Welcome to Scribd, the world's digital library. Read, publish, and share books and documents. See more
Standard view
Full view
of .
Look up keyword
Like this
0 of .
Results for:
No results containing your search query
P. 1
WRF Forecasting DNI Spain-CPV8 2012

WRF Forecasting DNI Spain-CPV8 2012

Ratings: (0)|Views: 23|Likes:
Published by smart_chris
Discusses the possibilities to forecast solar radiation in Spain for concentrating PV systems
Discusses the possibilities to forecast solar radiation in Spain for concentrating PV systems

More info:

Categories:Types, Research, Science
Published by: smart_chris on Sep 10, 2012
Copyright:Attribution Non-commercial


Read on Scribd mobile: iPhone, iPad and Android.
download as PDF, TXT or read online from Scribd
See more
See less





 CPV-8 Conference, Toledo, Spain, 2012
Evaluation Of DNI Forecast Based On The WRFMesoscale Atmospheric Model For CPV Applications
V. Lara-Fanego
, J. A. Ruiz-Arias
, A. D. Pozo-Vázquez
,C. A. Gueymard
and J. Tovar-Pescador 
 Department of Physics, University of Jaén. Campus Las Lagunillas A3-066, 23071, Jaén (Spain)
Solar Consulting Services, P.O. Box 392, Colebrook, NH 03576, USAvlara@ujaen.es
The integration of large-scale solar electricity production into the energy supply structures depends es-sentially on the precise advance knowledge of the available resource. Numerical weather prediction (NWP) models provide a reliable and comprehensive tool for short- and medium-range solar radiation forecasts. The methodologyfollowed here is based on the WRF model. For CPV systems the primary energy source is the direct normal irradi-ance (DNI), which is dramatically affected by the presence of clouds. Therefore, the reliability of DNI forecasts isdirectly related to the accuracy of cloud information. Two aspects of this issue are discussed here: (i) the effect of the model’s horizontal spatial resolution; and (ii) the effect of the spatial aggregation of the predicted irradiance.Results show that there is no improvement in DNI forecast skill at high spatial resolutions, except under clear-skyconditions. Furthermore, the spatial averaging of the predicted irradiance noticeably reduces their initial error.
DNI, WRF, NWP, forecast.
88.05.Lg, 96.60.Ub, 92.60.Vb, 94.20.Cf 
Technologies for harnessing the solar resourcehave experienced a significant development in recentyears. Their future looks even more promising. TheInternational Energy Agency expects that, accordingto a reference scenario, the world’s installed solar  power capacity will increase from 14 GW in 2008 to119 GW in 2035, with a 8.3% average annual in-crease [1]. Therefore, the challenge for the next fewyears is to achieve a high level of development andintegration, to make this resource competitive com- pared to traditional sources of energy, or even tomore established renewable sources, like wind. Amajor effort is being made in this regard [2]. Toachieve this goal, a key aspect concerns the resourceitself (technology aside). The safe and optimal inte-gration of large-scale solar electric power productioninto the energy grid of any country depends on theknowledge of the solar production capacity, whichin turn is directly related to the available resource.An important intrinsic characteristic of solar ra-diation is its very high variability over space andtime, itself directly dependent on weather character-istics. This intermittency in the resource makes a so-lar plant’s operation and management particularlydifficult. It also makes solar production troublesomefor grid system operators, since it is hardly con-trolled and may not be available when it would be of greatest value [1]. This ultimately translates into in-cremental exploitation and integration costs. There-fore, prior knowledge of the available resource of the near future is essential. Previous experience withthe wind energy sector has shown that accurate fore-casts play a key role toward the successful integra-tion of variable energy sources.CPV systems use the beam component of solar radiation—or direct normal irradiance (DNI)—astheir energy source. DNI is primarily affected byclouds, aerosols, and water vapor. Clouds are nor-mally the principal factor affecting the incident solar radiation at the earth’s surface, since they are mostoften completely opaque to DNI. In contrast, aero-sols are most influential under cloudless conditions.The uncertainty in the determination of the physical parameters associated with these atmospheric