Retrieval and application of cloud physical property parameters in Inner Mongolia based on Fengyun-4 Satellite and random forest

doi:10.11755/j.issn.1006-7639-2026-02-0264

Abstract

Abstract:

To meet the demand for high-precision monitoring of cloud physical characteristics in artificial weather modification operations in Inner Mongolia region, this study utilizes multi-source data from Fengyun-4A/B satellites (FY-4A/B), CloudSat/CALIPSO, Himawari-8, Moderate Resolution Imaging Spectroradiometer (MODIS), and the fifth-generation reanalysis data of the European Centre for Medium-Range Weather Forecasts (ERA5), and through preprocessing steps of radiometric calibration and geometric correction for FY-4 Satellite data, then applies artificial intelligence algorithms such as random forest to construct a cloud physical characteristic parameter inversion algorithm based on FY-4 Satellite. This algorithm achieves cloud detection and the retrieval of cloud top height, cloud top temperature, supercooled layer thickness, cloud optical thickness as well as cloud effective particle radius. Furthermore, the algorithm accuracy verification and adaptability analysis are conducted, and an operational platform for cloud parameter retrieval and a data release website are developed, forming a complete technical chain of “data-algorithm-platform-application”. The results show that the overall accuracy of the self-developed cloud detection algorithm is 90.07%, which is 1.11% higher than that of the official algorithm of the FY-4 Satellite; the determination coefficients (R2) of the inversion model for cloud top height and cloud top temperature are 0.928 and 0.922 respectively, and the root mean square errors are 0.901 km and 5.963 K respectively; the R2 of the inversion model for ice clouds and water clouds optical thickness are 0.693 and 0.582 respectively, and the R2 of the inversion of effective particle radius is 0.562 and 0.809, respectively.

Key words: FY-4 Satellite, cloud physical parameters, retrieval algorithm, random forest, Inner Mongolia

摘要：

为满足内蒙古地区人工影响天气业务对云物理特性高精度监测的需求，利用风云四号A/B星（FY-4A/B）、云卫星/云气溶胶激光雷达和红外探测者卫星（CloudSat/CALIPSO）、葵花-8号卫星（Himawari-8）、中分辨率成像光谱仪（Moderate Resolution Imaging Spectroradiometer，MODIS）及欧洲中期天气预报中心第五代再分析资料（ERA5）等多源数据，通过对FY-4卫星数据的辐射定标与几何纠正预处理，采用随机森林等人工智能方法，构建基于FY-4卫星的云物理特性参量反演模型，实现云检测及云顶高度、云顶温度、过冷层厚度、云光学厚度及云滴有效粒子半径的反演，并开展算法精度验证与适应性分析，开发了云参数反演运行平台与数据发布网站，形成“数据—算法—平台—应用”的完整技术链。结果表明：自研云检测算法总体精度达90.07%，较FY-4卫星官方算法提升1.11%；云顶高度与云顶温度反演结果的判定系数（R2）分别为0.928、0.922，均方根误差分别为0.901 km、5.963 K；冰云与水云光学厚度反演结果的R2分别为0.693、0.582，有效粒子半径反演的R2分别为0.562、0.809。

关键词: 风云四号卫星, 云物理参量, 反演算法, 随机森林, 内蒙古地区

CLC Number:

P407.8

ZHANG Jie, XU Zhili, BI Lige, GAO Jian, BAO Shanhu, SU Yue, ZHANG Wenbo. Retrieval and application of cloud physical property parameters in Inner Mongolia based on Fengyun-4 Satellite and random forest[J]. Journal of Arid Meteorology, 2026, 44(2): 264-272.

张杰, 许志丽, 毕力格, 高健, 包山虎, 苏玥, 张文博. 基于风云四号卫星与随机森林的内蒙古地区云物理特性参量反演及应用[J]. 干旱气象, 2026, 44(2): 264-272.

Figures/Tables 9

Fig.1 Scope of the study area

Tab.1 Data names and their attributes

数据名称	时空分辨率	时间范围	获取参数
FY-4A卫星L1数据	0.036°、15 min	2019年1月—2021年12月	14波段，几何观测信息云产品，用于数据对比
CloudSat/CALIPSO L2数据		2019年1月—2021年12月	云顶温度、云顶高度、云滴有效粒子半径
Himawari-8	0.05°、10 min	2019年1月—2021年12月	云光学厚度
MODIS MOD06/MYD06	0.05°	2019年1月—2021年12月	云参量，用于数据对比
土地利用（ERA5）	0.10°	2020年1月—2021年12月	土地类型
积雪数据（ERA5）	0.05°、逐月	2016年1月—2022年3月	积雪覆盖率
地表、海面温度（ERA5）	0.05°、逐月	2019年1月—12月	气温
SRTM高程数据	30 m		30 m分辨率高程信息，用于地形校正及高原地区大气路径辐射修正

