基于多源遥感特征参数与ACNN的土壤湿度反演研究

doi:10.12118/j.issn.1000-6060.2025.314

摘要/Abstract

摘要：

土壤湿度是农业、气象和水文领域的关键参数，提高其反演精度可以为干旱区农作物精准灌溉提供技术支撑。提出一种多源遥感特征参数与自适应卷积神经网络（Adam convolution neural networks，ACNN）的土壤湿度反演方法，从Sentinel-1/2获取11项特征参数并进行特征融合，利用皮尔逊相关系数与四分位距法筛选确定6项参数，利用ACNN与其他模型精度对比并反演银川市永宁县土壤湿度空间分布。结果表明：（1）对数双极化比参数对模型反演贡献度最高，揭示微波极化比参数的核心作用。（2） ACNN模型反演精度（决定系数为0.947，均方根误差为1.263，平均绝对误差为0.840）显著优于对比模型。（3）最少特征参数的反演效果优于基本参数和全部参数。（4）多源遥感参数与ACNN反演的土壤湿度空间分布与高斯特征可为区域农作物灌溉决策提供科学依据。通过验证多源遥感特征参数与深度学习耦合框架的可行性，为干旱区农业水资源管理提供技术方案。

关键词: 多源遥感, 特征参数, 土壤湿度, 反演, 自适应卷积神经网络

Abstract:

Soil moisture is a key parameter in agriculture, meteorology, and hydrology. Enhancing the accuracy of its inversion can provide essential support for precise irrigation in arid regions. This study presents a soil moisture inversion method that combines multi-source feature parameter fusion with an adaptive convolutional neural network (ACNN). We extracted 11 basic feature parameters from Sentinel-1/2 data and performed 29 feature fusion operations, selecting 6 parameters using the Pearson correlation coefficient and Interquartile Range methods. The ACNN model’s accuracy was compared with that of other models and then used to invert the spatial distribution of soil moisture in Yongning County, Yinchuan City. The findings reveal that (1) The necessity of the 6 multi-source fusion feature parameters is confirmed through RF modeling and ablation experiments, and the importance of single-source feature parameters is also assessed. Notably, the dual polarization ratio logarithmic parameter significantly enhances the model inversion, highlighting the central role of the microwave polarization ratio parameter. (2) An accuracy comparison of soil moisture inversion was conducted among 4 models, namely, BP, GABP, RF, and ACNN, using different training and test sets. The ACNN model achieved superior inversion accuracy (R²=0.947, RMSE=1.263, MAE=0.840) compared to the other models. (3) Evaluating the inversion accuracy of 40 feature parameters, 11 basic parameters, and 6 optimized parameters within the ACNN framework revealed that the six optimized parameters had the highest R² and the lowest RMSE and MAE. The model demonstrated that fewer feature parameters led to better accuracy and shorter computation times, outperforming both the basic and all-parameter approaches. (4) The measured and inverted soil moisture values at sampling sites showed minimal differences, and the inverted values at larger spatial scales aligned well with actual measurements. This consistency supports effective monitoring of crop growth conditions and irrigation scheduling. In addition, the soil moisture inversion results align with the conclusion of moisture briefing. Thus, the spatial distribution and Gaussian characteristics of soil moisture derived from multi-source remote sensing parameters and ACNN inversion can inform regional crop irrigation decisions. This study validates the feasibility of integrating multi-source remote sensing feature parameters with deep learning, offering a technical solution for managing agricultural water resources in arid areas.

Key words: multi-source remote sensing, feature parameters, soil moisture, inversion, ACNN

李占虎, 郭中华, 马嘉强, 李蕾蕾. 基于多源遥感特征参数与ACNN的土壤湿度反演研究[J]. 干旱区地理, 2026, 49(6): 1192-1202.

LI Zhanhu, GUO Zhonghua, MA Jiaqiang, LI Leilei. Soil moisture inversion based on multi-source remote sensing feature parameter and ACNN[J]. Arid Land Geography, 2026, 49(6): 1192-1202.