con-stituents is the main source of error in DNI predic-tions. This study focuses on how the latter is affect-ed by cloudiness forecasts. Numerical weather prediction (NWP) modelshave been proved to be powerful tools for solar radi-ation forecasting [3, 4]. One particular tool that iswidely used by the research community is theWeather Research and Forecasting (WRF) model[5]. WRF, like other NWP models, has a wide rangeof physical parameterizations, providing the possi- bility to achieve high spatial and temporal resolu-tions. It is commonly assumed that the higher theresolution, the better the physical description andresults will be. This should apply, for instance, to therepresentation of processes that lead to the formationof clouds. In turn, high resolutions are computation-ally very expensive. Therefore, an optimal spatialresolution may exist in solar forecasting.This contribution evaluates the role of the WRFmodel’s horizontal spatial resolution in the reliabil-ity of the DNI forecasts that it can (indirectly) gen-erate. Additionally, the intentional use of spatial av-eraging of the gridded WRF-derived solar field toimprove the model’s accuracy is evaluated. Themethodology applied is described first. A descriptionof the forecast results is presented in a second step.
METHODOLOGYObservations and Evaluation Procedure
This study is conducted for the Andasol Solar Thermal Power Plant (37.228º N, 3.069º W; 1100m.a.s.l.), Fig. 1. Ten-minute DNI measurements arecollected with an RSR2 radiometer. This instrumentis well maintained and calibrated. Data for the 12-month period 01/12/2009 to 30/11/2010 were firstcorrected for spectral effects, and finally filteredwith a series of quality control tests. For this study,irradiance values corresponding to solar zenith an-gles above 85º were filtered out to avoid the highmeasurement uncertainties associated with low-sunconditions. The original 10-minute data were alsoaveraged to obtain hourly values. From a climato-logical standpoint, 2010 was an exceptionallyrainy—and therefore cloudy—year.Two forecast horizons are studied separatelyhere: hours 1–24 (day 1, or “day ahead”), and hours25–48 (day 2). Sky conditions are characterized bythe clearness index (k 
) to separate clear-sky (0.65 <
), cloudy (0.4
0.65) and complete overcast(k 
< 0.4) conditions. The forecast reliability is ob- jectively evaluated in terms of mean bias error (MBE) and root mean squared error (RMSE), andtheir relative values (in %), obtained by normaliza-tion to the mean of the ground measurement irradi-ance for the considered period. The forecast errors(residuals) are calculated as the difference betweenforecasted values and observations. A positive MBEis thus indicative of an overestimation of the mod-eled DNI. Finally, the trivial persistence model isused as the skill reference model.
WRF Configuration
The model’s domain configuration is representedin Fig. 1. The dynamical downscaling is driven bythe use of four nested domains with progressivelydecreasing horizontal resolutions of 27, 9, 3, and 1km for the outermost to innermost domains. The at-mospheric column is decomposed into 28 verticallevels. The ECMWF/IFS weather forecasts are usedas initial and boundary conditions. For each day of the evaluation period a WRF (ARW, version 3) sim-ulation of 60 hours is run. The first 12 forecastedhours are considered as model spin up, and discard-ed. The next 48 hours are evaluated independentlyfor the first and second 24-hour periods. The WRF parameterizations are selected based on [6]. In par-ticular, Dudhia’s scheme is used for the shortwaveradiation parameterization.
DNI Derivation
 NWP models (e.g., WRF) do not usually provideDNI as an output variable. Therefore, DNI needs to be derived in a post-processing step based on WRF’scomprehensive forecasted information and an exter-nal radiative model [4, 7]. Recent studies have rig-orously analyzed the performance of a large set of different radiative models [8, 9]. Results showed thatmeteorological radiative models achieve a very high performance in DNI estimation under clear-sky con-ditions. In contrast, statistical/empirical modelsshow lower performance but more simplicity. Sincethis work focuses on the WRF aspects related tocloud modeling, the simplest way to derive DNI is preferred; nevertheless better results can be expected by using the first kind of radiative models mentionedabove [10]. An empirical statistical model [11] issimply applied here to obtain DNI from the WRFglobal horizontal irradiance (GHI) forecasts.
