UC San Diego

Solar energy integration into the power grid requires forecasting of solar power due to the natural variability of solar irradiance, primarily due to clouds. Recently, rooftop solar photovoltaic (PV) installations have grown dramatically in California, with a majority of these systems concentrated near the densely-populated coast. A common weather phenomena in coastal California are vast sheets of convective low clouds called Stratocumulus (Sc) clouds. The formation and dissipation of Sc clouds greatly impact solar power production.

The challenges surrounding the modeling of Sc clouds come from the complex interplay among the governing physics-- namely, surface-driven convection, cloud-top triggered convection, microphysical processes, and entrainment across the inversion layer. This thesis uses tools including satellite images, statistical analysis, and numerical weather prediction (NWP) models to predict Sc clouds and generate solar irradiance forecast. First, a novel approach of tracking the evolution of stationary clouds to produce short-term solar forecast using satellite images is presented. The method predicts Sc dissipation time accurately by overcoming the limitation of the frozen cloud assumption in traditional cloud motion vector satellite solar forecasting techniques. Second, an observation-based analog ensemble forecast is investigated to better understand the meteorological variables contributing to Sc cloud lifetime. Intra-day analog ensemble solar forecasts suggest that boundary layer averaged heat, moisture, and height are key to capturing Sc dissipation time. Finally, Sc forecast beyond intra-day is investigated through the use of a numerical weather prediction model-- Weather Research and Forecasting (WRF). In WRF, the planetary boundary layer (PBL) scheme parameterizes mixing processes in the PBL that impact the heat and moisture profile in the PBL. An eddy-diffusivity/mass-flux (EDMF) framework is adopted to model realistic mixing in the PBL. The framework models the non-convective environment through eddy-diffusivity and the convective area using a mass-flux. Specifically, cloud-top triggered downdrafts are developed and integrated into the mass-flux model. Turbulent downdrafts help mix the heat and moisture in the top part of the PBL and result in better representation of thermodynamic profiles and cloud thickness.