地球科学进展 ›› 2026, Vol. 41 ›› Issue (1): 49 -60. doi: 10.11867/j.issn.1001-8166.2026.010

研究论文 上一篇    下一篇

基于深度学习和小波变换的被动微波遥感雪深反演
赵欣冉1,2(), 尤元红1,2(), 黄春林3,4, 唐志光5, 汪左1,2, 侯金亮3   
  1. 1.安徽师范大学 地理与旅游学院,安徽 芜湖 241002
    2.资源环境与地理信息工程 安徽省工程技术 研究中心,安徽 芜湖 241002
    3.中国科学院西北生态环境资源研究院,甘肃省遥感重点实验室,甘肃 兰州 730000
    4.兰州交通大学 测绘与地理信息学院,甘肃 兰州 730000
    5.湖南科技大学 地理空间信息技术国家地方联合工程实验室,湖南 湘潭 411201
  • 收稿日期:2025-05-26 修回日期:2025-12-19 出版日期:2026-01-10
  • 通讯作者: 尤元红 E-mail:xinranzhao@anhu.edu.cn;xinranzhao@ahnu.edu.cn;youyh@ahnu.edu.cn
  • 基金资助:
    国家自然科学基金项目(42130113);国家自然科学基金项目(42371398);测绘遥感信息工程湖南省重点实验室开放基金项目(E22405)

Microwave Remote Sensing of Snow Depth Based on Deep Learning Models and Wavelet Transform

Xinran Zhao1,2(), Yuanhong You1,2(), Chunlin Huang3,4, Zhiguang Tang5, Zuo Wang1,2, Jinliang Hou3   

  1. 1.School of Geography and Tourism, Anhui Normal University, Wuhu Anhui 241002, China
    2.Anhui Provincial Engineering Research Center for Resources, Environment and Geographic Information Systems, Wuhu Anhui 241002, China
    3.Key Laboratory of Remote Sensing of Gansu Province, Northwest Institute of Eco-Environment and Resources, Chinese Academy of Sciences, Lanzhou 730000, China
    4.School of Geomatics and Geoinformation, Lanzhou Jiaotong University, Lanzhou 730000, China
    5.National-Local Joint Engineering Laboratory of Geospatial Information Technology, Hunan University of Science and Technology, Xiangtan Hunan 411201, China
  • Received:2025-05-26 Revised:2025-12-19 Online:2026-01-10 Published:2026-03-10
  • Contact: Yuanhong You E-mail:xinranzhao@anhu.edu.cn;xinranzhao@ahnu.edu.cn;youyh@ahnu.edu.cn
  • About author:Zhao Xinran, research areas include snow depth retrieval, deep learning, and multi-source remote sensing data assimilation. E-mail: xinranzhao@anhu.edu.cn
  • Supported by:
    the National Natural Science Foundation of China(42130113);The Open Fund of Hunan Provincial Key Laboratory of Geo-Information Engineering in Surveying, Remote Sensing and Information Engineering(E22405)

微波亮温与雪深之间的关系十分复杂,地形和积雪状态的多变性对其影响极为显著,传统方法难以精确反演。为有效刻画微波亮温与雪深之间的复杂非线性关系,提出了一种融合小波变换与残差卷积神经网络的被动微波雪深反演模型,利用小波变换提取多频微波亮温的多尺度特征,以表征雪深的整体变化与局地扰动,并通过残差结构提升深层特征传递与非线性拟合能力。在此基础上,选取北半球不同积雪气候类型下的代表性站点,评估了该模型的适用性。结果表明,融合小波变换与残差卷积神经网络模型在站点尺度实验中,雪深反演的均方根误差为0.23 m,相关系数为0.93,平均偏差为0.01 m,纳什效率系数为0.86,各项误差统计指标相较于传统卷积神经网络和残差卷积神经网络模型均有较大改善。不同积雪深度下的性能均优于传统深度学习模型,且该模型在积雪较厚和变化较快时能够表现出更准确的估算能力。此外,该模型在不同积雪气候类型下的时序响应能力和空间迁移性能均优于现有主流雪深反演产品,可以为不同积雪气候类型下的雪深监测提供参考依据。

