duration / possible duration, both for actual horizon) are plotted against their METEOSAT counterparts (Heliosat duration / possible duration, both for unobstructed horizon), at the Norwegian stations. During those months when the possible duration exceeds 75% of the astronomical duration (possible duration for unobstructed horizon), a reasonable agreement is seen for monthly ground truth relative sunshine duration ranging from 1 to 9%. Moreover, apart from a slight spring versus autumn asymmetry, the METEOSAT minus ground truth deviation shows no significant seasonal trend (Fig. 8c). It thus appears that METEOSAT sunshine duration, along with knowledge (from topography) of possible duration, yield trustworthy sunshine duration data. Some more scatter is seen, however, for those midwinter months when horizon screening makes the possible duration less than 75% of the astronomical duration (Fig. 8a). In such months, during which the METEOSAT duration tends to be lower than ground truth, noon solar elevation is low and intermittent snow cover occurs in the station surroundings. The underestimation may thus be due partly to limited accuracy because of the low solar elevations, and partly to intermittent snow cover being interpreted as cloud cover by the Heliosat procedure. Moreover, particularly important during these winter months is the fact that the METEOSAT relative duration refers to the part of the day when the sun is above an unobstructed horizon, while the ground truth data refer to the part of the day when the sun is above the actual horizon. The actual horizon preferentially blocks off the sun at low solar elevation and thus contributes to higher average solar elevation in the ground truth case than in the METEOSAT case. Under a fixed atmosphere, one should therefore expect, as in fact observed, a higher relative sunshine duration (lower beam extinction) in the ground truth case than in the METEOSAT case. 4.7 Modelled versus observed illuminance 7

8 Starting from half-hourly global irradiances, derived by the Heliosat procedure from METEOSAT data, we applied our diffuse fraction model, luminous efficacy model and slope model to finally obtain illuminances on horizontal and vertical planes. Similarly, these illuminances were also modelled from ground truth global and diffuse irradiances by our luminous efficacy and slope models. These two sets of modelled illuminances were subsequently compared to corresponding ground truth illuminances. The various illuminances (global, diffuse, beam, S9, W9, E9, N9) are modelled from ground truth global and diffuse irradiances with minor MBD and moderate RMSD throughout the year at Geneva (Fig. 9a-b). The high RMSD on S9 during December and January is, however, an exception (see below). Moreover, from February to November, the MBDs stay minor even if the model input is changed from ground truth global and diffuse irradiances (Fig. 9a) to METEOSAT global irradiance (Fig. 9c). This change of input yields, on the other hand, a quite significant RMSD increase. But, similarly to the case shown in Fig. 6, it is found (not shown here) that the frequency distributions of METEOSAT half-hourly illuminances differ only moderately from the distribution of ground truth values, in spite of the quite significant RMSD between the two. The METEOSAT derived illuminances for December and January are, however, reminders of the particular difficulties encountered at low solar elevation. As mentioned above, only minor MBDs are found between ground truth illuminances and illuminances modelled from ground truth global and diffuse irradiance (Fig. 9b), corroborating the adequacy of the luminous efficacy model and slope model even during these midwinter months. The Heliosat procedure, however, overestimates (Fig. 9d) the average global illuminance by 3. klux (23% of ground truth). The diffuse fraction model adequately turns this excess global illuminance into a minor excess of.9 klux (9%) in average horizontal diffuse illuminance and into a more significant excess of 2.1 klux (68%) in average horizontal beam illuminance. At these low winter solar elevations, this beam excess propagates via the slope model and yields a quite significant MBD of 8.3 klux (55%) for illuminance on the south facing vertical. 5. Summary and concluding remarks Hourly recordings of sunshine duration and global irradiance at Bergen, Norway, show negligible MBDs compared to data derived from METEOSAT by the Heliosat procedure. There is, however, a slight tendency that METEOSAT derived global irradiance exceeds ground truth under overcast sky while the opposite is true under approximately cloud-free sky. These minor deviations in global irradiance propagate, via a diffuse fraction model, into the METEOSAT diffuse irradiance, which at any solar elevation shows MBDs relative to ground truth amounting to some 5-1% of the average observed global irradiance. The RMSD (relative to ground truth) of satellite-derived hourly global irradiance at Bergen (6.4 N) conforms reasonably well with the RMSDs of 2-25% typically reported world-wide in current literature. This RMSD level is similar to that seen between ground truth stations some 2-3 km apart. The spatial variation of annual mean global irradiance at seven high latitude (58-64 N) stations is nicely reproduced by the METEOSAT data, with overall MBDs at all individual stations being less than 3% of the overall ground truth average, or less than 1% of the range of individual ground truth averages. Even monthly relative sunshine duration is nicely reproduced, provided horizon screening does not make the monthly possible sunshine duration less than 75% of the astronomical duration. A diffuse fraction model, a luminous efficacy model and a slope model is successfully tested against independent data. Horizontal and vertical illuminances, derived for Geneva by these models from METEOSAT global irradiances, show MBDs relative to ground truth well within a range of ±4% of observed average global illuminance during the 1 months February to November. However, the RMSDs increase quite significantly when illuminances are derived from METEOSAT global irradiance rather than from ground truth global and diffuse irradiances. But the frequency distribution of illuminances derived from 8