Fig.2 The flowchart of cloud detection algorithm

Tab.2 Comparison of cloud detection results between the self-developed algorithm and the FY-4 official product over China

日期	自研算法平均	FY-4官方算法平均	精度差异（自研算法减FY-4官方算法）
2018年8月（夏）	92.61	90.92	1.69
2018年11月（秋）	87.99	87.43	0.56
2019年2月（冬）	88.68	88.30	0.38
2019年5月（春）	91.01	89.20	1.81
平均	90.07	88.96	1.11

Fig.3 The scatter plots of estimated cloud top height (a) and cloud top temperature (b) vs. CALIPSO Satellite measurements (Frequency represents the probability density of sample points occurring, reflecting the degree of concentration of data distribution)

Fig.4 The architecture diagram of the model

Fig.5 The cloud detection image (a) and the false color composite image (b) of FY-4A Satellite at 12:00 on July 1, 2019 (The area surrounded by the red line is Inner Mongolia，the same as below)

Fig.6 The retrieval results of cloud top height (a, Unit: km) and cloud top temperature (b, Unit: ℃) of FY-4A Satellite at 12:00 on July 1, 2019

Fig.7 The retrieval results of supercooled layer thickness (a, Unit: km), cloud optical thickness (b) and cloud effective particle radius (c, Unit: μm) of FY-4A Satellite at 12:00 on July 1, 2019

References 16

[1]	陈英英, 周毓荃, 毛节泰, 等, 2007. 利用FY-2C静止卫星资料反演云粒子有效半径的试验研究[J]. 气象, 33(4):29-34.
[2]	邓军, 白洁, 刘健文, 2006. 基于EOS/MODIS的云雾光学厚度和有效粒子半径反演研究[J]. 遥感技术与应用, 21(3): 220-226.
[3]	范思睿, 王维佳, 2018. 利用FY-4A卫星反演产品对飞机增雨作业的分析[J]. 高原山地气象研究, 38(4):60-66.
[4]	刘健, 董超华, 张文建, 2003. 利用FY-1C资料反演水云的光学厚度和粒子有效半径[J]. 红外与毫米波学报, 22(6):436-440.
[5]	叶晶, 李万彪, 严卫, 2009. 利用MODIS数据反演多层云光学厚度和有效粒子半径[J]. 气象学报, 67(4):613-622.
[6]	赵凤生, 丁强, 孙同明, 等, 2002. 利用NOAA-AVHRR观测数据反演云辐射特性的一种迭代方法[J]. 气象学报, 60(5):594-601.
[7]	周青, 赵凤生, 高文华, 2010. 利用FY-2C卫星数据反演云辐射特性[J]. 大气科学, 34(4):827-842.
[8]	FREY R A, ACKERMAN S A, LIU Y H, et al, 2008. Cloud detection with MODIS. Part I: Improvements in the MODIS cloud mask for collection 5[J]. Journal of Atmospheric and Oceanic Technology, 25(7):1057-1 072. DOI URL
[9]	MA Z S, LIU Q J, ZHAO C F, et al, 2018. Application and evaluation of an explicit prognostic cloud-cover scheme in GRAPES global forecast system[J]. Journal of Advances in Modeling Earth Systems, 10(3): 652-667. DOI URL
[10]	PILEWSKIE P, TWOMEY S, 1987. Cloud phase discrimination by reflectance measurements near 1.6 and 2.2 μm[J]. Journal of the Atmospheric Sciences, 44(22):3419-3 420.
[11]	PLATNICK S, KING M D, ACKERMAN S A, et al, 2003. The MODIS cloud products: Algorithms and examples from Terra[J]. IEEE Transactions on Geoscience and Remote Sensing, 41(2):459-473. DOI URL
[12]	RIEDI J, MARCHANT B, PLATNICK S, et al, 2010. Cloud thermodynamic phase inferred from merged POLDER and MODIS data[J]. Atmospheric Chemistry and Physics, 10(23): 11 851-11 865.
[13]	SHANG H Z, LETU H, CHEN L F, et al, 2020. Cloud thermodynamic phase detection using a directional polarimetric camera (DPC)[J]. Journal of Quantitative Spectroscopy and Radiative Transfer, 253:10717. DOI:10.1016/j.jqsrt.2020.107179.
[14]	STEPHENS G L, 2005. Cloud feedbacks in the climate system: A critical review[J]. Journal of Climate, 18(2): 237-273. DOI URL
[15]	STOWE L L, DAVIS P A, MCCLAIN E P, 1999. Scientific basis and initial evaluation of the CLAVR-1 global clear/cloud classification algorithm for the advanced very high resolution radiometer[J]. Journal of Atmospheric and Oceanic Technology, 16(6):656-681. DOI URL
[16]	XIE S C, LIU X H, ZHAO C F, et al, 2013. Sensitivity of CAM5-simulated Arctic clouds and radiation to ice nucleation parameterization[J]. Journal of Climate, 26(16): 5 981-5 999.