Deriving meteorological variables from numerical weather prediction model output: A nearest neighbor approach

Abstract

This paper presents the application of variations in a nearest neighbor resampler approach for generating local‐scale meteorological variables from numerical weather prediction model output. On the basis of measure of closeness and sampling strategy, six nearest neighbor models were designed. The proposed models were applied to downscale station daily precipitation and minimum and maximum temperature fields for the Chute‐du‐Diable meteorological station in northeastern Canada. Suites of deterministic diagnostic measures were employed for evaluating individual models as well as for intercomparison among the downscaling models. On the basis of intercomparison among models a relatively better nearest neighbor resampler was identified and the subsequent model was further investigated with a focus on downscaling daily precipitation. Suites of conventional and distribution‐based diagnostic measures were employed for evaluating the skill of the downscaled precipitation over the raw numerical model output. The comparative results showed that the downscaled precipitation had greater skill values based on different performance measures which include median bias, Brier skill score, ranked probability skill score, discrimination, reliability, and relative operating characteristics.