Search

The ability of four models to provide short-term (up to 6 h ahead) GHI and DNI forecasts in the Iberian Peninsula is assessed based on two years of data collected at four stations. The models follow (mostly) independent approaches: one pure statistical model (Smart Persistence), one model based on CMV derived from satellite images (Satellite), one NWP model (WRF-Solar) and a hybrid satellite-NWP model (CIADCast). Overall, results show Smart Persistence to be the best at the first lead steps, advective models (Satellite and CIADCast) at intermediate ones and the WRF-Solar at the end of the forecasting period. The break-even point between the advective models and WRF-Solar varies between 1 and 3 h for GHI and 3 and 5 h for DNI. Nevertheless, a detailed analysis shows enormous differences between models performance related to 1) the local geographic and topographic conditions of the evaluation stations; 2) the evaluated variable (GHI vs. DNI); and 3) the sky and synoptic weather conditions over the study area. Depending on the station and lead time, rRMSE values range from 25% to 70% for GHI and from 35% to 100% for DNI. For the same stations and leading time, rRMSE values for DNI are between 50% and 100% higher than the corresponding GHI counterparts. Depending on the synoptic pattern, rRMSE values are about 10/20% for GHI/DNI (3 h lead time, during high pressure conditions) to about 80/180% for GHI/DNI (during low pressure conditions). All models show a poor performance at a coastal station, attributed to a lack of ability to forecast clouds associated with sea-land breezes. To conclude, no single model proves to be the best performing model and, therefore, results show that the four models are, somehow, complementary. The advantages attained by this complementarity are further explored in a companion paper (Part II). The authors are supported by the Spanish Ministry of Economy and Competitiveness, project ENE2014-56126-C2-1-R and ENE2014-56126-C2-2-R (http://prosol.uc3m.es). The team from the University of Jaen is also supported by FEDER funds and by the Junta de Andalucía (Research group TEP-220). The authors thank all the provided support. The authors are in debt with the National Centers for Environmental Prediction (NCEP), EUMETSAT, Faculdade de Ciencias da Universidade de Lisboa, Grupo de Energía Solar of the Universidad Politécnica de Madrid and Abengoa Solar for providing the data used in this work.

In this article we explore the blending of the four models (Satellite, WRF-Solar, Smart Persistence and CIADCast) studied in Part 1 by means of Support Vector Machines with the aim of improving GHI and DNI forecasts. Two blending approaches that use the four models as predictors have been studied: the horizon approach constructs a different blending model for each forecast horizon, while the general approach trains a single model valid for all horizons. The influence on the blending models of adding information about weather types is also studied. The approaches have been evaluated in the same four Iberian Peninsula stations of Part 1. Blending approaches have been extended to a regional context with the goal of obtaining improved regional forecasts. In general, results show that blending greatly outperforms the individual predictors, with no large differences between the blending approaches themselves. Horizon approaches were more suitable to minimize rRMSE and general approaches work better for rMAE. The relative improvement in rRMSE obtained by model blending was up to 17% for GHI (16% for DNI), and up to 15% for rMAE. Similar improvements were observed for the regional forecast. An analysis of performance depending on the horizon shows that while the advantage of blending for GHI remains more or less constant along horizons, it tends to increase with horizon for DNI, with the largest improvements occurring at 6 h. The knowledge of weather conditions helped to slightly improve further the forecasts (up to 3%), but only at some locations and for rRMSE. The authors are supported by the Spanish Ministry of Economy and Competitiveness, projects ENE2014-56126-C2-1-R and ENE2014-56126-C2-2-R (http://prosol.uc3m.es). The University of Jan team are also supported by FEDER funds and by the Junta de Andalucia (Research group TEP-220).The authors are in debt with the National Centers for Environmental Prediction (NCEP), EUMETSAT, Faculdade de Ciencias da Universidade de Lisboa, Grupo de Energa Solar of the Universidad Politcnica de Madrid and Abengoa Solar for providing the data used in this work.