9 METEOSAT differs only moderately from that of ground truth values, and this frequency distribution conformity alone makes the satellite-derived data fulfil the demands of many SATEL-LIGHT data users. During December and January, however, moderate errors in METEOSAT horizontal beam irradiances yield even quite significant MBD on the south facing vertical surface. Acknowledgements This work is a part of the project SATEL-LIGHT in the JOULE Programme JOR3-CT9541, funded by the European Commission. We thank our project colleagues for valuable advice. In the SATEL-LIGHT project J.A. Olseth was also affiliated at The Norwegian Meteorological Institute. Global radiation data is made available from 6 Norwegian agrometeorlogical stations by Planteforsk, Aas. Data from the IDMP stations at Lisbon, Lyon, and Garston were received from D. Dumortier at ENTPE, Lyon. Data from the Swiss Research class IDMP station at Geneva were recived from P. Ineichen at GAP-E, University of Geneva. Data from the Swedish General class IDMP station Gävle were received from H. A. Löfberg at the Royal Institute of Technology, Department of Built Environment. 6. References Beyer, H.G., Costanzo, C., Heinemann, D., 1996: Modifications of the Heliosat procedure for irradiance estimates from satellite images. Solar Energy 56, Bird, R.E., Riordan, C., 1986: Simple solar spectral model for direct and diffuse irradiance on horizontal and tilted planes at the earth s surface for cloudless atmospheres. J. Climate Appl. Meteor. 25, Cano, D., Monget, J.M., Albuisson, M., Guillard, H., Regas, N., Wald, L., 1986: A method for the determination of the global solar radiation from meteorological satellite data. Solar Energy 37, Dumortier, D., 1995: Mesure, Analyse et ModJlisation du gisement lumineuz Application B l'jvaluation des performances de l'jclairage naturel des b>timents. Thesis, Laboratoire GJnie Civil et Habitat, ENTPE, Lyon, France. Dumortier, D., 1998: The Satellight model of turbidity variations in Europe. Report for the sixth SATELLIGHT meeting in Freiburg, Germany, September Fontoynont, M., Dumortier, D., Heinemann, D., Hammer, A., Olseth, J. A., Skartveit, A., Ineichen, P., Reise, C., Page, J., Roche. L., Beyer, H. G., Wald, L., 1998: SATELLIGHT: A www server which provides high quality daylight and solar radiation data for western and central Europe. Proc. 9th Conference on Satellite Meteorology and Oceanography in Paris, May 25-28, 1998, p Hammer, A., Heinemann, D., Westerhellweg, A., Ineichen, P., Olseth, J. A., Skartveit, A., Dumortier, D., Fontoynont, M., Wald, L., Beyer, H. G., Reise, C., Roche. L, Page, J., 1998: Derivation of daylight and solar irradiance data from satellite observations. Proc. 9th Conference on Satellite Meteorology and Oceanography in Paris, May 25-28, 1998, p Hammer, A., 2: Anwendungsspezifische Solarstrahlungsinformationen aus Meteosat-Daten. PhD Thesis, Department of Physics, Carl von Ossietzky University, Oldenburg, Germany. Hammer, A., Heinemann, D., Hoyer, C., 21: Effect of Meteosat VIS sensor properties on cloud reflectivity. Third SoDa meeting, Bern, Switzerland, 17 th - 19 th January 21. Available from Department of Physics, Carl von Ossietzky University, Oldenburg, Germany. Hay, J. E., 1979: Study of short-wave radiation on non-horizontal surfaces. Report No 79-12, Atmos- 9

