Renewable Energy


Since 2009 RAL has collaborated with university researchers, DOE labs, commercial partners, and other NCAR laboratories to develop methods to more accurately analyze and predict wind and solar power in support of the renewable energy industry. Projects have focused on resource assessment, analysis of the interaction between atmosphere and operating wind turbines, and real time wind, solar, and load forecasting to improve operations and economics of incorporating renewable energy into the power mix, and characterization and quantification of variability in wind and solar energy.

NCAR scientists have become recognized as world experts in applying meteorology concepts for enhancing renewable energy production. During 2017, they collaborated on several book chapters documenting the background of meteorological modeling for renewable energy (Haupt et al. 2017a, b, Haupt 2017, Troccoli et al. 2017) and have started yet another one.


Renewable Energy Forecasting

Figure 2. High-level architecture of the blended DPV/load forecast system.
Figure 1. High-level architecture of the blended DPV/load forecast system.

Since 2008, NCAR RAL has been developing renewable energy forecasting systems. The renewable power forecasting system enables more economical utilization of resources and more reliable grid operation while still meeting the needs of the utility’s electricity customers.  In collaboration with Xcel Energy Services, Inc.,  RAL has developed, deployed and transferred wind and solar power forecasting systems, which allows powering down traditional coal- and natural gas–powered plants when sufficient winds and solar irradiance are predicted (Mahoney et al. 2012, Haupt et al. 2013, Haupt and Mahoney 2015, Haupt et al. 2017, Haupt and Kosovic 2017). Operational implementation of the initial day-ahead forecasting system resulted in significant savings for the utility and the ratepayer. Although the most recent collaboration with Xcel Energy was completed during FY15, during FY17 the team continued publications on the results, including documenting the impact of assimilating wind speed observations from the wind farms into the Weather Research and Forecasting (WRF) model, finding that it can make a substantial improvement in the forecast (Cheng et al. 2017) and NCAR’s work in building both distributed photovoltaic (DPV) solar and load forecasting systems and when it is necessary to combine the two (Haupt et al. 2017c). Figure 1 provides a view of the architecture of the load plus distributed solar forecasting system.

Figure 3. Normalized RMSE (left) and Bias (right) at six sites, indicated by the broken line colors, from data averaged over a 6-month period.
Figure 2. Normalized RMSE (left) and Bias (right) at six sites, indicated by the broken line colors, from data averaged over a 6-month period.

The DPV forecasting system takes a “top down” approach, due to the lack of detailed data regarding site specific solar power production because most of it is “behind the meter” and appears as a decrease in load. For the same reasons, it is difficult to completely independently assess the accuracy of the DPV forecasts because very few sites report power output from their solar panels. Figure 2 displays the normalized RMSE at six sites of the Boulder Valley School District (BVSD) where production data are available. These errors represent the median of six months (Jun - Nov 2014) of Intraday (0-15hrs) and Day Ahead (24-39hrs) forecasts initialized at 1100 UTC. Note that most nRMSE values are less than 3%. Also note that Day Ahead forecasts do not exhibit degraded accuracy. 

Figure 4. Scaled PSCo net loads at 2 UTC vs. the weighted average temperature from ten METAR sites, with the color scale representing the Julian day.
Figure 3. Scaled PSCo net loads at 2 UTC vs. the weighted average temperature from ten METAR sites, with the color scale representing the Julian day.

The load forecast is highly dependent on weather conditions, as seen in Figure 3, which plots electric load as a function of temperature, with each point colored by Julian day of the year. For temperatures colder than about 12°C, the load values increase in an approximately linear fashion. Similarly, as the temperatures rise above 12°C, they increase in a quadratic or perhaps exponential manner. As expected, Julian days associated with late fall, winter, and early spring are predominantly on the left side of the plot, while late spring, summer and early fall values appear on the right side. One can also discern a more subtle pattern: light blue and blue-green points (denoting spring to early summer) tend to represent smaller loads than those experienced at other times of year for the same temperature. This may be in part a reflection of humans tolerating greater temperature variations during a seasonal transition, or may demonstrate the impact of other weather variables (e.g., longer-term temperature trends, cloudiness or precipitation). In any case, this plot illustrates both the importance of daily, weekly and seasonal usage patterns and temperature, while also suggesting the complexity of accurately predicting electrical loads. This connection of weather to load was critical for constructing the load forecasting system. The load forecast performance was evaluated by analyzing the real-time results over a two-month period after the latest version of the forecast was implemented. The real-time system’s day-ahead performance, evaluated for the 20-43 hour lead-times for each 11 UTC forecast, shows errors within approximately 1.25% of maximum monthly load.

