基于 FY-3D 卫星和机器学习的辽宁省土壤水分监测技术研究

Soil Moisture Monitoring Technology in Liaoning Province Based on FY-3D Satellite and Machine Learning

摘要

干旱是国家粮食安全的重要威胁之一，土壤水分精准监测技术既对抗旱减灾至关重要，同时也是高标准农田节水灌溉的重要支撑。基于FY-3D遥感数据，采用统计分析方法筛选与土壤水分变化显著相关的植被遥感指数和土壤参数等关键特征因子，利用随机森林、XGBoost以及ConvLSTM深度学习模型构建了0~20 cm深度的土壤体积含水量反演模型。研究结果表明：在裸土-播种期与成熟期，土壤参数与地表温度（LST）成为主要特征因子；出苗-拔节期与抽雄-灌浆期，植被指数对土壤水分反演的影响明显增强。双时间节点较单时间节点建模的反演结果 R²提升了0.45 以上，显著提升了模型精度。ConvLSTM 模型具有较强的稳定性与泛化能力，全时期模型反演结果 R²均值达到0.71，RMSE低于0.046（cm³/cm³），MAE低于0.032（cm³/cm³）。基于ConvLSTM模型的双时间节点建模能够高精度反演辽宁省土壤水分，实现了逐日的大面积土壤水分监测，可有效地支撑气象抗旱减灾业务。

Abstract

Drought is one of the major threats to national food security. Accurate soil moisture monitoring technology is not only essential for drought mitigation and disaster reduction but also serves as a critical foundation for water-saving irrigation in high-standard farmland. Based on FY-3D remote sensing data, statistical analysis methods are used to screen key feature variables such as vegetation indices and soil parameters that are significantly correlated with soil moisture variations. These features are then used to develop Random Forest, XGBoost, and ConvLSTM deep learning models for retrieving soil volumetric water content at a depth of 0-20 cm. The results indicate that during the Bare soil - Sowing period and Maturity period, soil parameters and land surface temperature (LST) are the dominant factors for soil moisture retrieval. In contrast, during Emergence - jointing stage and Tasseling - Grouting stage, the impact of vegetation indices on soil moisture inversion is significantly enhanced. Compared with single-time-point modeling, the R² of dual-time-point modeling increases by more than 0.45, significantly enhancing model accuracy. The ConvLSTM model demonstrates strong stability and generalization ability, with mean R² values of 0.71 across different periods, RMSE below 0.046 (cm³/cm³), and MAE below 0.032 (cm³/cm³).The dual-time-point ConvLSTM modeling enables high-accuracy retrieval of soil moisture in Liaoning Province, achieving large-scale, daily soil moisture monitoring, and providing effective support for meteorological drought mitigation and disaster reduction operations.

关键词

FY-3D / ConvLSTM / 土壤水分反演

Key words

FY-3D / ConvLSTM / soil moisture inversion

参考文献

[1] 陈国茜, 杨耀晨, 李素雲, 等. 基于随机森林模型的青海高原土壤水分反演技术研究[J]. 遥感技术与应用, 2024, 39(06): 1417-1428.

[2] 郝萍萍, 韩金朝, 刘晓, 等. 土壤水分含量测定方法的技术现状与发展趋势[J]. 江苏农业科学, 2024, 52(17): 11-20.

[3] 江笑薇, 白建军, 刘宪峰. 基于多源信息的综合干旱监测研究进展与展望 . 地球科学进展[J], 2019, 34(03): 275-287.

[4] 张浩杰, 薛华柱, 赵天杰, 等. 联合极轨和倾斜轨道被动微波传感器的青藏高原土壤水分变化监测[J]. 遥感技术与应用, 2024, 39(06): 1404-1416.

[5] 赵天杰. 被动微波反演土壤水分的L波段新发展及未来展望[J]. 地理科学进展, 2018, 37(02): 198-213.

[6] 付平凡, 杨晓静, 姜波, 等. 融合多源数据的高分辨率土壤水分模拟模型构建及应用[J]. 农业工程学报, 2025, 41(05): 96-106.

