Using the Ensemble Kalman Filter on Fire Weather Days to Explore Model Error over the Northeast United States

The Ensemble Kalman Filter (EnKF) is a popular data assimilation method for improving model initial conditions or exploring model error. One such unexplored pathway is during fire weather, since these days typically exhibit less complex meteorological conditions with fewer clouds and a deeper planetary boundary layer (PBL). However, FWI days over the Northeast United States (NEUS) exhibit cooler and wetter biases in the PBL compared to climatology using the National Centers for Environmental Prediction (NCEP) Short Range Ensemble Forecast (SREF). These biases can result to underestimated fire threat when using forecast ensembles like the SREF. For this reason, an EnKF is used to both correct and explore potential biases and error in the modeled PBL over the Northeast United States (NEUS).

The data assimilation platform comes from the Pennsylvania State University Ensemble Kalman Filter (PSU-EnKF) Version 4.0 with forward iterations performed using the Weather Research and Forecasting (WRF) Advanced Research WRF (ARW) core Version 3.5. This setup includes an outer domain at 27 km grid spacing covering the eastern two-thirds of the United States and an inner domain at 9 km grid spacing covering the NEUS. Fire weather days are quantified using a fire weather index (FWI), which consists of a binomial logistic regression trained on near-surface weather variables to determine the probability of fire occurrence.

Two 45-member PSU-EnKF runs are conducted throughout April of 2012 during a period of drought and abundant FWI days. These runs include one with no parameter estimation and one with Simultaneous State and Parameter Estimation (SSPE) of physically relevant parameters within the ACM2 PBL scheme. These estimated parameters are associated with minimum allowed mixing and governing mixing profile in the daytime PBL, as recommended by Nielsen-Gammon et al. (2010). Results are presented that show the rapid growth of typical FWI model biases after data assimilation and the impact of SSPE on model bias and error.