Finally, we assessed how much the solar forecast impacts the net load forecast. Does it impact the forecast currently and, as the capacity grows, will that growth necessitate direct inclusion of the DPV forecast as an input to the load forecast? We found that DPV is beginning to become discernable in the net load, but, the DPV capacity is growing rapidly in Colorado, increasing by a factor of ten in five years.  Many of the same variables used in the DPV forecast are already included in the load forecast, however. Unsurprisingly, an initial analysis suggested that the load forecast therefore implicitly includes information regarding the output of the DPV without explicitly including that forecast.  It is only during the most rapid growth in DPV deployment that explicitly including the DPV forecast in the load forecast became necessary.

Solar Power Forecasting

In 2013, RAL embarked on a major DOE-funded effort to advance the state-of-the science of solar power forecasting.  This work is in partnership with the National Renewable Energy Laboratory, Brookhaven National Laboratory, National Oceanographic and Atmospheric Administration; universities – Penn State, Colorado State, Hawaii, Washington, and University of Buffalo; utilities – Long Island Power and Light, Public Service of Colorado, Sacramento Municipal Utility District, Southern California Edison, and the Hawaiian Electric System; independent system operators (ISOs) – New York Power Authority, Xcel Energy, California ISO, and Hawaiian Electric; and commercial forecast providers – Schneider Electric, Atmospheric and Environmental Research, Global Weather Corporation, and MDA Information Systems.

Fig. 5. Value chain of implementing a weather decision support system for solar power. At the bottom are the components of the NCAR team’s system that build toward providing an economic impact of this system.
Figure. 4. Value chain of implementing a weather decision support system for solar power. At the bottom are the components of the NCAR team’s system that build toward providing an economic impact of this system.

The primary objective of this project is to develop a solar power forecasting system that advances the state-of-the-science through cutting-edge research, tests it in several high penetration solar utilities and ISOs, and disseminates the research results widely to raise the bar on solar power forecasting technology. This is a Big Data problem  (Haupt and Kosovic 2017). To reach this goal requires basic and use-inspired research in targeted core areas. Metrics have been developed in collaboration with DOE, the other DOE-funded team led by IBM, and thoroughly vetted by the stakeholders (Jensen et al. 2016). These metrics measure improvements in solar forecasts, the resulting power predictions, and value to the utility or ISO. A major advance has been developing WRF-Solar (Jimenez et al. 2016a, b) and showing the comparative value of the forecasting methods through a series of case studies (Lee et al. 2016). For the shortest ranges, NCAR built a regime dependent machine learning forecasting system (McCandless et al. 2015, 2016a,b)

This project was successfully completed in FY16 and the team completed documentation of the new Sun4Cast sysem in the literature (Haupt et al. 2017d) and making the models OpenSource (  The project focused on identifying elements of a value chain as displayed in Figure 4, beginning with the weather that causes a deviation from clear sky irradiance and progressing through monitoring of observations, modeling, forecasting, dissemination and communication of the forecasts, interpretation of the forecasts, and through decision-making, which produces outcomes that have an economic value. The system was evaluated using metrics developed specifically for this project, which provided rich information on model and system performance. 

The Sun4Cast® system requires substantial software engineering to blend all of the new model components as well as existing publicly available NWP model runs (Figure 5). To do this we use an expert system for the Nowcasting blender and the Dynamic Integrated foreCast (DICast®) system for the NWP models. These two systems are then blended, using an empirical power conversion method to convert the irradiance predictions to power, and then applying an analog ensemble (AnEn) approach to further tune the forecast as well as to estimate its uncertainty. The AnEn module decreased Root Mean Square Error (RMSE) by 17% over the blended Sun4Cast power forecasts and provided skill in the probabilistic forecast with a Brier Skill Score of 0.55. In addition, we developed a Gridded Atmospheric Forecast System (GRAFS) in parallel, leveraging cost share funds.

Lidar Support for Wind Energy

RAL researchers worked closely with faculty and students from the University of Colorado and the National Renewable Energy Laboratory’s National Wind Technology Center to deploy lidars in field studies relevant for boundary layer meteorology and wind energy applications. In 2017 RAL’s lidar was deployed in support of the Wind Forecast Improvement 2 (WFIP2) project. The WFIP2 project was a year-long, multi-laboratory, multi-university field study in the area of Columbia River Gorge. Observations from this study are now being used to improve high-resolution wind forecasting in complex terrain. While the WFIP2 project is focused on improving numerical weather prediction for wind energy applications, these improvements will have broader impact by improving overall weather prediction.

