A cloud classification method based on random forest for FY-4A. Issue 9 (3rd May 2021)

• Record Type: Journal Article
• Authors: Yu, Zhuofu; Ma, Shuo; Han, Ding; Li, Guanlin; Gao, Ding; Yan, Wei
• Is Part Of: International journal of remote sensing. Volume 42:Issue 9(2021)
• Page Start: 3353
• Page End: 3379
• Publication Date: 2021-05-03
• Subjects: Remote sensing -- Periodicals; Télédétection -- Périodiques; 621.3678
• DOI: 10.1080/01431161.2020.1871098
• Languages: English
• ISSNs: 0143-1161

Abstract

ABSTRACT: Cloud is complicated due to its various types and distribution on different layers. Various classification principles and overlapping of clouds on multiple layers make it difficult to classify clouds correctly. A cloud classification method based on random forest (RF) for FengYun-4A (FY-4A) is presented in this paper, aiming at obtaining cloud classification products with high temporal resolution and extensive coverage. The method classifies clouds into multi-layer clouds and 8 categories of single-layer clouds (Deep convective cloud, Nimbostratus, Cumulus, Stratocumulus, Stratus, Altocumulus, Altostratus and Cirrus). Additionally, multi-layer clouds are further classified into 12 categories of two-layer clouds (combinations of 8 single-layer clouds). CloudSat cloud classification products are used as the target class and to evaluate the method. In order to classify clouds correctly, it is necessary to capture cloud properties comprehensively. Thus we perform comparative pre-experiments to analyse the effects of FY-4A cloud products on cloud classification. It is demonstrated that clouds products can improve cloud classification. Therefore, cloud optical thickness (COT), cloud effective radius (CER) and cloud top height (CTH) are used as dataset together with multispectral data in cloud classification. Cloud classification models based on different algorithms and different channels combinations are compared. The results show that RF models perform better than K -Nearest Neighbour (KNN) models and Back Propagation Neural Network (BPNN) models in cloud classification, and models using all channels' data perform better than models using data of selected channels combination. The method can provide cloud types and distribution for a FY-4A scan full disk in low time cost. Especially, it gives more specific types for multi-layer clouds, which can provide reference for subsequent research on cloud classification and FY-4A satellite.