图/表 14

图1

表1

表2

图2

表3

图3

图4

图5

表4

表5

图6

表6

图7

图8

参考文献 33

[1]	Wang W Z, Ma C F, Wang X F, et al. A soil moisture experiment for validating high-resolution satellite products and monitoring irrigation at agricultural field scale[J]. Agricultural Water Management, 2024, 304: 109071, doi: 10.1016/j.agwat.2024.109071.
[2]	杨春曦, 刘华, 谢可心, 等. 便携式土壤湿度检测装置用于精准灌溉决策系统[J]. 农业工程学报, 2018, 34(22): 84-91.
	[Yang Chunxi, Liu Hua, Xie Kexin, et al. Portable detection device of soil moisture for precision irrigation decision system[J]. Transactions of the Chinese Society of Agricultural Engineering, 2018, 34(22): 84-91. ]
[3]	Bayat B, Tol D V C, Verhoef W. Remote sensing of grass response to drought stress using spectroscopic techniques and canopy reflectance model inversion[J]. Remote Sensing, 2016, 8(7): 557, doi: 10.3390/rs8070557.
[4]	吕娜, 郭梦京, 赵馨, 等. 内陆淡水湖博斯腾湖水质遥感反演及时空演变特征[J]. 干旱区地理, 2024, 47(6): 953-966. doi: 10.12118/j.issn.1000-6060.2023.386
	[Lü Na, Guo Mengjing, Zhao Xin, et al. Remote sensing inversion of water quality and spatiotemporal evolution characteristics of the Bosten Inland Freshwater Lake[J]. Arid Land Geography, 2024, 47(6): 953-966. ] doi: 10.12118/j.issn.1000-6060.2023.386
[5]	Li Z Q, Tu Y H, Liu N. Inversion of the soil moisture based upon neural network[C]// 2000 5th International Symposium on Antennas, Propagation, and EM Theory. ISAPE 2000 (IEEE Cat. No.00EX417). Beijing: Institute of Electrical and Electronics Engineers, 2000: 398-401.
[6]	刘娣, 孙佳倩, 余钟波. 基于改进BP神经网络的多层土壤湿度反演[J]. 节水灌溉, 2023(11): 19-27. doi: 10.12396/jsgg.2023191
	[Liu Di, Sun Jiaqian, Yu Zhongbo. Multi-layer soil moisture inversion based on improved back propagation neural network technique[J]. Water Saving Irrigation, 2023(11): 19-27. ] doi: 10.12396/jsgg.2023191
[7]	Kaheil H Y, Gill K M, McKee M, et al. A new bayesian recursive technique for parameter estimation[J]. Water Resources Research, 2006, 42(8): W08423, doi: 10.1029/2005wr004529.
[8]	杜绍杰, 赵天杰, 施建成, 等. Sentinel-1和Sentinel-2协同反演地表土壤水分[J]. 遥感技术与应用, 2022, 37(6): 1404-1413. doi: 10.11873/j.issn.1004-0323.2022.6.1404
	[Du Shaojie, Zhao Tianjie, Shi Jiancheng, et al. Sentinel-1 and Sentinel-2 synergistic retrieval of surface soil moisture[J]. Remote Sensing Technology and Application, 2022, 37(6): 1404-1413. ]
[9]	Notarnicola C, Angiulli M, Posa F. Soil moisture retrieval from remotely sensed data: Neural network approach versus Bayesian method[J]. IEEE Transactions on Geoscience and Remote Sensing, 2008, 46(2): 547-557. doi: 10.1109/TGRS.2007.909951
[10]	Kolassa J, Reichle R, Liu Q, et al. Estimating surface soil moisture from SMAP observations using a neural network technique[J]. Remote Sensing of Environment, 2018, 204: 43-59. doi: 10.1016/j.rse.2017.10.045 pmid: 29290638
[11]	Wang Z H, Gao Q, Crosetto M, et al. High-resolution surface soil moisture retrieval: A hybrid machine learning framework integrating change detection and downscaling for precision water management[J]. International Journal of Applied Earth Observation and Geoinformation, 2025, 142: 104702, doi: 10.1016/j.jag.2025.104702.
[12]	Zhang Z H, Wang J J, Ding J L, et al. Soil moisture retrieval and spatiotemporal variation analysis based on deep learning[J]. Agricultural Water Management, 2025, 317: 109622, doi: 10.1016/j.agwat.2025.109622.
[13]	刘瑞亮, 贾科利, 李小雨, 等. 组合光学和微波遥感的耕地土壤含盐量反演[J]. 干旱区地理, 2024, 47(3): 433-444. doi: 10.12118/j.issn.1000-6060.2023.375
	[Liu Ruiliang, Jia Keli, Li Xiaoyu, et al. Inversion of soil salt content by combining optical and microwave remote sensing in cultivated land[J]. Arid Land Geography, 2024, 47(3): 433-444. ] doi: 10.12118/j.issn.1000-6060.2023.375
[14]	Zhang C M, Wan S J, Liu J P, et al. A method for ground soil moisture inversion based on integration of active and passive remote sensing data[J]. Sensor Letters, 2014, 12(3): 749-755. doi: 10.1166/sl.2014.3134
[15]	Kolassa J, Gentine P, Prigent C, et al. Soil moisture retrieval from AMSR-E and ASCAT microwave observation synergy. Part 1: Satellite data analysis[J]. Remote Sensing of Environment, 2016, 173: 1-14. doi: 10.1016/j.rse.2015.11.011
[16]	Wang R, Zhao J H, Yang H J, et al. Inversion of soil moisture on farmland areas based on SSA-CNN using multi-source remote sensing data[J]. Remote Sensing, 2023, 15(10): 2515, doi: 10.3390/rs15102515.