Wind Forecast Improvement Project in Complex Terrain Near the Columbia River Gorge

Fig 6. Large-eddy simulation of a topographic wake and mountain waves observed on March 07 – 08, 2016 during WFIP2 filed study in the Columbia River Gorge Area. Contour plot of vertical velocity is shown at 1200 m above ground level. The wake of Mt. Hood is well resolved as well as the mountain waves resulting from the Westerly flow over the Cascade Range.
Figure 5. Large-eddy simulation of a topographic wake and mountain waves observed on March 07 – 08, 2016 during WFIP2 filed study in the Columbia River Gorge Area. Contour plot of vertical velocity is shown at 1200 m above ground level. The wake of Mt. Hood is well resolved as well as the mountain waves resulting from the Westerly flow over the Cascade Range.

NCAR is collaborating with Vaisala Inc. in developing improved planetary boundary layer parameterizations for high-resolution mesoscale simulations of flows in complex terrain with application to wind forecasting. In general physics parameterizations in mesoscale models including parameterizations of turbulent stresses and fluxes are based on the assumption of horizontal homogeneity and are therefore essentially one-dimensional parameterization. Computational resources now enable mesoscale simulations with grid cell sizes of 1 km or less. While such simulations resolve well large-scale features, the effects of atmospheric boundary layer turbulence on the mesoscale flow must be parameterized. Accurate parameterization of turbulent stresses and fluxes is essential for accurate wind forecasting. However, as grid cell size decreases over heterogeneous surfaces including complex terrain the homogeneity assumption is violated. Therefore, the goal is to develop a fully three-dimensional PBL parameterization that will account for the effects of surface heterogeneities and enable more seamless coupling between mesoscale and microscale simulations. During 2017 RAL researchers have developed and implemented a fully three-dimensional boundary layer parameterization in the Weather Research and Forecasting model. The model developments are being validated using observations from the WFIP2 field study and large-eddy simulations (LES). To carry out LES studies RAL researchers have applied for and were awarded an NCAR Strategic Capability Computing grant on NCAR’s Yellowstone supercomputer as well as an Advanced Scientific Computing grant on NCAR’s new Cheyenne supercomputer. These grants were used to carry out several high-resolution, large-domain LES of selected case study periods from the WFIP2 field study. 

Mesoscale to Microscale Coupling for Renewable Energy

Figure 7. Process for coupling mesoscale to microscale simulations for wind plant management.
Figure 6. Process for coupling mesoscale to microscale simulations for wind plant management.

A collaboration with DOE commenced in FY15 that focuses on blending information from mesoscale model simulations into microscale simulations in order to provide a capability to more accurately model details of flow that impacts a wind plant (Figure 6). NCAR is leading a collaboration of five DOE national laboratories (Argonne, Los Alamos, Lawrence Livermore, National Renewable Energy Lab, and Pacific Northwest National Laboratory) to accomplish mesoscale and microscale coupled simulations of carefully selected cases that are representative of wind farm conditions. In FY17 the team focused on nonstationary cases in complex terrain, specifically modeling the coupled mesoscale-microscale system for cases that included conditions such as cold pools, gravity waves, mountain wakes, and other phenomena forced by the complex terrain. NCAR performed formal statistical assessment tailored to the needs of wind farm production. A major finding is that it is important to couple the mesoscale model to the microscale in order to produce the full forcing needed to model the details of the flow and turbulence at the right scales. More specifically, nesting the microscale in the mesoscale is an important feature, because the nesting allows continued coupling of the physical parameterizations into the scales of the wind plant. The impact of this project is expected to include improved wind plant operation and performance.  The team completed documentation of the results of the first and second year results (Haupt et al. 2015, Haupt et al. 2017e)  and will also fully document the work from FY17. 

