顾及空间邻域特征的MODIS PWV神经网络校正模型

刘洋; 张文渊; 李文云; 张书毕; 张克非; 王楠迪; 高雨; 王梦瑶

doi:10.11834/jrs.20254201

模型与方法 | 浏览量 : 0 下载量: 160 CSCD: 0

R-PDF
PDF
导出
分享
收藏
专辑

顾及空间邻域特征的MODIS PWV神经网络校正模型
MODIS PWV neural network correction model considering spatial neighborhood characteristics
2026年30卷第1期页码：144-155
收稿：2024-05-23，

纸质出版：2026-01-07
DOI： 10.11834/jrs.20254201
稿件说明：

移动端阅览

刘洋，张文渊，李文云，张书毕，张克非，王楠迪，高雨，王梦瑶.2026.顾及空间邻域特征的MODIS PWV神经网络校正模型.遥感学报，30（1）： 144-155 DOI： 10.11834/jrs.20254201.

Liu Y，Zhang W Y，Li W Y，Zhang S B，Zhang K F，Wang N D，Gao Y and Wang M Y. 2026. MODIS PWV neural network correction model considering spatial neighborhood characteristics. National Remote Sensing Bulletin， 30（1）：144-155 DOI： 10.11834/jrs.20254201.

摘要

卫星遥感技术是大气水汽含量探测的主流手段之一，具有高空间分辨率的显著优势。然而，该技术探测精度较低，难以满足大气水汽变化研究需求。已有研究陆续利用高精度GNSS PWV“站点式”数据对“面式”遥感水汽数据进行适当性校正，获取精确的卫星遥感水汽产品。但现有研究多基于GNSS测站与遥感像素点的“点式”空间匹配数据构建校正模型，忽略了大气水汽局域强相关性的重要影响，导致其校正能力有限。鉴于此，本文以水汽空间邻域相关性为切入点，借助机器学习技术的非线性处理优势，构建顾及空间邻域特征的MODIS水汽产品神经网络校正模型。该模型以BP神经网络算法为框架，选定尺度范围内MODIS产品的云层信息、地表覆盖类型、传感器空间姿态等非线性影响要素作为模型输入参数。基于美国西部地区GNSS和MODIS PWV数据的实验结果表明：本文所提模型校正后的MODIS PWV均方根误差为2.13 mm，相较于广泛应用的线性校正模型，均方根误差降低了46.21%；与目前“点式”匹配模型的校正结果相比，均方根误差降低了12.35%。从时间和空间维度对比的结果表明：本文所提模型校正产品的均方根误差稳定于2.0—3.0 mm，论证了顾及空间邻域特征校正模型在校正遥感水汽产品方面的优越性，可反映出水汽分布的精细化时空信息。

Abstract

Precipitable Water Vapor (PWV) is a crucial indicator for characterizing precipitation potential and serves as a key parameter for predicting extreme weather events and climate change. Fine-scale spatiotemporal distribution information of PWV provides essential data support for scientific and effective analysis. Remote sensing water vapor retrieval technology with high spatial resolution and high-precision Global Navigation Satellite System (GNSS) water vapor detection technology have become the mainstream methods for detecting PWV. However

both technologies have certain limitations: MODIS PWV products are susceptible to nonlinear factors such as cloud type and land cover type

resulting in limited observational accuracy

and GNSS PWV products cannot effectively reflect the high spatial-resolution information on water vapor distribution due to the distribution of GNSS stations. Therefore

the high-precision GNSS PWV is used as a reference value to calibrate MODIS PWV

and it can effectively improve accuracy. However

existing research has paid little attention to the influence of local spatial correlation of remote sensing water vapor products

leading to limited correction capability. In summary

this paper proposes a new correction model based on the strong local spatial correlation of water vapor products

which can significantly enhance correction accuracy and subsequently reflect fine-scale spatiotemporal information on water vapor distribution.

