Development of Cloud Propagation Forecast System

Growing concerns over rising fossil fuel costs, energy security, resource depletion, climate change and sustainable economic development have stimulated the Renewable Energy (RE) sector. Solar energy is on the rise in the United States, but its variable nature and relative unpredictability remains a major obstacle for integration onto the regional energy grid. A primary source of variability is cloud cover, and accurate short-term prediction can significantly ease integration. Our cloud forecasting methodology uses a multi-source fusion of persistence, propagated cloud features, and short-term ensemble forecasts. Persistence and propagative components are derived from a Cloud Mask Generator (CMG) model developed in house, which combines multi-satellite, multi-spectral remotely sensed imagery, numerical assimilations and conventional environmental observations to perform cloud cover analysis. Propagation is Lagrangian, using motion vectors derived from cloud-top brightness temperature cross-correlations over time. Persistence is conditional, depending on a regional cloud field feature decomposition using a Laplacian-pyramid approach to isolate cloud field features by scale. In addition to persistence and propagative elements, the NOAA NCEP's Short Range Ensemble Forecast (SREF) data set is accessed and used as a predictor set with scale decomposition. All sources of data are fused using a self-validating and self-training regression system, automatically training using current data as validation of prior forecasts and then producing forecasts for specified future times. The CPF's architecture supports forecast lead times ranging from now-casting at +15 minutes past imagery receipt time out to day-ahead forecasts at +36 hours, with a reliability correction post-processing step. CPF can be run in line-of-sight, skydome, or probabilistic modes to suit the needs of final irradiance modeling.