[7] 李万涛, 杨明龙, 唐秀娟, 等. 基于Sentinel 多源遥感数据的农田地表土壤水分反演[J]. 南方农业学报, 2025, 56(01): 87-96.

[8] 蔡斌, 陆文杰, 郑新江. 气象卫星条件植被指数监测土壤状况[J]. 国土资源遥感, 1995(04): 45-50+ 66.

[9] 张涛, 张立新, 蒋玲梅, 等. 裸露土壤参数空间分布对被动微波遥感反演土壤水分的影响[J]. 中国科学:地球科学, 2012, 42(09): 1403-1413.

[10] 张添, 黄春林, 沈焕锋. 土壤水分对土壤参数的敏感性及其参数优化方法研究[J]. 地球科学进展, 2012, 27(06): 678-685.

[11] 刘昀昊, 李雪冬, 费龙, 等. 基于特征选择与遗传神经网络的土壤水分反演[J]. 中国农业气象, 2024, 45(10): 1095-1108.

[12] 刘娣, 孙佳倩, 余钟波. 基于机器学习模型的多层土壤湿度反演[J]. 河海大学学报(自然科学版), 2024, 52(03): 7-14.

[13] Hanade Houmma I, El Mansouri L, Gadal S, et al. Prospective analysis of spatial heterogeneity influence on the concordance of remote sensing drought indices: a case of semi-arid agrosystems in Morocco (Moulouya and Tensift watersheds) [J]. Geocarto International, 2022, 37(27): 14899-14924.

[14] Chen J, Jin Q, Chao J. Design of deep belief networks for short-term prediction of drought index using data in the Huaihe River basin[J]. Mathematical Problems in Engineering2012, 2012(1): 235929.

[15] Wu T, Xu L, Lv Y, et al. Integrating causal inference with ConvLSTM networks for spatiotemporal forecasting of root zone soil moisture[J]. Journal of Hydrology, 2025, 659: 133246.

[16] ElSaadani M, Habib E, Abdelhameed A M, et al. Assessment of a spatiotemporal deep learning approach for soil moisture prediction and filling the gaps in between soil moisture observations[J]. Frontiers in Artificial Intelligence, 2021, 4: 636234.

[17] Yinglan A, Wang G, Hu P, et al. Root-zone soil moisture estimation based on remote sensing data and deep learning[J]. Environmental Research, 2022, 212: 113278.

[18] 赵天杰, 施建成, 徐红新, 等. 闪电河流域水循环和能量平衡遥感综合试验[J]. 遥感学报, 2021, 25(04): 871-887.

[19] 王岩, 王敬宜, 冯锐, 等. 基于FY-3D/MERSI数据的东北地区干旱监测方法研究[J]. 干旱地区农业研究, 2023, 41(04): 289-297+ 317.

[20] 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.

[21] Kogan F N. Application of vegetation index and brightness temperature for drought detection [J]. Advances in Space Research, 1995, 15(11): 91-100.

[22] Bajgain R, Xiao X M, Wagle P, et al. Sensitivity analysis of vegetation indices to drought over two tall grass prairie sites.ISPRS [J]. ISPRS Journal of Photogrammetry and Remote Sensing, 2015, 108: 151-160.

[23] 黄友昕, 胡茂胜, 沈永林, 等. MODIS 干旱指数结合RBFNN 反演冬小麦返青期土壤湿度[J]. 农业工程学报, 2019, 35(12): 81-88.

[24] Carlson T N, Gillies R R, Perry E M. A method to make use of thermal infrared temperature and NDVI measurements to infer surface soil water content and fractional vegetation cover[J]. Remote Sensing Reviews, 1994, 9(1/2): 161-173.

[25] Zhang Z, Xu W, Shi Z, et al. Establishment of a comprehensive drought monitoring index based on multisource remote sensing data and agricultural drought monitoring[J]. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2021, 14: 2113-2126.