[17]	Inoubli R, Bennaceur L, Jarray N, et al. A comparison between the use of machine learning techniques and the water cloud model for the retrieval of soil moisture from Sentinel-1A and Sentinel-2A products[J]. Remote Sensing Letters, 2022, 13(10): 980-990. doi: 10.1080/2150704X.2022.2115862
[18]	Jia T Y, Shamseldin Y A, Liu T X, et al. Soil moisture inversion method for semi-arid regions using multi-temporal Sentinel-1 and Sentinel-2 data[J]. Journal of Hydrology, 2025, 661: 133603, doi: 10.1016/j.jhydrol.2025.133603.
[19]	Zhao J H, Zhang C Y, Min L, et al. Retrieval of farmland surface soil moisture based on feature optimization and machine learning[J]. Remote Sensing, 2022, 14(20): 5102-5117. doi: 10.3390/rs14205102
[20]	Wang Y L, Zhao J H, Guo Z W, et al. Soil moisture inversion based on data augmentation method using multi-source remote sensing data[J]. Remote Sensing, 2023, 15(7): 1899-1915. doi: 10.3390/rs15071899
[21]	Mattar C. A combined optical-microwave method to retrieve soil moisture over vegetated areas[J]. IEEE Transactions on Geoscience and Remote Sensing, 2012, 50(5): 1404-1413. doi: 10.1109/TGRS.2011.2179051
[22]	童玲, 陈彦, 贾明权. 雷达遥感机理[D]. 北京: 科学出版社, 2014.
	[Tong Ling, Chen Yan, Jia Mingquan. Mechanism of radar remote sensing[D]. Beijing: Science Press, 2014. ]
[23]	Yadav K S, Mishra S S, Gupta S. An efficient estimator for population variance using parameters of an auxiliary variable[J]. Journal of Statistics and Management Systems, 2019, 22(6): 1005-1013. doi: 10.1080/09720510.2017.1406643
[24]	Narayanarao B, Subhadip D, Dipankar M, et al. Soil moisture retrieval over croplands using dual-pol L-band GRD SAR data[J]. Remote Sensing of Environment, 2022, 271(3): 112900, doi: 10.1016/j.rse.2022.112900.
[25]	汉牟田, 张波, 杨东凯, 等. 利用GNSS干涉信号振荡幅度反演土壤湿度[J]. 测绘学报, 2016, 45(11): 1293-1300. doi: 10.11947/j.AGCS.2016.20160145
	[Han Mutian, Zhang Bo, Yang Dongkai, et al. Soil moisture retrieval utilizing GNSS interference signal amplitude[J]. Acta Geodaetica et Cartographica Sinica, 2016, 45(11): 1293-1300. ] doi: 10.11947/j.AGCS.2016.20160145
[26]	周美玲, 张德宁, 王浩, 等. 光学和微波遥感数据联合反演植被覆盖表层土壤含水率[J]. 灌溉排水学报, 2024, 43(1): 45-51.
	[Zhou Meiling, Zhang Dening, Wang Hao, et al. Inversion of surface soil moisture under vegetated areas based on optical and microwave remote sensing data[J]. Journal of Irrigation and Drainage, 2024, 43(1): 45-51. ]
[27]	Graldi G, Zardi D, Vitti A. Retrieving soil moisture at the field scale from Sentinel-1 data over a semi-arid mediterranean agricultural area[J]. Remote Sensing, 2023, 15(12): 2997, doi: 10.3390/rs15122997.
[28]	McNally A, Arsenault K, Kumar S, et al. A land data assimilation system for sub-Saharan Africa food and water security applications[J]. Scientific Data, 2017, 4(2): 170012-170030. doi: 10.1038/sdata.2017.12
[29]	自治区农业技术推广总站. 第十二期农田土壤墒情简[EB/OL]. [2024-07-02]. http://nynct.nx.gov.cn/xwzx/zwdt/202407/t20240-702_4583002.html.
	[General Station of Agricultural Technology Extension in the Autonomous Region. The 12^th briefing on soil moisture in agricultural fields[EB/OL].[2024-07-02]. http://nynct.nx.gov.cn/xwzx/zwdt/202407/t20240-702_4583002.html.
[30]	王峰, 李萍, 熊昱, 等. 不同干旱程度对夏玉米生长及产量的影响[J]. 节水灌溉, 2017(2): 1-4, 8.
	[Wang Feng, Li Ping, Xiong Yu, et al. Effects of different drought degrees on summer maize growth and yield[J]. Water Saving Irrigation, 2017(2): 1-4, 8. ]
[31]	Wang L G, Gao Y. Soil moisture inversion using multi-sensor remote sensing data based on feature selection method and adaptive stacking algorithm[J]. Remote Sensing, 2025, 17(9): 1569, doi: 10.3390/rs17091569.
[32]	李奎, 张瑞, 段金亮, 等. 利用SAR影像与多光谱数据反演广域土壤湿度[J]. 农业工程学报, 2020, 36(7): 134-140.
	[Li Kui, Zhang Rui, Duan Jinliang, et al. Wide-area soil moisture retrieval using SAR images and multispectral data[J]. Transactions of the Chinese Society of Agricultural Engineering (Transactions of the CSAE), 2020, 36(7): 134-140. ]
[33]	李伟, 宋睿, 刘明江, 等. 光学遥感在区域灌溉监测中应用的研究进展[J]. 排灌机械工程学报, 2024, 42(11): 1157-1165.
	[Li Wei, Song Rui, Liu Mingjiang, et al. Research progress on application of optical remote sensing in regional irrigation monitoring[J]. Journal of Drainage and Irrigation Machinery Engineering, 2024, 42(11): 1157-1165. ]

