Hawaii’s dependence on imported energy, high electricity costs, and abundance of sunshine have led to wide-spread deployment of photovoltaic (PV) systems. However, because of the variable, distributed and “behind the meter” nature of solar power generation, an inherent uncertainty accompanies increased PV penetration. This uncertainty is particularly concerning for the small and isolated electric grids in Hawaii. The Hawaii Natural Energy Institute (HNEI) is developing an operational, high fidelity, ground-level irradiance monitoring and forecasting system that provides timely information at sub-hourly to daily horizons for the prediction of solar energy generation in Hawaii.

Point Person: Dax Matthews

The HNEI Solar Forecasting System utilizes (1) numerical weather prediction, combined with (2) satellite and (3) ground based observations, to predict ground-level irradiance at various forecasting horizons.

1) Numerical weather prediction (NWP) forecasts are used to provide irradiance forecasts from hours to days ahead. Forecasts in this range are used to determine the need for operating reserves and scheduling utility-scale solar plant generation. While NWP models are not as accurate as observation-based predictions for short-range forecasts, at horizons longer than 4-6 hours NWP is necessary to account for the more complex dynamics that drive cloud fields.
Each night, we run the Weather Research and Forecasting (WRF) solar model — a version of WRF with improved radiation schemes that directly compute global (GHI), diffuse (DIF) and direct (DNI) irradiance, and new shallow convection and cloud microphysics schemes. By 8 am a 48-hour, 10-km resolution, state-wide prediction is generated — a nested 2-km grid covering the islands of Oahu, Hawaii, Maui, Molokai and Lanai will be added to the operational system in February 2015.

2) The use of satellite images for solar forecasting is a fundamental technique for modern systems and has been shown to produce useful predictions from ~30 minutes to 4-6 hours. Distribution utilities can use such forecasts to gain situational awareness and to address potential reliability concerns within the networks they manage. 
We use GOES images to monitor current cloud conditions, and through well-tested algorithms, estimate current surface irradiance values and cloud motion vectors. We combine this information with WRF output to estimate cloud top height (CTH) and produce a high-resolution velocity field, from which we predict future cloud locations, cloud shadows, and surface irradiance using a cloud motion model and the Heliosat method. Nominally, every 30 minutes (from 7am to 7pm), a 6 hour, 1 km, state-wide forecast is generated from the most recent GOES image; the latency is less than 30 minutes.

3) Total sky imaging (TSI) systems can provide significantly higher resolution cloud images compared to satellites, and have been used to predict solar plant outputs in realtime to up to ~30 minutes ahead. This resolution and range of forecasting allows for the prediction of ramp events caused by sudden changes in solar resources. 
We have installed a TSI-880 Automatic Total Sky Imager and Eliasson CBME 80 cloud ceilometer on the roof of the Hawaiian Institute of Geophysics (HIG) building at the University of Hawaii of Manoa campus. The sky imager provides overhead sky conditions and the ceilometer measures cloud base height. By applying image navigation and classification techniques, and cloud tracking and motion algorithms, 30 minute forecasts are generated. The sky conditions and cloud base heights affect the spatial resolution (~100 m) and area coverage (7-15 km) of the sky images.