Abstract
<title>Abstract</title> <p>Super-resolution (SR), a technique for estimating a high-resolution (HR) image from a low-resolution image, has been used in meteorology for downscaling and resolution enhancement of observations. SR Gaussian process regression with a steering kernel (SRGP-SK) generates more accurate HR images than the original SR Gaussian process regression (SRGP), but it has not yet been applied to meteorological data. This study applied SRGP-SK to radar/rain gauge–analyzed precipitation for convective and stratiform cases and evaluated the results using the structural similarity index (SSIM) and radially averaged power spectral density (PSD). SRGP-SK achieved SSIM values comparable to those of bicubic interpolation and higher than those of SRGP while reconstructing finer precipitation structures; it recovered variations down to a wavelength of 6 km, compared with 8 km for bicubic interpolation. This difference may correspond to an approximately threefold increase in the number of convective cells. Among the kernel functions compared, the Matérn 5/2 kernel yielded the highest geometric mean PSD ratio. Further investigation of this point could identify the statistical scaling characteristics of the precipitation field. This study evaluates only two precipitation cases; examining more cases is necessary before SRGP-SK can be applied more broadly.</p>