Soil moisture is recognized as a key parameter in agriculture, hydrology, meteorology, and other fields. An efficient means for large-scale monitoring is offered by remote sensing, yet traditional extraction methods are hindered by low accuracy and poor adaptability. In response to the above issues, a soil moisture network extraction service based on the GPT4All large model was proposed, through which accuracy and reliability were enhanced by the integration of AI with remote sensing. Landsat 8 images were processed using remote sensing cloud computing platform to obtain soil line parameters, vegetation indices, coverage, vertical drought indices, and other relevant data. The GPT4All model was employed to optimize the mean pixel dryness index. The GPT4All model was innovatively employed to optimize the modified perpendicular drought index, with dynamic adjustment of correction coefficients, contrast, noise suppression, and moisture thresholds based on regional features and historical data. Compliant with Open Geospatial Consortium standards, the system was equipped with web service interfaces and RESTful APIs, enabling seamless integration between web and server for convenient online extraction and analysis. Taking Yongxing Town in Kaifeng City, Henan Province, China, as the study area, the results show that the GPT4Al large model extraction method outperforms traditional methods in metrics such as root mean square error, mean absolute error, and correlation coefficient. The soil moisture extraction service enhanced by integrating the GPT4All large model demonstrates significantly improved automation and adaptability, enabling high-precision monitoring across diverse climates and land cover regions. This provides universally accessible high-precision soil moisture sensing capabilities for users at all levels, with broad application prospects in agricultural irrigation decision-making, drought monitoring, flood warning, and other fields.
ZHOU Lian-jie, DU Jia-kuan, LI Jia, DENG Tian-min, ZHANG Xiang.
Soil Moisture Network Extraction Service Based on GPT4All Model[J].
Science Technology and Engineering, 2026, 26(28): 12296-12305 DOI:10.12404/j.issn.1671-1815.2505713
传统的土壤湿度监测主要依赖地面站点观测与遥感反演两种技术手段。地面站点观测精度较高,但受限于空间分布的稀疏性(全球仅约5 000个监测站),难以满足大范围、连续尺度的监测需求。相比之下,遥感技术为大区域土壤湿度监测提供了新的途径,但其反演数据常面临精度不足、时空分辨率受限等问题[5]。综上所述,当前的土壤湿度监测面临着地面观测“点状稀疏”与遥感反演“精度不稳”的双重局限。为了突破这一瓶颈,亟须发展多源数据融合技术。通过协同地基观测、主被动遥感(如Soil Moisture Active Passive、Sentinel系列卫星)及新兴监测手段(无人机、物联网传感器),并结合机器学习等智能算法优化反演模型,从而为农业、气象及防灾减灾等领域提供可靠的数据支撑。近年来,人工智能技术的快速发展为解决上述问题提供了新思路。特别是大语言模型(large language model,LLM)在处理复杂非线性关系方面表现出巨大潜力。GPT4All是一种轻量级大语言模型,凭借其部署方便、资源需求相对较低、推理速度迅速等优势,为提升土壤湿度遥感提取精度提供了技术可行性。将GPT4All大模型与土壤湿度遥感提取技术相融合,不但能提高土壤湿度数据的精确水平,还可凭借网络服务的形式,让数据处理和分析能力惠及范围更广泛的用户群体。
网络服务与用户界面是系统与用户交互的通道,直接影响用户体验。系统构建了完整的网络服务后端与友好的Web前端界面,提供了便捷的土壤湿度提取服务。网络服务后端借助Flask框架得以实现,给出RESTful API和WPS服务这两类的接口,RESTful API遵循表现层状态转移(representational state transfer,REST)架构风格,通过超文本传输协议(hypertext transfer protocol,HTTP)方法实现操作,并采用JSON格式进行数据交换。系统实现了GET与POST两种HTTP方法,分别用于查询数据和处理请求递交。针对土壤湿度提取等计算密集型任务,系统采用异步处理模式:用户提交请求后立刻获得处理ID,并通过该ID查询任务状态与结果,避免了长时间的同步等待。WPS服务接口基于PyWPS库开发,严格遵循OGC WPS标准规范[20]。系统将土壤湿度提取功能组合成一个WPS处理服务,定义了输入参数(时间范围、地理范围、处理选项)和输出结果的格式(JSON 结果数据、Base64编码预览图像),如图5所示。WPS服务支持3种操作模式:GetCapabilities(返回服务元数据)、DescribeProcess(返回处理流程详细描述)及Execute(执行任务并返回结果)[21]。
JiangKaisheng, LiuYu, LiuYing, et al. Exploring the relationship between vegetation coverage and climate factors in northwest China[J]. Surveying and Mapping Science, 2023, 48(6): 172-180.
