.png){kind=icon}
.png){kind=icon}
Return
Developing a machine learning-based model for surface wind speed retrieval based on FY4B-GIIRS satellite data
.png){kind=icon}
PRE
AI
DOI：10.1080/01431161.2025.2549534.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
This study utilizes high-resolution hyperspectral infrared radiance level 1 (L1) data to improve wind speed retrieval accuracy and temporal resolution over a region in southeast China. The methodology presented herein involves several steps including, data pre-processing, model selection, hyperparameter tuning, training, and evaluation. Radiance data from Geostationary Interferometric Infrared Sounder (GIIRS) was normalized, and key features, such as radiance ratios, latitude, longitude, elevation, and brightness temperature (calculated from radiance) were extracted to serve as inputs for several Machine Learning (ML) models. Only clear or near clear sky points were considered in the retrieval algorithm during August 8-11, 2022. The target variable, wind speed, was derived from ERA5 reanalysis data to ensure consistency. Multiple ML algorithms, including Random Forest (RF), Gradient Boosting (GB), Adaptive Boosting (Adab), and Multi-Layer Perceptron Artificial Neural Network (MLP-ANN), were systematically evaluated, with hyperparameter tuning performed via Grid Search. Among the tested models, RF demonstrated superior performance, achieving a Root Mean Square Error (RMSE) of 0.459 and a correlation coefficient (R) of 0.895 on the test dataset. After RF, GB had the second rank with RMSE of 0.529 and correlation coefficient of 0.85. ANN and Adab ranked next, respectively. Beyond its accuracy, a key advantage of this approach is its ability to provide timely access to L1 GIIRS data, which can be valuable for wind speed estimation.
Keywords:
Surface wind speed
FY-4B
GIIRS
ERA5
hyperspectral infrared radiance
machine learning
Random Forest
Journal
IF:
2.6
Papers:
1.2W
Citations:
2.7W