Configuration of the WRF domains. The spa-tial resolutions are 27 km, 9 km, 3 km and 1 km for do-mains D01, D02, D03 and D04, respectively. The radio-metric station is located at the center of all domains.
RESULTSDependence On Horizontal Resolution
Figures 2 and 3 show the performance results for the day-ahead forecast horizon over the whole 12-month period. Extremely large errors are obvious for complete overcast conditions (Fig. 2). Since DNI isvery sensitive to the presence of clouds, any misrep-resentation of cloudiness—in either space or time— in the model’s predictions may cause significant er-rors. The high variability of cloud type and cloudamount enhances this effect. The most important re-sult is that the errors (RMSE and MBE) increasewith spatial resolution, contrarily to what wouldhave been expected. An interesting exception is thecase of clear-sky conditions, under which the RMSEdecreases (Fig. 3). These results mean that cloudsare not better resolved at higher spatial resolution bythe model, at least in terms of their effect on solar radiation. In contrast, under cloudless conditions thetopographic effects, which are better resolved at fin-er resolutions, become more relevant. This is of par-ticular interest to CPV, since this technology cannormally be installed on uneven terrain.
MBE is found positive in all cases. Under clear-sky conditions, WRF tends to overestimate bothGHI and DNI, most probably because of a too lowaerosol optical depth (AOD) in the model. Thisoverestimation also occurs under cloudy conditions,which means that the model predicts less cloudinessthan will occur in reality. Nevertheless, it is im- portant to note that the WRF-based DNI forecastsmost generally outperform the persistence model.
Relative errors of the DNI day-ahead fore-casts over the complete data period. For RMSE, the orangecolor corresponds to clear skies, the green color to cloudyconditions, the blue color to complete overcast, and the redcolor to all-sky conditions. The inner bars indicate theMBE. The persistence model errors are indicated by hori-zontal segments: red for RMSE and blue for MBE.
Same as Fig. 2 but with a different scale for  better visualization.
The dependence of the forecast performance onthe forecast horizon is an important topic, since en-ergy sales in the daily electricity market must bemade
24 hours early. From this standpoint, the se-cond day of forecast may be more important than thefirst one.Figure 4 compares the performance results of DNI forecasts according to time horizon (24h vs.48h), for all possible spatial resolutions. These re-sults show that the DNI forecasts are remarkablystable over time. Both forecast horizons exhibitsimilar dependence on spatial resolution. It should be pointed out that these results correspond to rawmodel outputs.Better performance is achievable in DNI fore-casts with post-processing, as demonstrated in thenext section. Moreover, the exceptionally rainy con-ditions that prevailed during the study period canexplain a part of the large errors.
All-sky performance results for the 24h and48h forecast horizons. Bars correspond to RMSE, and bluesegments to MBE. The red dashed lines correspond to the persistence model’s RMSE (its MBE is
Spatial Averaging (Post-Processing)
The inability of the WRF model to simulate high-frequency spatial-temporal cloud changes (thus, DNIchanges) can be worked around by applying a spatialfiltering algorithm. In particular, DNI values aregathered from the model’s grid by averaging theDNI forecasts over windows of varying incrementalsize, with the target station always at the center of these windows. Spatial averaging of the predictedsolar radiation is a commonly used filtering tech-nique to remove the high-frequency variability inforecasts. Figure 5 shows the performance results for the 48h forecast horizon.
RMSE and MBE for the 48h forecast horizonand a spatial average of surrounding points from 1 (nearestgrid point from the station location) to 47 (square side of farthest grid points).
Both MBE and RMSE are reduced by the spatialaveraging process, for all initial spatial resolutions.This improvement depends on spatial resolution.The best results are obtained again for the coarser initial resolution (27 km) using an averaging win-

You're Reading a Free Preview

/*********** DO NOT ALTER ANYTHING BELOW THIS LINE ! ************/ var s_code=s.t();if(s_code)document.write(s_code)//-->