In this paper

correlation analysis of remote sensing water vapor products within a local spatial domain is first conducted to determine an appropriate spatial neighborhood influence scale. Nonlinear influencing factors such as cloud information

land cover type

and sensor spatial orientation within this spatial domain from MOIDS products were taken into account

and a MODIS PWV correction model that considers the aforesaid spatial neighborhood characteristics is constructed. The proposed correction model is based on backpropagation (BP) neural network

with the difference between GNSS PWV and MODIS PWV matched data as the output variable.

Taking GNSS PWV as the reference value

the proposed correction model achieves a root mean square error (RMSE) of 2.13 mm. Compared with the uncorrected data (RMSE = 4.97 mm)

the correction results of the traditional linear model (RMSE = 3.96 mm)

and the point-based matching BP model (RMSE = 2.43 mm)

the quality of MODIS PWV is significantly improved

with enhancements of approximately 57.14%

46.21%

and 12.35%

respectively. In terms of spatial and temporal performance

the proposed model demonstrates stable correction effects. Across the entire spatial domain of the study area

the RMSE of most correction results from the proposed model is below 3 mm

and the RMSE of correction results in different months generally falls within the range of 2.0—2.5 mm

indicating good stability.Four hours is the time threshold

taking the high-precision radiosonde PWV as the reference value and considering the poor spatial resolution of the radiosonde PWV. The RMSE of the corrected MODIS PWV is 2.50 mm

representing a 53.79% improvement over the uncorrected data (RMSE = 5.41 mm). Furthermore

the RMSE of most corrected PWV lies within the range of -3 and 3 mm

which further demonstrates that the correction model can effectively improve the quality of MODIS water vapor products and meet requirements for practical applications.

First

the experiment demonstrates that considering the strong local spatial correlation of atmospheric water vapor can effectively improve the correction accuracy of MODIS PWV. Second

the correction results obtained using the proposed model exhibit good stability and can reflect fine-scale spatiotemporal information on water vapor distribution.

关键词

Keywords

references

Chang L , Xiao R Y , Prasad A A , Gao G P , Feng G P and Zhang Y . 2019 . Cloud mask-related differential linear adjustment model for MODIS infrared water vapor product . Remote Sensing of Environment , 221 : 650 - 664 [ DOI: 10.1016/j.rse.2018.12.005 http://dx.doi.org/10.1016/j.rse.2018.12.005 ]

Gurbuz G and Jin S G . 2017 . Long‐time variations of precipitable water vapour estimated from GPS, MODIS and radiosonde observations in Turkey . International Journal of Climatology , 37 ( 15 ): 5170 - 5180 [ DOI: 10.1002/joc.5153 http://dx.doi.org/10.1002/joc.5153 ]

Liang H , Cao Y C , Wan X M , Xu Z F , Wang H S and Hu H . 2015 . Meteorological applications of precipitable water vapor measurements retrieved by the national GNSS network of China . Geodesy and Geodynamics , 6 ( 2 ): 135 - 142 [ DOI: 10.1016/j.geog.2015.03.001 http://dx.doi.org/10.1016/j.geog.2015.03.001 ]

Liu B , Wang Y , Lou Z S and Zhan W . 2019 . The MODIS PWV correction based on CMONOC in Chinese mainland . Acta Geodaetica et Cartographica Sinica , 48 ( 10 ): 1207 - 1215

刘备 , 王勇 , 娄泽生 , 占伟 . 2019 . CMONOC观测约束下的中国大陆地区MODIS PWV校正 . 测绘学报 , 48 ( 10 ): 1207 - 1215 [ DOI: 10.11947/j.AGCS.2019.20180386 http://dx.doi.org/10.11947/j.AGCS.2019.20180386 ]

Long T F , Jiao W L , Zhang Z M and He G J . 2017 . Method for calculating observation zenith angle and azimuth angle of pixel of satellite image . CN , 104121884 B

龙腾飞 , 焦伟利 , 张兆明 , 何国金 . 2017 . 卫星影像像元观测天顶角和方位角的计算方法 . 中国 , 104121884 A

Ma S , Yue Y C , Shangguan M , Zou Y X and Ding M H . 2021 . A correction model for MODIS precipitable water vapor based on GNSS data . Journal of Nanjing University of Information Science and Technology (Natural Science Edition) , 13 ( 2 ): 154 - 160