10 pheric Environment Service, Downsview, Ontario. Iehlé, A., Bauer, O., Wald, L., 1997: Final Report of ARMINES/ENSMP to the University of Oldenburg for the SATEL-LIGHT programme. Groupe télédétection et modélisation, ENSMP, Sophia-Antipolis; France. Ineichen. P., Perez, R., 1999: Derivation of Cloud Index from Geostationary Satellites and Applicaion to the Production of Solar Irradiance and Daylight Illuminance Data. Theoretical and Applied Climatology 64,1/2, Kasten, F., 1996: The Linke turbidity factor based on improved values of the integral Rayleigh optical thickness. Solar Energy 56, Laine, V., Venäläinen, A., Heikinheimo, M., Hyvärinen, O., 1999: Estimation of Surface Solar Global Radiation from NOAA AVHRR Data in High Latitudes. J. Appl. Meteor. 38, 12, Olseth, J.A., Skartveit, A., 1989: Observed and modelled luminous efficacies under arbitrary cloudiness. Solar Energy 42, Olseth, J.A., Skartveit, A., 1997: Spatial distribution of photosynthetically active radiation over complex topography. Agric. Forest Meteorol. 86 (1977) Page, J., 1996: Algorithms for the Satellight programme. Technical Report for the second SATELLIGHT meeting in Bergen, Norway, June Skartveit, A., Olseth, J. A., 1986: Modelling slope irradiance at high latitudes. Solar Energy 36, Skartveit, A., Olseth, J. A., 1994: Luminous efficacy models and their application for calculation of photosynthetic active radiation, Solar Energy 52, Skartveit, A., Olseth, J. A., 1996: Illuminance/irradiance at a high latitude IDMP station. Second SATELLIGHT meeting, Bergen, June 24-25, 1996 ( Skartveit, A., Olseth, J. A., Tuft, M. E., 1998: An hourly diffuse fraction model with correction for variability and surface albedo. Solar Energy 63, Stephens, G. L., Ackerman, S., Smith, E. A., 1984: A short-wave parameterization revised to improve cloud absorption. J. Atmos. Sci. 41, World Meteorological Organization (WMO), 1982: Climatological normals (CLINO) for climate and climate ship stations for the period WMO / OMM - No 117. Zelenka, A., Perez, R., Seals, R., Renné, D., 1999: Effective Accuracy of Satellite-Derived Hourly Irradiances. Theor. Appl. Climatol. 62, (1999). 1

