UTILIZING BAYESIAN NEURAL NETWORKS TO MODEL THE OCEANATMOSPHERE INTERFACE.

The ocean-atmosphere interface (OAI) is a dynamic boundary of complex energy and chemical exchange and is important to understand mechanisms that influence it. Research is on-going to improve how the OAI is represented within tropical cyclone (TC) prediction models and ensembles. Motivation for improvement stems from a rapidly changing thermodynamic environment caused by climate change. Such changes are not widely understood, as no scientist has observed or measured these changes on long time scales. We assert the possibility of climate change, its underlying uncertainties and modified atmospheric variability can potentially lead to rapid intensification. We argue simplification of OAI to capture model ensemble data uncertainty through probabilistic modeling via Bayesian Neural Network (BNN). We retrieved area- averaged satellite data from NOAA and NASA, created a data set of several parameters-atmospheric air temperature (AirTemp), atmospheric temperature anomaly (ATA), atmospheric carbon dioxide (CO₂), sea surface temperature (SST), tropical cyclone heat potential (TCHP), mid-layer wind shear (WindShear), convective available heat potential (CAPE), vertical motions (VerticalMotion), precipitable water content (PWC) and our derived OAI parameter as inputs into a BNN via R programming language. We used the BNN to model the OAI and inferenced potential favorability of an OAI given conditional probabilities. The BNN network rejected ATA and WindShear. Results showed probabilities acceptable within expert interpretations of parameter interactions to predict favorable OAI conditions. [ABSTRACT FROM AUTHOR]