Performance of reanalysis and mesoscale models for wind resource assessment off the coast of Hawaii

Abstract. The eastern Hawaii coast in the United States is characterized by considerable wind resource fuelled by persistent trade winds, making it an important area for wind energy research. The need is strong for reanalyses, higher-resolution regional simulations, and purpose-built wind datasets in locations where observations have been historically limited, such as Hawaii's offshore environments. However, studies using offshore observations in other parts of the world have shown that significant errors can occur in reanalyses and wind datasets, which can lead to inaccurate estimates of wind energy generation, payback periods, and extreme weather risks at project locations. The degree of such errors is influenced by a number of factors, including spatial resolution and the handling of processes within the planetary boundary layer (PBL). In this work, we provide a wind resource characterization from year-long lidar buoy measurements off the eastern coast of Oahu, Hawaii, an environment previously unobserved at the rotor level, and use it to establish the performance of four simulation datasets with distinct spatial resolutions and PBL representations. The Oahu deployment location is meteorologically unique and less complex compared with previous offshore wind performance study locations, being strongly characterized by the trade winds with minimal land-atmosphere interaction influences. Despite the unique and consistent conditions, we hypothesize that distinct simulation datasets will exhibit diverse ranges of errors similar to what has been seen for other offshore locations. We find the European Centre for Medium-Range Weather Forecasts (ECMWF) Reanalysis version 5 (ERA5) to strongly underestimate observed wind speeds at the Oahu location (bias = -1.53 m s^-1 at a hub height of 140 m), while a regional Weather Research and Forecasting (WRF) Model simulation produced by the University of Hawaii (UH-WRF) provides a significantly smaller wind speed bias (-0.25 m s^-1), highlighting the value of running regional, higher-resolution simulations. Despite not temporally overlapping with the Oahu deployment, the long-term annual average 140 m wind speeds from the 2023 National Offshore Wind dataset (NOW-23) and Global Wind Atlas version 3 (GWA3) produce smaller magnitude biases (+0.39 m s^-1 and -0.10 m s^-1, respectively) than ERA5. The large bias noted for ERA5 is driven by significant underestimation of fast wind speeds, which the study site is largely characterized by, along with discontinuities in the ERA5 diurnal cycle. We also speculate that the relative sparsity of observations for data assimilation in this remote part of the world could influence the performance of ERA5.