One of the key challenges in forecasting solar variability, is that deterministic forecasts (those that produce one value only) do not provide end users with an appropriate level of decision making information. Not all forecasts have the same likelihood of being true, and different weather and climatic conditions directly impact forecast likelihoods.
Probabilistic Forecast Data - P10, P90
At Solcast, we represent this uncertainty with confidence intervals, that represent a 10% and 90% probability bound, around a median forecast value. In our forecasting data, these show up as 'GHI10' or 'GHI90' (for Solar Radiation Data) or 'PvEstimate10' or 'PvEstimate90' for PV power forecasting (Utility Scale, Rooftop and Grid Aggregations).
The '10' scenarios represent the lower bound of what is expected in the forecast. The '90' scenarios represent the upper bound of what is expected in the forecast data. In our demonstration graphs in the API Toolkit, these are used to create the shading present in relevant images, here's an example from a solar farm site in Taiwan:
How are the '10' and '90' values produced?
All of Solcast's forecasting data are 'Rapid Update', meaning they are updated every 10-15 minutes. For each forecasting update, Solcast re-computes a semi-dynamical satellite based nowcasting ensemble consisting of 18 members. Members have different properties, such as one that represents persistence, others that move clouds more quickly, or slowly, etc, to generate a range of outputs. By reviewing the probability distribution across these 18 members, we can determine the amount of uncertainty in a given forecast. Once beyond the 4-6 hour range, the ensemble is determined by a suite of numerical weather models.
Below: An example of two different ensemble members (cloud opacities)
This information enhances decision making, as a wider range between 10 and 90 values means the forecast has less certainty. Moreover, as an example, the user can select the '10' scenario for their use case, when trying to minimise ramp rates or manage risk in a remote grid.