Abstract

The NOAA NCEP has newly implemented Global Ensemble Forecast System version 12 (GEFSv12) on September 23, 2020 and produced a 31-year (1989-2019) ensemble reforecast dataset to support stakeholders for sub-seasonal forecasts and hydrological applications. To provide seamless numerical guidance to... Show moreThe NOAA NCEP has newly implemented Global Ensemble Forecast System version 12 (GEFSv12) on September 23, 2020 and produced a 31-year (1989-2019) ensemble reforecast dataset to support stakeholders for sub-seasonal forecasts and hydrological applications. To provide seamless numerical guidance to a broad range of users and partners, the integration time of the GEFSv12 was extended from weeks 1 and 2 (medium range) to weeks 3 and 4 (sub-seasonal). The GEFSv12 reforecast products are based on NCEP’s Global Forecast System version 15.1 (GFSv15.1) configuration, and it uses the Finite Volume 3 (FV3) dynamical core. The resolution of the forecast system is ~25 km (C384 grid) with 64 vertical hybrid levels. The Climate Forecast System (CFS) reanalysis and GEFSv12 reanalysis serve as initial conditions for the periods (1989–1999) and (2000–2019) respectively. For the reforecast period 1989-1999, the initial perturbations were produced using breeding vectors and ensemble transforms with a rescaling (BV-ETR) technique while ensemble Kalman filter (EnKF) analysis from reanalysis (or Global Data Assimilation System (GDAS)) are used for the reforecast period 2000-2019. These reforecasts were initialized at 0000 UTC once per day out to 16 days with 5 ensemble members, except on Wednesdays when the integrations were extended to 35 days with 11 members. It is well known that the numerical weather models’ raw products are not skillful at extended lead times and required statistical post-processing to address the systematic errors to obtain more reliable and skillful forecast guidance. In this study, we proposed a neural network (NN)-based calibration method which can incorporate non-linear relationships between arbitrary predictor variables and forecast distribution parameters that are automatically learned in a data-driven way rather than requiring pre-specified link functions. The NN technique improved the GEFSv12 reforecast products of maximum temperature (Tmax) and precipitation on extended to sub-seasonal forecasts over CONUS in depicting weather and climate extremes (heat waves and rainfall extremes) for the reforecast period 1989-2019. This study will focus on comparing the proposed neural network approach with other traditional statistical post-processing methods. Show less