Development and Application of K-means Ensemble Prediction Product Based on GRAPES-Global Ensemble Prediction System
- Article
- Figures
- Metrics
- Preview PDF
- Reference
- Related
- Cited by
- Materials
Abstract:
Based on the GRAPESGlobal Ensemble Forecast System (GRAPESGEPS), and the nationwide coldwave process from 13 to 16 February 2020, the Kmeans cluster products are developed. In this paper, the Sum of the Squared Errors (SSE) criterion function is applied to determine the most appropriate clustering numbers and the Kmeans cluster algorithm is used to classify the ensemble samples. Results indicate that, all types of Kmeans cluster products related to the 500 hPa geopotential height present the Ωshaped circulation situation and the cold advection situation behind the lowpressure system. In addition, Type 1 clustering products with the highest probability reflect the observed circulation situation most efficiently. For 850 hPa temperature, all categories can present the spatial characteristics of 850 hPa temperature, which increase gradually from North China to South China. In addition, Type 1 clustering products with the highest probability can reflect the spatial distribution of 850 hPa temperature and possess the least errors related to the observation. For 10 m wind speed clustering products, at higher wind speeds, the dispersion of the aggregate samples is larger, and the wind speeds of different kinds have significant differences. The Type 1 clustering products with the highest probability can reflect the spatial distribution and intensity of 10 m wind speed in Tianjin and its surrounding areas exactly and provide valuable prediction information for forecasters. With Kmeans cluster results, we can realize the aggregation of forecast sample information and provide the intuitive guidance of weather prediction for the forecasters.
Keywords:
Project Supported:
Clc Number:
You are thevisitors
Supported by:Beijing E-Tiller Technology Development Co., Ltd.
Supported by:Beijing E-Tiller Technology Development Co., Ltd.