With a view toward a stepwise increase of complexity beyond the canonical, steady flat terrain cases assessed during year one, the second year’s activities focused on nonstationary cases and in year three (FY17) efforts targeted nonstationary cases over complex terrain. The meteorology of complex terrain can be quite complicated, well beyond the obvious inhomogenous nature of the lower boundary condition. The complexity generates phenomena that require the physics incorporated in the mesoscale models. During stable conditions, cold pools commonly form in the pockets between the mountains. The breakup of these cold pools depends on various meteorological features beyond just the changes in heating, which includes shadowing effects. It also depends on the conditions being advected into the region, including the passage of weather patterns. In addition, the terrain generates topographic wakes and waves. During some conditions, hydraulic jumps occur, which depend on the baroclinicity that can be modeled with the mesoscale models. Thus, a goal of FY17 is to study the impact of coupling the mesoscale models to the microscale models and determining to what extent we can capture these complex features that will modify the energy available at the wind farms. To test the coupling methods for complex terrain, the team has utilized measurements from the WFIP2 project. The data available from that experimental campaign provided the observations required to assess and fine-tune coupled model performance for this challenging environment.

Wind Resource Assessment in the Developing World

In 2014, NCAR embarked on a project with the National Renewable Energy Laboratory (NREL) to assess and quantify the wind resource in Bangladesh. NREL deployed sodars on site as well as several meteorological observation towers. NCAR’s role is to use modeling capabilities to assess the resource and to assimilate data from the new observational network to calibrate the models. The first step was to compare the wind resource from three separate historical reanalyses, which blend information from historical observations with models. The next step involved working closely with NREL to downscale and assimilate the observations using the Real Time Four Dimensional Data Assimilation (RTFDDA) of RAL. NCAR is currently assimilating the observations from the deployed instruments to compute at high resolution 2 years’ worth of modeled wind speeds throughout the country to determine the best locations to deploy wind energy. 


FY2018 will continue to be an exciting time for renewable energy research at RAL. New collaborations with national laboratories, university scientists, private companies, and foreign research institutions and companies will advance the state-of-the-science necessary to make a large penetration of renewable energy capacity feasible. In FY2018 significant efforts will include advancing comprehensive renewable power forecasting capabilities. An emphasis will be on porting the research to other regions of the world, especially the Middle East and Asia. In addition, the analog ensemble methodology will be further advanced and applied to a range of renewable energy related projects to quantify uncertainty.

Other plans include:

  • Continuing to work with NREL with resource assessment and developing measurement programs in developing countries, including Bangladesh.
  • Expansion of the wind forecasting capability into new areas, including international, complex terrain and desert sites.
  • Continued collaboration with DOE laboratories to develop best practices for coupling mesoscale with microscale simulations, focusing on complex terrain and nonstationary conditions.
    • Continued collaboration with Vaisala Inc. and DOE laboratories to improve mesoscale simulations of flows in complex terrain as part of the DOE-funded WFIP2 project.
    • Initializing a new project devoted to blending the satellite data assimilation MADCast facility with WRF-Solar to develop a more robust and accurate version of MAD-WRF to improve nowcasts of solar irradiance. 



Cheng, W.Y.Y., Y. Liu, A. Bourgeois, Y. Wu, and S.E. Haupt, 2016: Short-term Wind Forecast of a Data Assimilation/Weather Forecasting System with Wind Turbine Anemometer Measurement Assimilation, submitted to Renewable Energy, recently revised.

Haupt, S.E., 2017: Short-Range Forecasting for Energy, in Weather and Climate Services for the Energy Industry, A. Troccoli, ed., Palgrave Macmillan, London, UK, in press.

Haupt, S.E. and B. Kosovic, 2017: Variable Generation Power Forecasting as a Big Data Problem, IEEE Transactions on Sustainable Energy, 8 (2), pp. 725-732. DOI: 10.1109/TSTE.2016.2604679.

Haupt, S.E., B. Kosovic, J.A. Lee, and P. Jimenez, 2017a: Mesoscale Modeling of the Atmosphere, in Modeling and Simulation in Wind Plant Design and Analysis, P. Veers, Ed., IET Press, Submitted.

Haupt, S.E., P. Jimenez, J.A. Lee, and B. Kosovic, 2017b: Principles of Meteorology and Numerical Weather Prediction, in Renewable Energy Forecasting: From Models to Applications, G. Kariniotakis, Ed., Elsevier, London, UK, pp. 3-28.

Haupt, S.E., S. Dettling, J. Williams, J. Pearson, T. Jensen, T. Brummet, B. Kosovic, G. Wiener, T. McCandless, and C. Burghardt, 2017c: Impact of Distributed PV on Demand Load Forecasts, Solar Energy,157, 542-551.

