Toward improving hurricane forecasts using the JPL Tropical Cyclone Information System (TCIS): A framework to address the issues of Big Data

Accurate forecasting of extreme weather requires the use of both regional models as well as global General Circulation Models (GCMs). The regional models have higher resolution and more accurate physics - two critical components needed for properly representing the key convective processes. GCMs, on the other hand, have better depiction of the large-scale environment and, thus, are necessary for properly capturing the important scale interactions. But how to evaluate the models, understand their shortcomings and improve them? Satellite observations can provide invaluable information. And this is where the issues of Big Data come: satellite observations are very complex and have large variety while model forecast are very voluminous. We are developing a system - TCIS - that addresses the issues of model evaluation and process understanding with the goal of improving the accuracy of hurricane forecasts. This NASA/ESTO/AIST-funded project aims at bringing satellite/airborne observations and model forecasts into a common system and developing on-line tools for joint analysis. To properly evaluate the models we go beyond the comparison of the geophysical fields. We input the model fields into instrument simulators (NEOS3, CRTM, etc.) and compute synthetic observations for a more direct comparison to the observed parameters. In this presentation we will start by describing the scientific questions. We will then outline our current framework to provide fusion of models and observations. Next, we will illustrate how the system can be used to evaluate several models (HWRF, GFS, ECMWF) by applying a couple of our analysis tools to several hurricanes observed during the 2013 season. Finally, we will outline our future plans. Our goal is to go beyond the image comparison and point-by-point statistics, by focusing instead on understanding multi-parameter correlations and providing robust statistics. By developing on-line analysis tools, our framework will allow for consistent model evaluation, providing results that are much more robust than those produced by case studies - the current paradigm imposed by the Big Data issues (voluminous data and incompatible analysis tools). We believe that this collaborative approach, with contributions of models, observations and analysis approaches used by the research and operational communities, will help untangle the complex interactions that lead to hurricane genesis and rapid intensity changes - two processes that still pose many unanswered questions. The developed framework for evaluation of the global models will also have implications for the improvement of the climate models, which output only a limited amount of information making it difficult to evaluate them. Our TCIS will help by investigating the GCMs under current weather scenarios and with much more detailed model output, making it possible to compare the models to multiple observed parameters to help narrow down the uncertainty in their performance. This knowledge could then be transferred to the climate models to lower the uncertainty in their predictions. The work described here was performed at the Jet Propulsion Laboratory, California Institute of Technology, under contract with the National Aeronautics and Space Administration.