An evaluation of convection-allowing ensemble forecast sensitivity to initial conditions

Abstract

This dissertation aims to advance understanding of initial conditions (ICs) for convection-allowing ensembles (CAEs). To do so, experiments with 80-member limited-area ensemble Kalman filters (EnKFs) were performed over the entire conterminous United States for a 4-week period. The EnKF data assimilation systems differed in terms of their cycling strategies (continuous or partial cycling) and horizontal grid spacings (15- or 3-km horizontal grid spacing). EnKF analyses initialized 36-h, 3-km, 10-member CAE forecasts that were evaluated with a focus on precipitation, providing insights about CAE forecast sensitivity to ICs. Additionally, EnKF analyses were leveraged to isolate CAE forecast sensitivity to resolution of both IC perturbations and central initial states about which IC perturbations were centered. A “blending” approach was also used to produce new sets of CAE ICs by combining small scales from continuously cycling EnKF analyses with large scales from Global Ensemble Forecast System (GEFS) ICs using a low-pass filter.

Key results are as follows:• CAE forecasts initialized from continuously cycling 3-km EnKF analyses were more skillful and reliable than those initialized from downscaled GEFS and continuously cycling 15-km EnKF ICs through 12–18 and 6–12 h, respectively. Conversely, after 18 h, GEFS-initialized forecasts were better than forecasts initialized from continuously cycling EnKFs. Blended 3-km ICs led to ~18–36-h forecasts possessing comparable quality as GEFS-initialized forecasts while preserving short-term forecast benefits of unblended continuously cycling 3-km EnKF analyses. • Continuously cycling EnKF analyses initialized ~1–18-h forecasts that were comparable to or somewhat better than those with partial cycling EnKF ICs. Conversely, ~18–36-h forecasts with partial cycling EnKF ICs were comparable to or better than those with unblended continuously cycling EnKF ICs. However, blended ICs yielded ~18–36-h forecasts that were statistically indistinguishable from those with partial cycling ICs. • It is more important for central initial states than for IC perturbations to possess convection-allowing horizontal grid spacing for short-term CAE forecasting applications.

These collective findings have important implications for model developers working on next-generation CAEs and suggest paths toward potentially saving computing resources, streamlining processes for improving CAE ICs, and unifying short-term and next-day CAE forecasting systems.