[26] Xing N, Huang W, Xie Q, et al. A transformed triangular vegetation index for estimating winter wheat leaf area index[J]. Remote Sensing, 2019, 12(1): 16.

[27] Kogan F. World droughts in the new millennium from AVHRR‐based vegetation health indices[J]. Eos, Transactions American Geophysical Union, 2002, 83(48): 557-563.

[28] Gao B C. NDWI: a normalized difference water index forremote sensing of vegetation liquid water from space[J]. Remote Sensing of Environment, 1996, 58(3): 257-266.

[29] 王卓颖, 刘杨晓月, 杨骥, 等 . 风云卫星系列及CLDAS 土壤水分产品多尺度精度验证与评价:以青藏高原那曲地区为例[J]. 中国农业气象, 2021, 42(09): 788-804.

[30] 全国农业气象标准化技术委员会(SAC/TC 539). 农业干旱等级:GB/T 32136-2015[S]. 北京: 中国标准出版社, 2015.

[31] 王晓, 刘海新, 孙振宇, 等. 融合多源数据与机器学习算法的综合干旱监测模型构建——以河北省为例[J]. 水资源与水工程学报, 2024, 35(06): 207-219.

[32] Adab H, Morbidelli R, Saltalippi C, et al. Machine learning to estimate surface soil moisture from remote sensing data[J]. Water, 2020, 12(11): 3223.

[33] Luo L, Li Y, Guo F, et al. Research on robust inversion model of soil moisture content based on GF-1 satellite remote sensing[J]. Computers and electronics in agriculture, 2023, 213: 108272.

[34] 王思超, 蔡玉林, 朱子正, 等. XGBoost 和 DSCGAN 模型协同的 SMAP 土壤水分降尺度方法[J]. 地球信息科学学报, 2024, 26(09): 2213-2225.

[35] Zhang Y, Xie D, Tian W, et al. Construction of an integrated drought monitoring model based on deep learning algorithms[J]. Remote Sensing, 2023, 15(3) 667.

[36] 王然, 赵建辉, 杨会巾, 等. 基于RIME-CNN-SVR模型的麦田土壤水分反演[J]. 农业工程学报, 2024, 40(15): 94-102.

[37] Kasim A A, Leng P, Li Y X, et al. Remote sensing of root zone soil moisture: A review of methods and products[J]. Journal of Hydrology, 2025: 133002.

[38] 张菊英, 覃光华, 郭立. 土壤水分监测站点优选与预测模型选择的研究[J]. 中国农村水利水电, 2025(03): 37-43+54.

[39] Wu Z, Cui N, Zhang W, et al. Estimating soil moisture content in citrus orchards using multi-temporal sentinel-1A data-based LSTM and PSO-LSTM models[J]. Journal of Hydrology, 2024, 637: 131336.

[40] Kumar K C A, Reddy G P O, Masilamani P, et al. Integrated drought monitoring index: A tool to monitor agricultural drought by using time-series datasets of space-based earth observation satellites[J]. Advances in Space Research, 2021, 67(1): 298-315.

[41] Zhao Q, Ma X, Yao W, et al. A drought monitoring method based on precipitable water vapor and precipitation[J]. Journal of Climate, 2020, 33(24): 10727-10741.

[42] Gavahi K, Abbaszadeh P, Moradkhani H. How does precipitation data influence the land surface data assimilation for drought monitoring? [J]. Science of the Total Environment, 2022, 831: 154916.

[43] 王霞迎, 折育霖, 张双成, 等 . 基于 Sentinel-1 和 Landsat 8遥感数据的高原地区土壤水分反演[J]. 测绘通报, 2024(07): 140-146.

基金资助

中国气象局人才项目“中国气象局农业气象重点创新团队”(CMA2024ZD02)

科院基本科研业务费“农田生态系统-气候-水资源相互作用与农田节水灌溉”(2024Z001)

国家自然基金面上项目“玉米开花后干旱胁迫导致叶片早衰的定量诊断及产量模拟研究”(42275202)