Evaluation of Uniformly High-Resolution Hurricane Forecasts Using NMM-B

Operational numerical weather prediction (NWP) models for hurricane forecasting currently rely on nesting, wherein small high-resolution grids are placed inside of relatively coarse-resolution large scale grids. Nests are particularly useful for hurricane forecasting since nests can be placed around storms to resolve storm-scale features at high resolution. Nesting is a pragmatic way of optimizing forecast skill while adhering to strict timeliness requirements for hurricane forecasts. Improvements in forecasting skill in operational models over the past several years are partially attributable to the high resolutions made possible by nesting, but it is still of interest to evaluate the negative impacts of using nests as opposed to running a single large high-resolution grid. In doing so, we may be able to improve nesting. In this work, we discuss our efforts to evaluate the impact of nesting using the Non-hydrostatic Multiscale Model on the B-Grid (NMM-B). NMM-B is poised to become an operational model in the future. As such, many aspects of the successful Hurricane Weather Research and Forecasting (HWRF) model have been ported to it. Hence, in addition to nested NMM-B configurations, we use HWRF as a baseline for comparison for this work. Our work also serves as the basis for a regional nature run that will be based off of NASA's G5NR global nature run.

In this presentation, we will summarize the steps taken to run forecasts on a non-nested, uniform 3km grid spanning over 60 degrees in length and 140 degrees in width. We provide an overview of the modifications made to the modeling system to make the experiments possible and we present the results obtained from the cases run. We then give a basic evaluation of the regional nature run created using the same model configuration.

An objective of this work is to measure the degradation of forecast track and intensity prediction accuracy due to nesting. We do this by studying major hurricane cases. For example, we study hurricane Earl from 2010. The HWRF model has generally performed well at predicting hurricane Earl. We compare the performance to nested and uniform 3km NMM-B forecasts and note a significant improvement in forecast track. We also use hurricane Joaquin from 2015 in our evaluation. This proved to be a difficult system for models to predict. Specifically, HWRF, GFS, GFDL, and our nested NMM-B model all predicted that the system would make landfall. In reality, the system stayed out at sea. Again, the uniform 3km configuration showed notable improvements in track. These results present an opportunity to improve the model by determining what caused the eventual shift towards land and to specifically improve nesting by determining what caused the nested forecasts with NMM-B to perform worse than their uniform 3km counterparts.

These experiments were motivated by the need to create a regional “nature run” off of NASA's G5NR global nature run, in order to run observing system simulation experiments (OSSEs). The nature run serves as the “truth” for OSSEs, so it should use a state of the art modeling system. The results obtained from the uniform-3km experiments suggest that it is an ideal platform for the nature run. To address the “identical/fraternal twin” problem, we use a different physics package to run the nature run. We also evaluate real cases with this configuration and compare results with the HWRF physics configuration.