Accurate retrieval of snow depth from passive microwave observations remains a major challenge because of the pronounced nonlinearity between microwave brightness temperature and snow physical properties, which is further modulated by complex terrain and the temporal evolution of snowpacks. Conventional retrieval approaches, including physically based radiative transfer models, are often constrained by high computational costs and strong sensitivity to uncertain input parameters, resulting in degraded performance under heterogeneous snow conditions and across different snow climate regimes. To overcome these limitations, this study develops a passive microwave snow depth retrieval model that integrates Wavelet transform with a Residual Convolutional Neural Network (Wavelet-ResNet-CNN) to better represent multi-scale features and complex nonlinear relationships. Specifically, the wavelet transform is applied to multi-frequency brightness temperature observations to extract scale-dependent information, allowing the model to capture both large-scale snow depth variability and localized perturbations associated with rapid snow accumulation and melt processes. The resulting multi-scale features are then ingested into a residual convolutional neural network, in which residual connections facilitate deep feature propagation and enhance nonlinear fitting capability while mitigating performance degradation in deep architectures. The proposed model is evaluated using representative ground-based stations spanning different snow climate classes in the Northern Hemisphere to assess its applicability under diverse snow conditions at the site scale. Model performance is quantified using multiple statistical metrics, including Root Mean Square Error (RMSE), correlation coefficient (R), mean bias, and Nash-Sutcliffe Efficiency (NSE), and is benchmarked against conventional convolutional neural network and residual convolutional neural network models. The results show that the Wavelet-ResNet-CNN achieves an RMSE of 0.23 m, an R of 0.93, a mean bias of 0.01 m, and an NSE of 0.86, indicating consistent improvements over the reference deep learning models across all evaluation metrics. Additional analyses demonstrate that the proposed model outperforms traditional deep learning approaches across a wide range of snow depths, with particularly notable gains during periods of deep snow accumulation and rapid snow depth changes. Furthermore, compared with existing snow depth retrieval products, the model exhibits enhanced temporal responsiveness and improved spatial transferability across different snow climate types, highlighting its potential for robust passive microwave snow depth monitoring in regions with diverse snow regimes.

中图分类号: 