The intra-day modes of variability of the solar resources in the Iberian Peninsula, their associated weather patterns and their impact on the solar power output are assessed in this work. The analysis is performed for yearly and seasonal variability. Firstly, the modes of variability are identified by means of hierarchical cluster analysis. It is computed with two years of measured global horizontal irradiance (GHI) and direct normal irradiance (DNI) data gathered at four stations. Notably, three-hour statistics describing mean and variability of solar radiation are used as input to the cluster analysis. Secondly, synoptic weather patterns associated with each group resulting from the cluster analysis are assessed using sea level pressure and cloudiness data. Finally, the solar PV power yield associated with each mode is evaluated. The yearly analysis reveals the existence of four modes of variability of the solar resource in the study area. The four modes are shown to have a distinctive weather pattern and also specific impacts on solar power generation in the study area. Seasonal analyses show results similar to the annual analysis, but with marked seasonal differences.

A new method for short-term solar radiation forecasting (referred to as Cloud Index Advection and Diffusion, CIADCast) is proposed and validated. The method is based on the advection and diffusion of Meteosat Second Generation (MSG) cloud index estimates using the Weather Research and Forecasting (WRF) numerical weather prediction (NWP) model. The forecasted cloud index is transformed in global horizontal irradiance (GHI) and direct normal irradiance (DNI) forecasts by means of the Heliosat-2 method. The cloud index maps are inserted in the WRF vertical layer which corresponds to the cloud height provided by a ceilometer. GHI and DNI are forecasted up to 6 h ahead with 15 min of time resolution. The method was tested using 25 days of radiometric data collected at three stations located in southern Spain. Benchmarking models such as smart persistence, a cloud motion vector (CMV) based approach and the WRF-Solar suite of the WRF model are also evaluated. Results were analyzed in the light of the different topographic characteristics of the evaluation stations areas. Results proved that CIADCast is able to provide enhanced forecasts in areas with low topographic complexity, where cloud advection by the atmospheric mesoscale dynamics is not perturbed by mountain features. In these areas, CIADCast generally outperforms the other models, especially for DNI and partially cloudy conditions. On the other hand, in regions with complex topography, where the mesoscale cloud pattern is influenced by the mountains, the performance of the CIADCast model is poor and the use of persistence or the WRF-Solar model proved to be more appropriate.

The temporal resolution of aerosol optical depth (AOD) affects the prediction of surface irradiance and hence impacts solar resource assessments. Most current global AOD databases used for solar radiation estimation span from monthly to yearly time scales, which are much coarser than the typical hourly or sub-hourly time step used for irradiance calculations. The relative variation of the AOD’s mean and standard deviation (SD) calculated from data at monthly and daily temporal resolutions is assumed here as a proxy for the expected differences in the calculated long-term average of solar irradiance when using monthly and daily AOD data. This is specifically analyzed in Part 2 of this study. In this Part 1, the changes in the mean and SD of AOD caused by an alteration of the time scale from daily to monthly time resolution are investigated first. A database from 214 AERONET sites with long historical record series is used, as well as data from worldwide multi-year evaluations of AOD from a numerical weather model coupled with an aerosol transport model. The AERONET AOD data is fitted to the log-normal distribution and the results are separately analyzed for multiple AOD representation time scales from daily to monthly, and for sites with prevalence of coarse, mixed or fine aerosols. When the averaging period is increased from one to 30 days, three effects are noticed regardless of the aerosol regime: (i) the range of possible AOD values narrows; (ii) the probability around the modal AOD increases; and (iii) the AOD distribution’s mode moves toward higher values. On average, the mean AOD at sites dominated by coarse or mixed aerosols does not significantly vary (it does by only less than 1%) when the averaging time increases from one day to 30 days, although with significant dispersion from site to site (≈±5%). In contrast, again on average, it declines ≈4% at the fine aerosol sites. Conversely, the SD decays exponentially fast at all sites—faster at the fine aerosol sites—on average. For an averaging period of 30 days, the SD of AOD over Europe and inland areas of Asia can be as low as 40% of the daily SD, while for most other land areas it stays at ≈60% of the daily SD. The impact that this change in the AOD SD has on the prediction of surface irradiance is investigated in this study’s Part 2.

