Toward improved regional rainfall modeling in Hawai‘i: Lessons learned from a multi-method investigation

Abstract

Reliable regional rainfall projections are critical for effective climate planning in Hawaiʻi, where the combination of complex topography and highly variable climate presents significant challenges for modeling efforts. Unlike the continental United States, Hawaiʻi lacks coordinated, large-scale downscaling efforts, resulting in limited and inconsistent downscaled climate information. This dissertation addresses this gap by evaluating and comparing multiple downscaling approaches to better understand how methodological choices influence future rainfall projections. In doing so, it aims to both improve future downscaling practices and contribute new high-resolution precipitation datasets to the growing suite of climate resources available for the Hawaiian Islands.Chapter 2 uses a self-organizing maps (SOM) framework to classify wet-season circulation patterns and examine how their frequencies may shift under CMIP6 projections. This method links large-scale atmospheric variability to local rainfall and generates a new set of end-of-century precipitation projections (2070–2100). Chapter 3 develops two additional projection sets using multiple linear regression (MLR) and generalized additive models (GAM) with CMIP5 RCP4.5 data, highlighting how differences in statistical techniques affect spatial and seasonal rainfall outcomes. Chapter 4 evaluates the Intermediate Complexity Atmospheric Research (ICAR) model as a simplified dynamical downscaling tool, testing its performance using wind inputs from ERA5, WRF, and SOM-based reconstructions. Together, these studies demonstrate the critical role of method selection in shaping downscaled projections and emphasize the need for transparency and evaluation in regional climate modeling. By contributing new rainfall projections and comparative insights, this work supports more robust and regionally appropriate climate information for resilience planning in Hawaiʻi.