图1 不同积雪气候条件类型下的积雪观测站点分布图
Fig. 1 Distribution of snow observation stations across different snow climate types
表1 不同积雪气候条件类型下的积雪站点的主要信息
Table 1 Main information of each snow observation station under different types of snow climatic conditions
图2 Wavelet-ResNet-CNN网络结构图
Fig. 2 Structure of the Wavelet-ResNet-CNN Net
表2 不同特征因子输入方案的性能指标比较
Table 2 Performance comparison of different input feature combinations
图3 不同输入因子组合对雪深估算性能的影响
M1:多频被动微波垂直极化亮温数据;M2:多频被动微波垂直极化亮温数据+纬度+经度;M3:多频被动微波垂直极化亮温数据+高程+坡度+坡向+地表粗糙度;M4:多频被动微波垂直极化亮温数据+纬度+经度+高程+坡度+坡向+地表粗糙度。
Fig. 3 Influence of different input factor combinations on snow depth estimation performance
M1: Multi-frequency passive microwave vertically polarized brightness temperature data; M2: Multi-frequency passive microwave vertically polarized brightness temperature data, latitude, and longitude; M3: Multi-frequency passive microwave vertically polarized brightness temperature data, elevation, slope, aspect, and surface roughness; M4: Multi-frequency passive microwave vertically polarized brightness temperature data, latitude, longitude, elevation, slope, aspect, and surface roughness.
图4 不同深度学习模型在雪深估计任务中的性能对比(包括误差分布特征与拟合关系)
(a) CNN模型;(b) ResNet-CNN模型;(c) Wavelet-CNN模型;(d) Wavelet-ResNet-CNN模型。小提琴图表示不同雪深分段下的模拟误差分布,灰色虚线表示零误差参考线,小提琴内部实线分别表示误差分布的中位数及四分位距(IQR)。
Fig. 4 Performance comparison of different deep learning models in snow depth estimationincluding error statistics and fitting accuracy
(a) CNN; (b) ResNet-CNN; (c) Wavelet-CNN; (d) Wavelet-ResNet-CNN. The violin plots illustrate the distribution of simulation errors across different snow depth intervals. The gray dashed line denotes the zero-error reference, and the solid lines inside the violins represent the median and Interquartile Range (IQR) of the error distribution.
图5 不同模型在雪深估计中的分析
Fig. 5 Analysis of different models in snow depth estimation
图6 不同深度学习模型在典型站点的雪深季节性模拟对比
(a) SASP站;(b) SDA站; (c) WFJ站
Fig. 6 Seasonal snow depth estimation of different deep learning models at typical sites
图7 4种模型在不同积雪气候类型站点下的雪深估算误差比较
1~4列为各模型在SASP站的误差,5~8列为SDA站的误差,9~12列为WFJ站的误差;从左往右分别为CNN、Wavelet-CNN、ResNet-CNN和Wavelet-ResNet-CNN。
Fig. 7 Comparison of snow depth estimation errors of four models at stations with different snow-climate
Columns 1~4 show the errors of each model at the SASP station, columns 5~8 at the SDA station, and columns 9~12 at the WFJ station. From left to right, the models are CNN, Wavelet-CNN, ResNet-CNN, and Wavelet-ResNet-CNN.
图8 Wavelet-ResNet-CNN模型雪深估算结果与站点观测和雪深遥感产品的比较
Fig. 8 Comparison of snow depth estimates from the Wavelet-ResNet-CNN model with in-situ observations and satellite-based snow depth products
[1] Wang Zekun, Gan Fuping, Yan Baikun, et al. Inversion of snow depth and snow water equivalent based on passive microwave remote sensing and its application progress[J]. Remote Sensing for Natural Resources202234(3): 1-9.
王泽坤, 甘甫平, 闫柏琨, 等. 雪深和雪水当量抑制微波反演及进展应用[J]. 自然资源遥感202234(3): 1-9.
[2] Bormann K J, Brown R D, Derksen C, et al. Estimating snow-cover trends from space[J]. Nature Climate Change20188(11): 924-928.
[3] Cohen J, Rind D. The effect of snow cover on the climate[J]. Journal of Climate19914(7): 689-706.
[4] Hansen J, Nazarenko L. Soot climate forcing via snow and ice albedos[J]. Proceedings of the National Academy of Sciences of the United States of America2003101(2): 423-428.
[5] Chen Y N, Li B F, Fan Y T, et al. Hydrological and water cycle processes of inland river basins in the arid region of Northwest China[J]. Journal of Arid Land201911(2): 161-179.
