Weather regimes affect short-term satellite solar forecast error

What the study found: Large-scale North Atlantic weather regimes were associated with different levels of error in short-term satellite-based solar forecasts. The study reports that forecast error varied significantly by regime, showing that atmospheric pattern matters for forecast reliability.
Why the authors say this matters: The authors say these weather regimes can be predicted several days in advance, so the findings may help anticipate forecast error. They conclude this could support decisions for optimizing photovoltaic (PV) integration, electricity trading, and development of machine-learning forecast algorithms.
What the researchers tested: The researchers ran an 8-year backtest of satellite-based forecasts made four hours ahead with a 15-minute time step. They compared the forecasts with pyranometer data, which are ground measurements of solar radiation, and examined four North Atlantic weather regimes: Atlantic Ridge, Scandinavian Blocking, NAO+, and NAO-.
What worked and what didn't: Forecast errors differed across weather regimes. The relative RMSE (root mean square error, a measure of forecast error) difference between Scandinavian Blocking and Atlantic Ridge was 10-12% in summer from 2016-2020 and about 10% after 2020. In winter, the difference was around 20% before 2020 and 15% after 2020.
What to keep in mind: The abstract does not describe detailed limitations beyond the scope of the backtest and the regimes studied. The findings are specific to the satellite-based short-term forecasts and North Atlantic weather patterns described in the study.

Key points

• Forecast error changed significantly with North Atlantic weather regime.
• Scandinavian Blocking and Atlantic Ridge showed the reported RMSE differences in both summer and winter.
• The study used an 8-year backtest of four-hours-ahead satellite forecasts with 15-minute steps.
• Forecasts were checked against pyranometer measurements of solar radiation.
• The authors say the results may help anticipate error for PV integration, electricity trading, and machine-learning forecasts.