A New Nonparametric Method to Correct Model Data: Application to Significant Wave Height from the ERA-40 Re-Analysis

Abstract

Abstract A new nonparametric method to correct model data is proposed. At any given point in space and time the correction is determined from “analogs” in a learning dataset. The learning dataset contains model data and simultaneous observations. The method is applied to the significant wave height dataset of the 45-yr European Centre for Medium-Range Weather Forecasts Re-Analysis (ERA-40). Comparison of the corrected data with significant wave height measurements from in situ buoy and global altimeter data shows clear improvements in bias, scatter, and quantiles in the whole range of values. Temporal inhomogeneities are also removed.