Erin Jones - September 29

Weather and Climate Systems

Multi-Resolution and Multiscale Hybrid 4DEnVar for FV3GFS to Improve Global and Tropical Cyclone Numerical Prediction

Erin Jones

Wednesday, September 29th

3:00 pm

Join Google Meet:

https://meet.google.com/joz-jnzu-cgr

Or dial: (US) +1 314-649-9181 PIN: 293 300 648#

The genesis and intensification of several meteorological features, notably tropical cyclones and frontal systems, depends on the interaction between the feature and the larger-scale environmental flow.  Therefore, forecasting these phenomena relies on the proper estimation of multiple scales.  Creating accurate multiscale estimation is especially relevant and beneficial as the spatial and temporal resolution of numerical weather prediction (NWP) models increase and new observation platforms emerge.   This study utilizes the Finite-Volume Cubed Sphere Global Forecast System (FV3GFS), the United States’ Next Generation Global Prediction System, to investigate the impact of a new multi-resolution and multiscale four-dimensional ensemble variational (4DEnVar) data assimilation approach.  In an operational setting, computational constraints limit the size and resolution of the ensemble used to estimate the background forecast errors in ensemble-based data assimilation systems, though, ideally, a higher-resolution ensemble with a larger ensemble size would produce a more accurate forecast.

Prior research has indicated that utilizing a multi-resolution ensemble can improve a forecast when compared to a low-resolution ensemble while keeping the computational cost below that of a high-resolution ensemble of the same size.  This study further develops the multi-resolution ensemble 4DEnVar approach to include multiscale localization.  Two 4DEnVar experiments with similar computational costs are compared: one with an 80-member high-resolution background ensemble and the other with a 220-member multi-resolution background ensemble, including 180 low-resolution and 40 high-resolution members.  Preliminary results show that despite having similar costs, the 220-member multi-resolution 4DEnVar outperforms the 80-member high-resolution 4DEnVar globally.  Additionally, tropical cyclone forecast track errors are decreased for the multi-resolution 4DEnVar over the high-resolution 4DEnVar for forecast lead times of up to 96 hours. Diagnostics suggest the former can more properly correct a wider range of scales of the background errors.