Development of a Two-Step EEOF Alogrithm in Collaboration with the NOAA-CPC Aimed at Reducing Error Growth within the Models.

Errors in numerical weather model prediction (NWP) grow over long lead times. A time extended empirical orthogonal function (EEOF) analysis of GEFS V12 model error suggests that biases in NWP generate large-scale patterns of error which grow with further forecast leads. These biases then systematically matriculate into altered positions of modeled atmospheric features which lead to the error in the synoptic and seasonal time scales. While individual weather events and anomalous weather patterns appear as noise, it is possible to separate this noise from the inherent model biases in representation of propagating and standing features. For 200hPa (Z200), systematic biases often emerge as errors in atmospheric wave patterns in the higher latitudes, such as inaccurate Rossby Wave phase speeds and or group velocities, leading to a decreased forecast skill in the model. Once this wave pattern becomes out of sync with the verification, model error will continue to grow with further lead times, decreasing the forecast skill.

In collaboration with the NOAA-CPC, a statistical postprocessing algorithm is used to identify patterns of model error using linear transformation techniques and then the algorithm is used to predict model error based on the projection of model forecast anomalies from the model seasonal cycle of error onto the error patterns. With the highest confidence from the Z200 algorithm in the middle latitudes, the algorithm using EEOFs, aims to construct error eigenmodes that evolve with further lead time. To best adapt this algorithm to a Research to Operations workspace (R2O), the statistical postprocessing algorithm was projected onto real time GEFSV12 ensemble forecasts, to remove the model error calculated by the historical EEOF algorithm in real time. Additionally, the global algorithm was domain restricted to both the Southern and Northern Hemisphere’s, demonstrating the error which grows with forecast lead and showing the leading pattern of error while removing the seasonality aspect that comes with analyzing the Z200 mid-latitudes. The removal of these errors creates an adjusted forecast with reduced error. Further analysis aims to adapt this algorithm to other atmospheric parameters, as well as the upcoming GEFSV13 ensemble product.