Haupt, S.E., B. Kosovic, T. Jensen, J. Lazo, J. Lee, P. Jimenz, J. Cowie, G. Wiener, T. McCandless, M. Rogers, S. Miller, M. Sangupta, Y. Xue, L. Hinkelman, P. Kab, J. Heiser, 2017d: Building the Sun4Cast System: Improvements in Solar Power Forecasting, submitted to Bulletin of the American Meteorological Society, , Bulletin of the American Meteorological Society, early online release. doi: 10.1175/BAMS-D-16-0221.1

Haupt, S.E., A. Anderson, R. Kotamarthi, J.J. Churchfield, Y. Feng, C. Draxl, J.D. Mirocha, E. Quon, E. Koo, W. Shaw, R. Linn, L. Berg, B. Kosovic, R. Rai, B. Brown, B.L. Ennis, 2017e: Second Year Report of the Atmosphere to Electrons Mesoscale to Microscale Coupling Project: Nonstaionary Modeling Techniques and Assessment, Pacific Northwest Laboratory Report PNNL-26267, 156 pp.

Haupt, S.E., A. Anderson, L. Berg, B. Brown, MJ Churchfield, C Draxl, B.L. Ennis, Y. Fang, B. Kosovic, Rs. Kotamarthi, R. Linn, J.D. Mirocha, P. Moriarty, D. Munoz-Esparaza, R. Rai, W.J. Shaw, 2015: First Year Report of the A2e Mesoscale to Microscale Coupling Project, Pacific Northwest Laboratory Report PNNL-25108, 124 pp.

Haupt, S.E. and W.P. Mahoney, 2015: Wind Power Forecasting, IEEE Spectrum, Nov 2015, pp. 46-52.

Haupt, S.E., W.P. Mahoney, and K. Parks, 2013: Wind Power Forecasting, in Weather Matters for Energy, Troccoli, Dubus, and Haupt, eds, Springer, in press.

Jensen, T.L, T.L. Fowler, B.G. Brown, J. Lazo, S.E. Haupt. 2016: Metrics for evaluation of solar energy forecasts.  NCAR Technical Report TN-527+STR, 67 pp, doi:10.5065/D6RX99GG.

Jimenez, P.A., J.P. Hacker, J. Dudhia, S.E. Haupt, J.A. Ruiz-Arias, C.A. Gueymard, G. Thompson, T. Eidhammer, and A.J. Deng, 2016a:  WRF-Solar: An Augmented NWP Model for Solar Power Prediction, Bull. Amer. Met. Soc, July 2016, 1249-1264. DOI:10.1175/BAMS-D-14-00279.1.

Jimenez, P.A., S. Alessandrini, S.E. Haupt, A. Deng, B. Kosovic, J.A. Lee, and L. Delle Monache, 2016b:  Role of Unresolved Clouds on Short-Range Global Horizontal Irradiance Predictability, Monthly Weather Review, 144, 3099-3107. DOI: 10.1175/MWR-D-16-0104.1

Lee, J.A., S.E. Haupt, P.A. Jimenez, T.C. McCandless, M.A. Rogers, and S.D. Miller, 2017: Solar Energy Nowcasting Case Studies near Sacramento, Weather and Forecasting, 56, 85-108. DOI:

 McCandless, T.C., S.E. Haupt, and G.S. Young, 2016a:  A Regime-Dependent Artificial Neural Network Technique for Short-Range Solar Irradiance Forecasting, Applied Energy, 89, 351-359.

McCandless, T.C., G.S. Young, S.E. Haupt, and L.M Hinkelman, 2016b:  Regime-Dependent Short-Range Solar Irradiance Forecasting, Journal of Applied Meteorology and Climatology, 55, 1599-1613.

McCandless, T.C., S.E. Haupt, and G.S. Young, 2015: A Model Tree Approach to Forecasting Solar Irradiance Variability, Solar Energy, 120, 514-524. DOI:10.1016/j.solener.2015.07.0200038-092X

Mahoney W. P., K. Parks, G. Wiener, Y. Liu, W. L. Myers, J. Sun, L. Delle Monache, T. Hopson, D. Johnson, S. E. Haupt, 2012: A Wind Power Forecasting System to Optimize Grid Integration. IEEE Trans. Sustain. Energy, 3, 670-682.

Troccoli, A., M. Bruno Soares, L. Dubus, S.E. Haupt, M. Dadeck Boulahya, and S. Dorling, 2017: Forging a Dialogue Between the Energy Industry and the Meteorological Community, in in Weather and Climate Services for the Energy Industry, A. Troccoli, ed., Palgrave Macmillan, London, UK, in press.