Improvements to the post-processing of weather forecasts using machine learning and feature selection

Kazuma Iwase

doi:10.1016/j.atmosres.2026.109037

Improvements to the post-processing of weather forecasts using machine learning and feature selection

2026-05-08

0

PRE

AI

K

Kazuma Iwase

T

Tomoyuki Takenawa *

DOI：10.1016/j.atmosres.2026.109037

Original

Original request for help

Save

Abstract

En 中文

• ML post-processing models are developed for three variables at 18 sites in Japan. • Surrounding-grid predictors improve accuracy by using spatial context. • Correlation-based feature selection removes redundant surrounding-grid inputs. • LightGBM gives lower RMSE than MSM, MSMG, and tested NN/CNN baselines. • Tweedie weighting improves threshold-based precipitation prediction.

Keywords:

Machine Learning

Weather Forecasting

Feature Selection

LightGBM

Precipitation Prediction

Improvements to the post-processing of weather forecasts using machine learning and feature selection

Abstract

Journal

Organization

Cited Papers

Citing Papers