Projecting changes in future heavy rainfall events for Oahu, Hawaii: A statistical downscaling approach

A statistical model based on nonlinear artificial neural networks is used to downscale daily extreme precipitation events in Oahu, Hawaii, from general circulation model (GCM) outputs and projected into the future. From a suite of GCMs and their emission scenarios, two tests recommended by the International Panel on Climate Change are conducted and the ECHAM5 A2 is selected as the most appropriate one for downscaling precipitation extremes for Oahu. The skill of the neural network model is highest in drier, leeward regions where orographic uplifting has less influence on daily extreme precipitation. The trained model is used with the ECHAM5 forced by emissions from the A2 scenario to simulate future daily precipitation on Oahu. A BCa bootstrap resampling method is used to provide 95% confidence intervals of the storm frequency and intensity for all three data sets (actual observations, downscaled GCM output from the present-day climate, and downscaled GCM output for future climate). Results suggest a tendency for increased frequency of heavy rainfall events but a decrease in rainfall intensity during the next 30 years (2011-2040) for the southern shoreline of Oahu.