[6] Bo Yanchen, Feng Xuezhi, LI Xin, et al. The retrieval of snow depth in Qinghai Xizang (Tibet) Plateau from passive microwave remote sensing data and its results assessment[J]. Journal of Remote Sensing2001(3): 161-165.
柏延臣, 冯学智, 李新, 等. 基于被动微波遥感的青藏高原雪深反演及其结果评价[J]. 遥感学报2001(3): 161-165.
[7] Kang Shichang, Guo Wanqin, Zhong Xinyue, et al. Changes in the mountain cryosphere and their impacts and adaptation measures [J]. Climate Change Research202016(2): 143-152.
康世昌, 郭万钦, 钟歆玥, 等. 全球山地冰冻圈变化、影响与适应[J]. 气候变化研究进展202016(2): 143-152.
[8] Yang Jianwei, Jiang Lingmei, Pan Jinmei. Uncertainty analysis for retrieving snow water equiva‐lent based on passive microwave remote sensing[J]. Remote Sensing Technology and Application202338(6): 1 274-1 284.
杨建卫, 蒋玲梅, 潘金梅. 被动微波遥感反演雪水当量不确定性因素分析[J]. 遥感技术与应用202338(6): 1 274-1 284.
[9] Tanniru S, Ramsankaran R. Passive microwave remote sensing of snow depth: techniques, challenges and future directions[J]. Remote Sensing202315(4): 1052.
[10] Chang A T C, Foster J L, Hall D K, et al. Snow water equivalent estimation by microwave radiometry[J]. Cold Regions Science and Technology19825(3): 259-267.
[11] Foster J. Comparison of snow mass estimates from a prototype passive microwave snow algorithm, a revised algorithm and a snow depth climatology[J]. Remote Sensing of Environment199762(2): 132-142.
[12] Smith T, Bookhagen B. Assessing uncertainty and sensor biases in passive microwave data across High Mountain Asia[J]. Remote Sensing of Environment2016181: 174-185.
[13] Zhao Z, Hao X, Shao D, et al. A snow depth downscaling algorithm based on deep learning fusion of enhanced passive microwave and cloud-free optical remote sensing data in China[J]. Remote Sensing202416(24): 4756.
[14] Zhu L L, Zhang Y H, Wang J G, et al. Downscaling snow depth mapping by fusion of microwave and optical remote-sensing data based on deep learning[J]. Remote Sensing202113(4): 584.
[15] Hou Haiyan, Hou Jinliang, Huang Chunlin, et al. Retrieve snow depth of north of Xinjiang region from AMSR2 data based on artificial neural network technology[J]. Remote Sensing Technology and Application201833(2): 241-251.
侯海艳, 侯金亮, 黄春林, 等. 基于人工神经网络和AMSR2多频微波亮温的北疆地区雪深反演[J]. 遥感技术与应用201833(2): 241-251.
[16] Gao S, Li Z, Zhang P, et al. A novel global snow depth retrieval method considering snow metamorphism and forest influence[J]. Remote Sensing of Environment2023295: 113712.
[17] Yang J, Jiang L, Pan J, et al. Comparison of Machine Learning-based snow depth estimates and development of a new operational retrieval algorithm over China[J]. Remote Sensing202214(12): 2800.
[18] Xing D, Hou J L, Huang C L, et al. Estimation of snow depth from AMSR2 and MODIS data based on deep residual learning network[J]. Remote Sensing202214(20): 5089.
[19] Yao H M, Zhang Y M, Jiang L J, et al. Snow parameters inversion from passive microwave remote sensing measurements by deep convolutional neural networks[J]. Sensors22(13): 4769.
[20] Wang Y H, Gupta H V, Zeng X B, et al. Exploring the potential of long short-term memory networks for improving understanding of continental- and regional-scale snowpack dynamics[J]. Water Resources Research202258(3): e2021WR031033.
[21] Hou H Y, Hu G H, Hu Z D. Snow depth retrieval from microwave remote sensing by combining wavelet transform and machine learning models in Northern Xinjiang, China[J]. Journal of Applied Remote Sensing202418(2): 024517.