景号	Sentinel-1			Sentinel-2
景号	获取日期	极化方式	分辨率/m	获取日期	云量/%	分辨率/m
1	2024-06-25	同极、交叉	10×10	2024-06-24	3	10/20/60
2	2024-06-25	同极、交叉	10×10	2024-06-24	5	10/20/60

原特征参数	融合特征参数
入射角（θ）	余弦函数（cosθ）	正弦函数（sinθ）
同极后向散射系数（σ_VV）	指数同极极化后向散射系数（expσ_VV）	对数同极极化后向散射系数（logσ_VV）
交叉后向散射系数（σ_VH）	指数交叉极化后向散射系数（expσ_VH）	对数交叉极化后向散射系数（logσ_VH）
海拔（Ev）	指数后向散射系数极化比$\left[\mathrm{e}\mathrm{x}\mathrm{p}\left({\mathit{\sigma }}_{\mathrm{V}\mathrm{V}}\times {\mathit{\sigma }}_{\mathrm{V}\mathrm{H}}^{-1}\right)\right]$	对数后向散射系数极化比$\left[\mathrm{l}\mathrm{o}\mathrm{g}\left({\mathit{\sigma }}_{\mathrm{V}\mathrm{V}}\times {\mathit{\sigma }}_{\mathrm{V}\mathrm{H}}^{-1}\right)\right]$
	同极与交叉极化后向散射系数之和（σ_VV+σ_VH）	同极与交叉极化后向散射系数之差（σ_VV-σ_VH）
	同极与交叉极化后向散射系数之积（${\mathit{\sigma }}_{\mathrm{V}\mathrm{V}}\times {\mathit{\sigma }}_{\mathrm{V}\mathrm{H}}^{}$）	同极与交叉后向散射系数极化比（${\mathit{\sigma }}_{\mathrm{V}\mathrm{V}}\times {\mathit{\sigma }}_{\mathrm{V}\mathrm{H}}^{-1}$）

序号	特征参数	皮尔逊相关系数
1	$\mathrm{l}\mathrm{o}\mathrm{g}({\mathit{\sigma }}_{\mathrm{V}\mathrm{V}}\times {\mathit{\sigma }}_{\mathrm{V}\mathrm{H}}^{-1})$	-0.8954
2	logσ_VV	0.3974
3	logσ_VH	-0.3297
4	cosθ	0.3045
5	NDVI	0.2804
6	Ev	0.1724

输入特征参数	R²减少比	RMSE增加比	MAE增加比
$\mathrm{l}\mathrm{o}\mathrm{g}({\mathit{\sigma }}_{\mathrm{V}\mathrm{V}}\times {\mathit{\sigma }}_{\mathrm{V}\mathrm{H}}^{-1})$、logσ_VV、logσ_VH、cosθ、Ev	1.75	0.93	6.38
$\mathrm{l}\mathrm{o}\mathrm{g}({\mathit{\sigma }}_{\mathrm{V}\mathrm{V}}\times {\mathit{\sigma }}_{\mathrm{V}\mathrm{H}}^{-1})$、logσ_VV、logσ_VH、NDVI	3.30	12.86	6.42
cosθ、Ev、NDVI	4.80	34.92	19.62
$\mathrm{l}\mathrm{o}\mathrm{g}({\mathit{\sigma }}_{\mathrm{V}\mathrm{V}}\times {\mathit{\sigma }}_{\mathrm{V}\mathrm{H}}^{-1})$、logσ_VV、logσ_VH	5.71	18.76	20.36
cosθ、NDVI	18.56	68.02	44.73

序号	模型	R²	RMSE	MAE	MBE
1	BP	0.884	1.957	1.218	-0.106
2	GABP	0.906	1.880	1.069	-0.061
3	RF	0.929	1.492	0.962	0.015
4	ACNN	0.947	1.263	0.840	-0.024