Bibliography

Background and objective

Early morning forecast

Satellite-based forecast principles

Several solar irradiance forecasting methods have been established using geostationary meteorological satellite visible images and generally present better results than NWP [1]. Reuniwatt’s HourCast method is described by the following chart. Raw images are converted into cloud index using an algorithm based on the Heliosat-2 method [2]. Then, a cloud motion vector (CMV) field is computed from two subsequent cloud index images, using an optical flow algorithm [3]. This CMV field is applied on the current image to extrapolate cloud index evolutions up to 6 hours. Extrapolated cloud index images are converted into irradiance using a clear sky model.

Sylvain Cros, Mathieu Turpin, Florian Le GuillouReuniwatt, 14 rue de la Guadeloupe, 97490 Sainte-Clotilde, Reunion Island, France

Corresponding author: sylvain.cros@reuniwatt.com

[1] Lorenz E., Kühnert J., Heinemann D., Nielsen K. P., Remund J., Müller S.C., Cros S., Comparison of irradiance forecasts based on nume-rical weather prediction models with different spatio-temporal resolutions, In Proceedings of the 30th European PV Solar Energy Confe-rence, 14-18 September 2015, Hamburg, Germany.

[2] Rigollier, C., Lefèvre, M., & Wald, L. (2004). The method Heliosat-2 for deriving shortwave solar radiation from satellite images. Solar Energy, 77(2), 159-169.

[3] Cros S., Sébastien N., Liandrat O., Schmutz N., Cloud pattern prediction from geostationary meteorological satellite images for solar en-ergy forecasting, Proc. SPIE 9242, Remote Sensing of Clouds and the Atmosphere XIX; and Optics in Atmospheric Propagation and Adap-tive Systems XVII, 924202 (October 21, 2014); doi:10.1117/12.2066853.

[4] Hammer, A., Kühnert, J., Weinreich, K., & Lorenz, E. (2015). Short-term forecasting of surface solar irradiance based on Meteosat-SEVIRI data using a nighttime cloud index. Remote Sensing, 7(7), 9070-9090.

[5] Richter, F. and Dutton E.G. (2013) The Baseline Surface Radiation Network and its World Radiation Monitoring Centre at the Alfred We-genKönig-Langlo, G. , Sieger, R. , Schmithüsen, H. , Bücker, A. , Richter, F. and Dutton E.G. (2013) The Baseline Surface Radiation Network and its World Radiation Monitoring Centre at the Alfred Wegener Institute. GCOS - 174, WCRP Report 24/2013, 30 pp.

Providing intraday solar irradiance forecasts before sunrise: satellite-based night cloud index retrieval and forecast

HourCast’s earliest forecast on a given punctual site can only be done if a large surrounding area is sufficiently illuminated by the sun. Concretely, the optical flow analysis is accurate only if the solar zenith angle (SZA) is smaller than 85° on a square of at least 1000*1000 km² centred on the given site. In some cases, country-scale forecasts cannot be delivered before one hour after sunrise. A solution to overcome this problem consists in building a nighttime cloud index to ensure an earlier delivery of HourCast. This solution has been proposed by [4]. We implemented it in our forecasting scheme.

• Intraday solar irradiance forecasts are necessary to:• manage grid with photovoltaic power penetration (optimizing storage and reserve power);• adjust the previous day’s unbalanced bids on the Day-Ahead electricity markets.

• They are often needed early in the morning, even before sunrise.• Irradiance time-series modelling or satellite-based forecasts are generally more accurate than NWP models for short-term (up to several hours) forecasting.• However, these techniques require observations in the daytime and cannot be delivered before sunrise. • We implemented and assessed a satellite-based solar irradiance forecasting method usable in the nighttime using the Meteosat-10 satellite over Europe and the Meteosat-8

satellite over the Indian Ocean.

Night cloud index computationThe principle of the night cloud index proposed by [4] consists in using infrared channels at 10.8 and 3.9 µm. Brightness temperature differences (BTD) between these channels, corrected by the satellite viewing angle permit to reproduce cloud patterns including lower warm clouds. The 10.8 µm channel is also used to detect very cold clouds (T10.8 < 232 K). Considering that the proportion of clear and cloudy pixels is stable at a continental scale between subsequent night and day times, [4] statistically computed bijective functions attributing a cloud index value for each BTD or T10.8 value for three different cloud classes.

Forecast implementation and assessmentWe assessed this method by implementing HourCast using recommendations from [4]: • a night cloud index when SZA > 89.8°• a linear weighted combination between night and day cloud indices when 85° < SZA < 89.8°• a day cloud index when SZA < 85° We computed forecasts on three punctual sites and compared the results with corresponding ground measurements as summarized in the table on the right-hand side.

Results

• Scatterplots of GHI values show a linear agreement between 3-hour time horizon forecasts delivered at nighttime.

• However there is a strong underestimation for sites located in the Indian Ocean area.• Forecasts using a combined cloud index are likely affected by the optical flow analysis in

the twilight zone which can induce strong cloud pattern extrapolation errors.

Conclusion

• Nighttime assessments using early ground measurements show a reasonable accuracy confirming the interest of the satellite for applications using intraday solar irradiance forecasting.

• The high bias in Meteosat-8 IODC results confirms the necessity to adjust the cloud classification according to BTD thresholds on a particular region.

• Similar IR channels can be used on third generation meteorological satellites (Himawari-8, GOES-R, MTG…).

Global Horizontal Irradiance (GHI) values forecasted in the morning between 4:00 and 10:00 UTC for Carpentras and between 1:00 and 7:00 UTC for Pamandzi and Sainte-Marie. nRMSE, nMBE and R concern nighttime forecasts only (pink dots)

SiteLatitude(dec. deg.)

Longitude (dec. deg.)

Altitude (m) SatelliteGroundmeasurements

Coveredperiod

Carpentras 44.08 5.06 99Meteosat-10Prime

BSRN [5] 2016

Sainte-Marie -20.89 55.53 8Meteosat-8 IODC

Météo-FranceFebruary to July 2017

Pamandzi -12,81 45.58 7Meteosat-8 IODC Météo-France

February to July 2017

Similar cloud distributions between daytime and nighttime can be seen using infrared images

Meteosat-8 and -10 images provided by

Meteosat-10 HRV channel image showing the twilight zone or terminator

HourCast method principle