[22] Ahmad J A, Forman B A, Kwon Y. Analyzing machine learning predictions of passive microwave brightness temperature spectral difference over snow-covered terrain in high Mountain Asia[J]. Frontiers in Earth Science20197: 212.
[23] Sturm M, Liston G E. Revisiting the global seasonal snow classification: an updated dataset for Earth system applications[J]. Journal of Hydrometeorology202122: 2 917-2 938.
[24] Rautiainen K, Lemmetyinen J, Schwank M, et al. Detection of soil freezing from L-band passive microwave observations[J]. Remote Sensing of Environment2014147: 206-218.
[25] Wever N, Schmid L, Heilig A, et al. Verification of the multi-layer SNOWPACK model with different water transport schemes[J]. The Cryosphere20159(6): 2 271-2 293.
[26] Herrero J, Polo M J, Moñino A, et al. An energy balance snowmelt model in a Mediterranean site[J]. Journal of Hydrology2009371(1/2/3/4): 98-107.
[27] Wayand N E, Massmann A, Butler C, et al. A meteorological and snow observational data set from Snoqualmie Pass (921 m), Washington Cascades, USA[J]. Water Resources Research201551(12): 10 092-10 103.
[28] You Y H, Huang C L, Wang Z, et al. A genetic particle filter scheme for univariate snow cover assimilation into Noah-MP model across snow climates[J]. Hydrology and Earth System Sciences202327(15): 2 919-2 933.
[29] Wang Hongwei, Huang Chunlin, Hao Xiaohua, et al. Analyses of the spatiotemporl variations of snow cover in North Xinjiang[J]. Journal of Glaciology and Geocryology201436(3): 508-516.
王宏伟, 黄春林, 郝晓华, 等. 北疆地区积雪时空变化的影响因素分析[J]. 冰川冻土201436(3): 508-516.
[30] Liu Y Q, Peters-Lidard C D, Kumar S V, et al. Blending satellite-based snow depth products with in situ observations for streamflow predictions in the Upper Colorado River Basin[J]. Water Resources Research201551(2): 1 182-1 202.
[31] Skaugen T, Melvold K. Modeling the snow depth variability with a high-resolution lidar data set and nonlinear terrain dependency[J]. Water Resources Research201955(11): 9 689-9 704.
[32] Li Y, Zhao X Y, Zhao Q H. Snow depth inversion in forest areas from sentinel-1 data based on phase deviation correction[J]. Remote Sensing202214(23): 5930.
[33] Liu Xuejun, Gong Jianya, Zhou Qiming, et al. A study of accuracy and algorithms for calculating slope and aspect based on grid Digital Elevation Model (DEM)[J]. Acta Geodaetica et Cartographica Sinica200433(3): 258-263.
刘学军, 龚健雅, 周启鸣, 等. 基于 DEM 坡度坡向算法精度的分析研究[J]. 测绘学报200433(3): 258-263.
[34] Che Tao, Dai Liyun, Li Xin. Long-term series of daily snow depth dataset in China (1979-2024)[DS/OL]. National Tibetan Plateau/Third Pole Environment Data Center (2024-09-24)[2024-12-10]. .
车涛, 戴礼云, 李新. 中国雪深长时间序列数据集(1979-2024)[DS/OL]. 国家青藏高原数据中心(2024-09-24)[2024-12-10]. .
[1] 王永杰, 张晓东, 唐文彬, 赵少杰, 马晋, 孟义真, 王子卫, 周纪. 星载被动微波亮温影像轨道间隙填补:进展与展望[J]. 地球科学进展, 2025, 40(1): 99-110.
[2] 肖雄新, 张廷军. 基于被动微波遥感的积雪深度和雪水当量反演研究进展[J]. 地球科学进展, 2018, 33(6): 590-605.
[3] 祁晓凡,杨丽芝,韩 晔,尚 浩,邢立亭. 济南泉域地下水位动态及其对降水响应的交叉小波分析[J]. 地球科学进展, 2012, 27(9): 969-978.
[4] 罗坚,姜勇强,戴彩悌. 提升小波变换在气象格点数据无损压缩中的应用[J]. 地球科学进展, 2012, 27(4): 460-465.
[5] 王国栋, 康建成, Han Guoqi, 刘超, 闫国东. 中国东海海平面变化多尺度周期分析与预测[J]. 地球科学进展, 2011, 26(6): 678-684.
[6] 窦燕,陈曦. 基于站点的中国天山山区积雪要素变化研究[J]. 地球科学进展, 2011, 26(4): 441-448.
[7] 吴兆福,高飞,陶庭叶. 小波变换后的噪声信息在大坝变形监测精度评定中的应用[J]. 地球科学进展, 2008, 23(6): 590-594.
[8] 孙之文,施建成,蒋玲梅,杨虎,张立新. 被动微波遥感反演中国西部地区雪深、雪水当量算法初步研究[J]. 地球科学进展, 2006, 21(12): 1363-1369.
[9] 吴兆福;高飞;宫鹏. 基于边缘特征的遥感影像小波变换融合法[J]. 地球科学进展, 2005, 20(7): 705-709.
[10] 陈仁升,康尔泗,张济世. 小波变换在河西地区水文和气候周期变化分析中的应用[J]. 地球科学进展, 2001, 16(3): 339-345.
阅读次数
全文


摘要