Long-term solar irradiance modelled using monthly-average aerosol optical depth (AOD) is biased compared to what is obtained using daily AOD values. This is due to a modification of the AOD frequency distribution that results from the coarsening of the time scale, combined with the nonlinear sensitivity of irradiance to AOD. The resulting alteration of the mean and standard deviation of AOD has been evaluated worldwide in this study’s Part 1. Here, the focus is to evaluate how the progressive coarsening (in 1-day steps) of the temporal representation of AOD affects the calculated long-term values of both global and direct irradiances. It is shown that, on average, their long-term values can be underestimated by as much as ≈6 W m −2 (≈1.2%) and ≈40 W m −2 (≈8%) respectively, over regions of North Africa and Asia, with respect to the reference case for which daily AODs are used. This irradiance underestimation can be parameterized as a function of the AOD time scale by means of four sitedependent parameters. The proposed parameterization provides a simple way to correct the long-term irradiance datasets that are commonly obtained using monthly-average AOD data.

In this work, a new method to derive hourly global horizontal irradiance (GHI) estimates from Meteosat Second Generation (MSG) imagery is presented. The method is based on an optimized Artificial Neural Network (ANN) ensemble model using a selection of the best ANN models identified from an initial ensemble that discerns between different sky conditions and an additional ensemble that considers all sky conditions together. For benchmarking purposes, hourly GHI estimates computed with the Heliosat-2 method, accounting for the diurnal variability of ground albedo, are used. Data collected during the 3-year period from 2009 to 2011 at 28 radiometric stations located in northern Africa, Middle East and Europe, are used in the procedure. From these stations, 7 are used to train the ANN models and the other 21 for independent validation. Results obtained with the proposed ANN ensemble model reduced the RMSE value of the Heliosat-2 model a 22% for all-sky conditions and a 42% for overcast conditions.

This article combines and discusses three independent validations of global horizontal irradiance (GHI) multi-day forecast models that were conducted in the US, Canada and Europe. All forecast models are based directly or indirectly on numerical weather prediction (NWP). Two models are common to the three validation efforts – the ECMWF global model and the GFS-driven WRF mesoscale model – and allow general observations: (1) the GFS-based WRF- model forecasts do not perform as well as global forecast-based approaches such as ECMWF and (2) the simple averaging of models’ output tends to perform better than individual models.

In this work, we evaluate the reliability of three-days-ahead global horizontal irradiance (GHI) and direct normal irradiance (DNI) forecasts provided by the WRF mesoscale atmospheric model for Andalusia (southern Spain). GHI forecasts were produced directly by the model, while DNI forecasts were obtained based on a physical post-processing procedure using the WRF outputs and satellite retrievals. Hourly time resolution and 3 km spatial resolution estimates were tested against ground measurements collected at four radiometric stations along the years 2007 and 2008. The evaluation was carried out independently for different forecast horizons (1, 2 and 3 days ahead), the different seasons of the year and three different sky conditions: clear, cloudy and overcast. Results showed that the WRF model presents considerable skill in forecasting both GHI and DNI, overall, better than a trivial persistence model. Nevertheless, both MBE and RMSE values presented a marked dependence on the sky conditions and season of the year. Particularly, for 24 h lead time, the MBE of the forecasted GHI was 2% for clear-skies and 18% for cloudy conditions. However, the MBE of the forecasted DNI increased up to about 10% and 75% for clear and cloudy conditions, respectively. Regarding RMSE values, in the case of forecasted GHI, results ranged from below 10% under clear-skies to 50% for cloudy conditions. In the case of forecasted DNI, RMSE ranged from 20% to 100% for clear and cloudy skies, respectively. This proved the higher sensitivity of DNI to the sky conditions. In general, an increment of the MBE and RMSE values with the cloudiness was observed. This reflects a still limited ability of the WRF model to properly forecast cloudy conditions compared to clear skies. Nevertheless, the model was able to accurately forecast steep changes in the sky (cloudiness) conditions. Finally, WRF performed considerable better than the persistence model for clear skies both for GHI and DNI, with relative RMSE values about a half. However, for cloudy conditions, performance was similar.