Electrical and Mechanical College, Hainan University, Haikou 570228, China.
Hainan Meteorological Observatory, Haikou 570203, China.
Sensors (Basel). 2021 Dec 28;22(1):193. doi: 10.3390/s22010193.
Achieving high-performance numerical weather prediction (NWP) is important for people's livelihoods and for socioeconomic development. However, NWP is obtained by solving differential equations with globally observed data without capturing enough local and spatial information at the observed station. To improve the forecasting performance, we propose a novel spatial lightGBM (Light Gradient Boosting Machine) model to correct the numerical forecast results at each observation station. By capturing the local spatial information of stations and using a single-station single-time strategy, the proposed method can incorporate the observed data and model data to achieve high-performance correction of medium-range predictions. Experimental results for temperature and wind prediction in Hainan Province show that the proposed correction method performs well compared with the ECWMF model and outperforms other competing methods.
实现高性能的数值天气预报（NWP）对民生和社会经济发展都很重要。然而，NWP 是通过求解具有全球观测数据的微分方程得到的，无法在观测站获取足够的局部和空间信息。为了提高预测性能，我们提出了一种新颖的空间 LightGBM（Light Gradient Boosting Machine）模型，以校正每个观测站的数值预报结果。通过捕获站点的局部空间信息并使用单站单时策略，该方法可以将观测数据和模型数据结合起来，实现对中程预测的高性能校正。针对海南省温度和风的预测实验结果表明，与 ECMWF 模型相比，所提出的校正方法表现良好，优于其他竞争方法。