马赛 , 岳迎春 , 上官明 , 邹玉学 , 丁茂华 . 2021 . 基于GNSS的MODIS大气可降水量校正模型 . 南京信息工程大学学报(自然科学版) , 13 ( 2 ): 154 - 160 [ DOI: 10.13878/j.cnki.jnuist.2021.02.004 http://dx.doi.org/10.13878/j.cnki.jnuist.2021.02.004 ]

Ma X W , Yao Y B , Zhang B , Qin Y X , Zhang Q and Zhu H . 2022 . An improved MODIS NIR PWV retrieval algorithm based on an artificial neural network considering the land-cover types . IEEE Transactions on Geoscience and Remote Sensing , 60 : 5622412 [ DOI: 10.1109/tgrs.2022.3170078 http://dx.doi.org/10.1109/tgrs.2022.3170078 ]

Ming D P , Wang Q and Yang J Y . 2008 . Spatial scale of remote sensing image and selection of optimal spatial resolution . Journal of Remote Sensing (in Chinese) , 12 ( 4 ): 529 - 537

明冬萍 , 王群 , 杨建宇 . 2008 . 遥感影像空间尺度特性与最佳空间分辨率选择 . 遥感学报 , 12 ( 4 ): 529 - 537 [ DOI: 10.11834/jrs.20080470 http://dx.doi.org/10.11834/jrs.20080470 ]

Shen H Y , Wang Z X , Gao C Y , Qin J , Yao F B and Xu W . 2008 . Determining the number of BP neural network hidden layer units . Journal of Tianjin University of Technology , 24 ( 5 ): 13 - 15

沈花玉 , 王兆霞 , 高成耀 , 秦娟 , 姚福彬 , 徐巍 . 2008 . BP神经网络隐含层单元数的确定 . 天津理工大学学报 , 24 ( 5 ): 13 - 15 [ DOI: 10.3969/j.issn.1673-095X.2008.05.004 http://dx.doi.org/10.3969/j.issn.1673-095X.2008.05.004 ]

Van Noord N and Postma E . 2017 . Learning scale-variant and scale-invariant features for deep image classification . Pattern Recognition , 61 : 583 - 592 [ DOI: 10.1016/j.patcog.2016.06.005 http://dx.doi.org/10.1016/j.patcog.2016.06.005 ]

Vaquero-Martínez J and Antón M . 2021 . Review on the role of GNSS meteorology in monitoring water vapor for atmospheric physics . Remote Sensing , 13 ( 12 ): 2287 [ DOI: 10.3390/rs13122287 http://dx.doi.org/10.3390/rs13122287 ]

Wan W , Chen X W , Peng X F , Bai W H , Xia J M , Liang H , Zhang X M , Xiong P , Yang T , Cao Y C , Yin C , Zhao L M and Hong Y . 2016 . Overview and outlook of GNSS remote sensing technology and applications . Journal of Remote Sensing (in Chinese) , 20 ( 5 ): 858 - 874

万玮 , 陈秀万 , 彭学峰 , 白伟华 , 夏俊明 , 梁宏 , 张学民 , 熊攀 , 杨婷 , 曹云昌 , 尹聪 , 赵利民 , 洪阳 . 2016 . GNSS遥感研究与应用进展和展望 . 遥感学报 , 20 ( 5 ): 858 - 874 [ DOI: 10.11834/jrs.20166228 http://dx.doi.org/10.11834/jrs.20166228 ]

Wang J , Li Q and Wei J H . 2021 . Study on the data correction of MODIS precipitable water vapor of Qinghai Province . Journal of Qinghai University , 39 ( 2 ): 77 - 84

王佳 , 李琼 , 魏加华 . 2021 . 青海MODIS可降水量数据校正研究 . 青海大学学报 , 39 ( 2 ): 77 - 84 [ DOI: 10.13901/j.cnki.qhwxxbzk.2021.02.011 http://dx.doi.org/10.13901/j.cnki.qhwxxbzk.2021.02.011 ]

Wang R B , Xu H Y , Li B and Feng Y . 2018 . Research on method of determining hidden layer nodes in BP neural network . Computer Technology and Development , 28 ( 4 ): 31 - 35