11 Fig. 1 Map with geographical position of 5 IDMP stations (A - E) and 7 Norwegian stations (1-7). 11

12 ACCUMULATED FREQUENCY (%) Global (MOD - OBS) Geneva Diffuse Beam ACCUMULATED FREQUENCY (%) Geneva EFFICACY (lumen/w) Global Diffuse (MOD - OBS) EFFICACY (lumen/w) Beam -3 3 ACCUMULATED FREQUENCY (%) Global (MOD - OBS) Garston Diffuse Beam ACCUMULATED FREQUENCY (%) Garston EFFICACY (lumen/w) Global Diffuse (MOD - OBS) EFFICACY (lumen/w) Beam -3 3 Fig. 2 Left: Accumulated frequency distributions of observed global, diffuse and beam illuminance together with accumulated distributions of deviations (modelled - observed illuminances) for Garston (halfhourly values) and Geneva (1 min values) for solar elevation > 1. Illuminances are here modelled by our luminous efficacy model from observed global and diffuse irradiance, with observed beam data obtained as global - diffuse. Note that the horizontal axis is expanded in the deviation plot. Right: Corresponding distributions of luminous efficacies. Beam luminous efficacies are included only if normal incidence beam irradiance > 12 Wm

13 MOD - OBS (klux) Lisbon Lyon Geneva Garston Gävle MOD - OBS (lumen/w) Lisbon Lyon Geneva Garston Gävle Diffuse Beam Global EFFICACY (lumen/w) Diffuse Global Beam Fig. 3 Bottom: Median values of observed individual illuminances (left) and luminous efficacies (right) at 5 European IDMP stations. Top: Corresponding median deviations (modelled - observed) of illuminances and luminous efficacies. Illuminances are modelled in the same way as in Fig. 2, and beam luminous efficacies are again included only if normal incidence beam irradiance > 12 Wm -2. Note that the vertical axis is expanded in the median illuminance deviation plot. 13

14 ACCUMULATED FREQUENCY (%) Geneva N9 (MOD - OBS) -5 5 A = A =.1 ACCUMULATED FREQUENCY (%) Geneva E9 (MOD - OBS) -5 5 A = A =.1 ACCUMULATED FREQUENCY (%) Geneva S9 (MOD - OBS) -5 5 A = A =.1 ACCUMULATED FREQUENCY (%) Geneva W9 (MOD - OBS) -5 5 A = A =.1 Fig. 4 Accumulated frequency distributions of observed illuminances on north (N9), east (E9), south (S9), and west (W9) verticals, along with corresponding distributions of deviations (modelled - observed illuminances). 1 min average data (solar elevation > 1 ) from Geneva plotted for two values of Lambertian foreground albedo A (see text). Illuminances are here modelled by our luminous efficacy model and slope model from observed global and diffuse irradiance, with observed beam data obtained as global - diffuse. Note that the horizontal axis is expanded in the deviation plots. 14

15 HOURLY GLOBAL IRRADIANCE (W m-2) (METEOSAT) h > 3 N = HOURLY GLOBAL IRRADIANCE (W m-2) (SURFACE OBS.) HOURLY DIFFUSE IRRADIANCE (W m-2) (METEOSAT) h > 3 N = HOURLY DIFFUSE IRRADIANCE (W m-2) (SURFACE OBS.) HOURLY GLOBAL IRRADIANCE (W m-2) (METEOSAT) < h < 3 N = HOURLY GLOBAL IRRADIANCE (W m-2) (SURFACE OBS.) HOURLY DIFFUSE IRRADIANCE (W m-2) (METEOSAT) < h < 3 N = HOURLY DIFFUSE IRRADIANCE (W m-2) (SURFACE OBS.) HOURLY GLOBAL IRRADIANCE (W m-2) (METEOSAT) h < 15 N = HOURLY GLOBAL IRRADIANCE (W m-2) (SURFACE OBS.) HOURLY DIFFUSE IRRADIANCE (W m-2) (METEOSAT) h < 15 N = HOURLY DIFFUSE IRRADIANCE (W m-2) (SURFACE OBS.) Fig. 5 Global (left column) and diffuse (right column) hourly irradiances derived by the Heliosat procedure from METEOSAT data, plotted versus their ground truth counterparts within three solar elevation (h) intervals at Bergen during The 1:1 line is indicated, and only hours during which the sun is above the natural horizon during the entire hour are plotted, with number of hours printed as N and average values printed on the respective axes. 15

