Abstract

Immense resources and community focus have been devoted to improving predictions of tropical cyclone (TC) rapid intensification (RI) over the past decade, yet a forecast technique which can both accurately and reliably predict RI remains elusive. Nevertheless, considerable progress has occurred a... Show moreImmense resources and community focus have been devoted to improving predictions of tropical cyclone (TC) rapid intensification (RI) over the past decade, yet a forecast technique which can both accurately and reliably predict RI remains elusive. Nevertheless, considerable progress has occurred across a number of fronts. Significant improvements have been made to limited area 3-d full physics models, such as the Hurricane Weather Research and Forecasting System (HWRF) and the Coupled Ocean/Atmosphere Mesoscale Prediction System for Tropical Cyclones (COAMPS®-TC) models. Although such models exhibit a relatively low probability of detection (POD) and high false alarm rate (FAR), these models can now occasionally predict RI. This is a definite improvement from the situation a decade ago. Meanwhile, the forecasts of global ensembles have continued to improve and now show some skill for intensity, whereas their intensity skill was quite poor a decade ago. The global ensembles demonstrate more tangible improvements their forecasts of the TC environment, as shown by substantial continued improvements in track skill. Finally, statistical-dynamical approaches and other data mining approaches along the vein of post-processing have leveraged the improvements in global and regional models; these have shown some promise for skillful RI forecasts. Yet, the grand challenge of obtaining an RI prediction technique that is both reliable and accurate remains. This talk will report on the development and evaluation of new frameworks for predicting extreme RI. The two-year effort, funded by the Hurricane Forecast Improvement Program, seeks to develop advanced forecast aids for RI through several novel approaches. One approach leverages the uncertainty information provided by global ensemble model tracks and fields to simulate a very large set of synthetic forecasts, resulting in probabilistic forecasts for intensity and the point-wise winds (see Lin et al., this conference). Another approach applies the machine learning technique of Convolutional Neural Networks (CNNs) to the forecast fields of HWRF to try to “learn” what a rapid intensifying storm “looks like” and use such information in an advanced RI forecast aid. A third approach has been to improve the Coupled Hurricane Intensity Prediction Scheme (CHIPS) by adding the effects of outflow self-stratification and real-time ocean analyses. A fourth approach has applied the logistic regression technique to HWRF output to produce probabilistic forecasts for RI (HLOG - see MacDaniel et al., this conference). A fifth approach uses vortex wind structure information from the Extended Flight Level Dataset for Tropical Cyclones (FLIGHT+) to predict intensification rates (see Hendricks et al., this conference). The results of each of these activities will be briefly reviewed. The balance of the talk will be devoted to describing a new generalized prediction framework that seeks to combine the best attributes of multiple forecast input sources. The philosophy behind the new approach, called the General Rapid Intensification Prediction (GRIP; no relation to the NASA field campaign of the same acronym) framework, is to combine the proven general intensity forecast skill of existing 3-d full physics models with the substantial skill in predicting RI provided by probabilistic forecast techniques. The goal of the framework is to obtain deterministic forecasts during RI periods that are both more accurate and more reliable than the 3-d full physics models can provide, without sacrificing any utility to forecasters during periods when RI is not anticipated. The initial version of this framework involves a simple linear blending between the intensity forecasts of a selected 3-d full physics model (e.g., HWRF) and CHP6, an ensemble member of CHIPS that has demonstrated a remarkable capability to sometimes capture the upper bound of intensity during extreme RI events. The blending weight for these two deterministic components is obtained simply as the probability of RI as predicted by a probabilistic RI prediction tool, such as Statistical Hurricane Intensity Prediction Scheme Rapid Intensity Index (SHIPS-RII) or HLOG. When environmental conditions favor a high chance of RI, the output of GRIP will be a blended intensity forecast that is closer to the CHP6 intensity prediction. When the predicted probability of RI is low, GRIP’s predicted intensity should be near to the HWRF intensity prediction. This prediction framework can be generalized to consist of any skillful base deterministic intensity model (e.g., HWRF, COAMPS-TC, or even a consensus of such models), an “upper bound” intensity model (e.g., CHP6, another empirically-based limit model, or a theory-based limit model), and a probabilistic RI model (e.g., HLOG, SHIPS-RII, etc.) to determine the blending fraction. We posit that this approach will provide improved skill in intensity forecasts during RI periods while preserving the substantial skill of the 3-d full physics model(s) during non-RI periods. We will present evaluation results of an initial version of GRIP. If time permits, we will also evaluate the results of an extended version of the framework which incorporates forecast information from one or more global ensemble prediction systems. Show less