Short-Term Cloud Motion Forecasting Model
One of the tasks during the first year of the LACISE project is to explore algorithms and deep neural network architectures in order to develop and validate the foundations of a cloud motion forecasting model for very short prediction horizons. This is carried out by scientists from the Institute of Electronics and Computer Science (EDI, Latvia) together with partners from the Swiss Center for Electronics and Microtechnology (CSEM, Switzerland).
Cloud movement is among the most challenging atmospheric processes to predict, and its rapid variability significantly affects both weather conditions and technological systems that depend on solar radiation. Therefore, accurate forecasting of cloud displacement over very short time intervals, ranging from a few minutes to half an hour, is becoming increasingly important.
Short-term cloud forecasts are particularly critical in solar energy production, where cloud shadows can cause rapid power fluctuations. Timely prediction of cloud appearance or disappearance enables operators to stabilize the electricity grid, plan generation, and reduce the need for reserve capacity.
The developed cloud motion forecasting model (see figure) is capable of analyzing sequential sky observation camera images and calculating how cloud structures will evolve over the upcoming minutes. The figure illustrates five input frames and the model’s 90-second forecast, compared with the actual sky observed at the same moment. This solution significantly reduces uncertainty in solar energy planning, allowing for the timely anticipation of cloudy periods and their impact on electricity generated by solar panels.
Figure: cloud motion forecast for a 90-second horizon: model prediction versus
observed reality
In 2026, the forecasting model is planned to be expanded by incorporating parametric data from a solar irradiance sensor, wind speed sensor, temperature sensor, and solar panel performance indicators. This approach will enable the model to synchronously link cloud movement with real changes in solar irradiance generation, thereby substantially improving forecast accuracy for solar energy planning.