16 PERCENTAGE OF LEVEL EXCEEDED (%) PERCENTAGE OF LEVEL EXCEEDED (%) PERCENTAGE OF LEVEL EXCEEDED (%) h < < h < 3 h > HOURLY GLOBAL IRRADIANCE (W m-2) 1 9 b) h < < h < 3 h > HOURLY DIFFUSE IRRADIANCE (W m-2) h < < h < 3 h > 3 a) c) HOURLY DIFFUSE IRRADIANCE (W m-2) Fig. 6 The same data as in Fig. 5 plotted as accumulated duration curves for global and diffuse irradiance within three solar elevation (h) intervals. Fully drawn curves show ground truth data. The broken curves in a) and b) show METEOSAT derived data, while those in c) show diffuse irradiances obtained by running our diffuse fraction model on ground truth global irradiance (not on METEOSAT global as in b)). 16

17 GLOBAL IRRADIANCE (W m-2) (METEOSAT) Zone 2 Zone GLOBAL IRRADIANCE (W m-2) (Surface observed) 1:1 Fig. 7 Annual mean values of ground truth versus METEOSAT derived (Heliosat) global irradiance for each of the 7 Norwegian stations ( ), for Linke turbidity factors according to Dumortier (1988). A distinction is made between two geographical zones (Zone 1: stations 1-4 in Fig. 1; Zone 2: stations 5-7 and E in Fig. 1). 17

18 RELATIVE SUNSHINE DURATION (%) (METEOSAT) a) Bergen 1:1 Mar. - Oct. Nov. - Feb. RELATIVE SUNSHINE DURATION (%) (METEOSAT) b) 1:1 5 stations RELATIVE SUNSHINE DURATION (%) (Surface observed) RELATIVE SUNSHINE DURATION (%) (Surface observed) RELATIVE SUNSHINE DURATION (%) (METEOSAT - SURFACE OBSERVED) c) Max (5) Median (5) Average (Bergen) Min (5) MONTH Fig. 8 a) Ground truth versus METEOSAT derived (Heliosat) values of monthly relative sunshine duration at Bergen ( ). Open squares indicate months when the possible duration is less than 75% of the astronomical duration, while crosses indicate the remaining months. b) Ground truth versus METEOSAT derived (Heliosat) values of monthly relative sunshine duration at 5 Norwegian stations ( ) equipped with Solar 111 recorders. c) The distribution, for each calendar month, of monthly METEOSAT minus ground truth differences. These distributions are plotted as minimum, median and maximum difference at the five Solar 111 stations, and as average difference at Bergen. Only months when the possible duration is less than 75% of the astronomical duration are included in b) and c). 18

19 MODELLED FROM SURFACE GLOBAL & DIFFUSE (klux) N9 B E9/W9 S9 D G a) MODELLED FROM SURFACE GLOBAL & DIFFUSE (klux) N9 B E9/W9 D G S9 b) OBSERVED (klux) OBSERVED (klux) 5 25 MODELLED FROM METEOSAT GLOBAL (klux) c) MODELLED FROM METEOSAT GLOBAL (klux) d) OBSERVED (klux) OBSERVED (klux) Fig. 9 Average modelled illuminances at Geneva plotted against their surface observed counterparts, and with corresponding RMSDs drawn as error bars (note that the S9 error bar in d) is drawn downwards only, to save space). The figure is based on 1 year of 1 min illuminances/irradiances on (see text along figure top) a horizontal surface (Global, Diffuse, and Beam) and on 4 verticals (S9, W9, E9, and N9). Separate plots are made for December + January (right) and for the rest of the year (left). Moreover, illuminances are modelled from ground truth global and diffuse irradiance (top) and from METEOSAT derived (Heliosat) global irradiance (bottom). 19