王嵘冰 , 徐红艳 , 李波 , 冯勇 . 2018 . BP神经网络隐含层节点数确定方法研究 . 计算机技术与发展 , 28 ( 4 ): 31 - 35 [ DOI: 10.3969/j.issn.1673-629X.2018.04.007 http://dx.doi.org/10.3969/j.issn.1673-629X.2018.04.007 ]

Xu J F and Liu Z Z . 2022 . A back propagation neural network-based algorithm for retrieving all-weather precipitable water vapor from MODIS NIR measurements . IEEE Transactions on Geoscience and Remote Sensing , 60 : 5633614 [ DOI: 10.1109/tgrs.2022.3219405 http://dx.doi.org/10.1109/tgrs.2022.3219405 ]

Yang J , Shi L , Wang Q W and He Q Q . 2020 . Moderate resolution imaging spectroradiometer near-infrared water vapor linear regression model based on the global positioning system data . Tropical Geography , 40 ( 1 ): 137 - 144

杨娇 , 史岚 , 王茜雯 , 何其全 . 2020 . 基于GPS数据的MODIS近红外水汽线性回归模型 . 热带地理 , 40 ( 1 ): 137 - 144 [ DOI: 10.13284/j.cnki.rddl.003207 http://dx.doi.org/10.13284/j.cnki.rddl.003207 ]

Yao Y B and Zhao Q Z . 2022 . Research progress and prospect of monitoring tropospheric water vapor by GNSS technique . Acta Geodaetica et Cartographica Sinica , 51 ( 6 ): 935 - 952

姚宜斌 , 赵庆志 . 2022 . GNSS对流层水汽监测研究进展与展望 . 测绘学报 , 51 ( 6 ): 935 - 952 [ DOI: 10.11947/j.AGCS.2022.20220039 http://dx.doi.org/10.11947/j.AGCS.2022.20220039 ]

Yuan P , Blewitt G , Kreemer C , Hammond W C , Argus D , Yin X G , Van Malderen R , Mayer M , Jiang W P , Awange J and Kutterer H . 2023 . An enhanced integrated water vapour dataset from more than 10 000 global ground-based GPS stations in 2020 . Earth System Science Data , 15 ( 2 ): 723 - 743 [ DOI: 10.5194/essd-15-723-2023 http://dx.doi.org/10.5194/essd-15-723-2023 ]

Zhang B and Yao Y B . 2021 . Precipitable water vapor fusion based on a generalized regression neural network . Journal of Geodesy , 95 ( 3 ): 36 [ DOI: 10.1007/s00190-021-01482-z http://dx.doi.org/10.1007/s00190-021-01482-z ]

Zhang K F , Li H B , Wang X M , Zhu D T , He Q M , Li L J , Hu A D , Zheng N S and Li H Z . 2022 . Recent progresses and future prospectives of ground-based GNSS water vapor soundina . Acta Geodaetica et Cartographica Sinica , 51 ( 7 ): 1172 - 1191

张克非 , 李浩博 , 王晓明 , 朱丹彤 , 何琦敏 , 李龙江 , 胡安东 , 郑南山 , 李怀展 . 2022 . 地基GNSS大气水汽探测遥感研究进展和展望 . 测绘学报 , 51 ( 7 ): 1172 - 1191 [ DOI: 10.11947/j.AGCS.2022.20220149 http://dx.doi.org/10.11947/j.AGCS.2022.20220149 ]

Zheng Z Q , Zhang K F , Shi J Q and Zhang M H . 2023 . Analysis of GNSS water vapor detection accuracy and temporal sequence characteristics in different climate types in China . Science of Surveying and Mapping , 48 ( 10 ): 68 - 77

郑志卿 , 张克非 , 师嘉奇 , 张明浩 . 2023 . 中国不同气候区域GNSS水汽探测精度及时序特征分析 . 测绘科学 , 48 ( 10 ): 68 - 77 [ DOI: 10.16251/j.cnki.1009-2307.2023.10.008 http://dx.doi.org/10.16251/j.cnki.1009-2307.2023.10.008 ]