Quantify the effect of forecast meteorology on FRP prediction skill

Quantify the reduction in global daily FRP forecast skill when FRP-GraphCast-Lite is driven by operational forecast meteorology rather than reanalysis meteorology.

Background

The reported verification uses ERA5 reanalysis meteorology as predictor input, thereby isolating the learned FRP-prediction component from meteorological forecast errors. Operational deployment would instead use forecast meteorology from systems such as GFS or GEFS. The paper identifies the magnitude of the resulting skill degradation as unknown and states that it will be assessed in future work.

References

The verification reported here uses reanalysis meteorology as predictor input in order to isolate the skill of the FRP model from errors in the meteorological forecast, and therefore represents an upper bound on the skill achievable with forecast meteorology of comparable quality; the magnitude of that reduction has not yet been quantified and will be assessed by driving the model with GFS/GEFS forecast meteorology (see the final paragraph of Section 6).

— A Graph Neural Network for Global Daily Fire Radiative Power Prediction at Medium-Range Lead Times  (2610.11022 - Zhang et al., 8 Oct 2026) in Section 3.5, Operational Configuration and Inference; Section 6, Summary and Discussion