HeSaike, ZhangPeijie, ZhangWeiguang, et al. Paradigm shift in decision-making in the era of large models[J]. Journal of China University of Geosciences (Social Sciences Edition), 2023, 23(4): 82-91.
WangShitai, YangKexin, YinMin, et al. A robust regression soil moisture inversion method based on GNSS-IR multi-satellite fusion[J]. Bulletin of Surveying and Mapping, 2024(3): 69-74.
LiuYuanhua, YangCheng, NiuXinliang. A soil moisture retrieval method considering noise and sudden weather changes[J]. Science of Surveying and Mapping, 2023, 48(9): 191-201.
[15]
LiY, LuoL, GuoF, et al. Quality control and improvement of GNSS-IR soil moisture robust inversion model[J]. Advances in Space Research, 2025, 75(1): 244-258.
HeShanbao, PangJingjing, ChengXing, et al. Research on soil moisture retrieval methods from satellite remote sensing data[J]. Journal of Anhui Normal University (Natural Science Edition), 2024, 47(1): 1-10.
QinQiming, YouLin, ZhaoYue, et al. Automatic extraction algorithm of soil line based on two-dimensional spectral feature space[J]. Transactions of the Chinese Society of Agricultural Engineering, 2012, 28(3): 167-171.
[20]
JesusJ D, WalkerP, GrantM, et al. WPS orchestration using the Taverna workbench: the eScience approach[J]. Computers, 2012, 47(1): 75-86.
ZhaoAnzhou, ZhangXuepei, ZhangLianghan, et al. Dynamic monitoring of environmental quality in Beijing-Tianjin-Hebei Region based on multi-source remote sensing data[J]. Geomatics World, 2021, 28(4): 93-99.
[23]
RichardsonL, AmundsenM, RubyS. RESTful Web APIs[M]. Sebastopol, CA: O’Reilly Media Inc., 2013.
[24]
HastutiA W, IsmailN P, NagaiM. Analysis of coastline extraction indices using Sentinel-2 and Google Earth Engine, case study in Bali, Indonesia[J]. BIO Web of Conferences, 2024, 106: 13.
[25]
GorelickN, HancherM, DixonM, et al. Google Earth Engine: planetary-scale geospatial analysis for everyone[J]. Remote Sensing of Environment, 2017, 202: 18-27.
ZhangYun, LiPeng, LiZhenhong, et al. High-precision spatiotemporal continuous land-sea surface temperature inversion of the Yellow River Delta using Sentinel-3 data[J]. Bulletin of Surveying and Mapping, 2023(1): 120-126.
LiYali, WangXiaoqin, ChenYunzhi, et al. Correlation analysis between land surface temperature and vegetation coverage in Fujian Province[J]. Journal of Geo-Information Science, 2019, 21(3): 445-454.
[30]
AnandY, NussbaumZ, TreatA, et al. GPT4All: an ecosystem of open source compressed language models[J]. 2023, 7(5): 59-64.
[31]
XuN, DaccacheA, AhmadiA, et al. Soil moisture estimation with microwave remote sensing: a systematic review and Meta-analysis[J]. International Journal of Digital Earth, 2025, 18(1): 1-27.
[32]
CastronovaA M, GoodallJ L, ElagM M, et al. Models as web services using the Open Geospatial Consortium (OGC) Web Processing Service (WPS) standard[J]. Environmental Modelling & Software, 2013, 41(3): 72-83.
[33]
ChenY, FengX, FuB. An improved global remote-sensing-based surface soil moisture (RSSSM) dataset covering 2003—2018[J]. Earth System Science Data, 2021, 21(1): 98-101.
WuYingjie, ZhuYongchao, KongXiangning. Inversion of farmland soil water content based on Landsat-8 images[J]. Chinese Agricultural Science Bulletin, 2024, 40(32): 129-134.
[36]
ZhangC, ZhangC, ZhengS, et al. A complete survey on generative AI (AIGC): is ChatGPT from GPT-4 to GPT-5 all you need?[J]. arXiv Preprint, arXiv: 2303.11717, 2023.
FengYiming, QiaoKun, FengShi’ang, et al. Spatiotemporal dataset of vegetation coverage (FVC) in the circum-Tarim Basin from 1990 to 2021[J]. China Scientific Data, 2023, 8(2): 307-314.