地球信息科学学报 ›› 2021, Vol. 23 ›› Issue (4): 749-761.doi: 10.12082/dqxxkx.2021.200149

• 遥感科学与应用技术 • 上一篇    下一篇

基于广义回归神经网络的京津冀地区土壤湿度遥感逐日估算研究

邓雅文1(), 凌子燕2,*(), 孙娜1, 吕金霞1,3   

  1. 1.环境遥感与数字城市北京市重点实验室,北京师范大学地理科学学部,北京 100875
    2.北部湾环境演变与资源利用教育部重点实验室,南宁师范大学地理科学与规划学院,南宁 530001
    3.遥感科学国家重点实验室,北京师范大学地理科学学部,北京 100875
  • 收稿日期:2020-03-27 修回日期:2020-06-21 出版日期:2021-04-25 发布日期:2021-06-25
  • 通讯作者: 凌子燕
  • 作者简介:邓雅文(1999— ),女,湖南衡阳人,硕士生,研究方向为生态水文遥感。E-mail: 202021051199@mail.bnu.edu.cn
  • 基金资助:
    国家自然科学基金项目(41571077);国家重点研发计划项目(2016YFC0503002);环境遥感与数字城市北京 重点实验室开放课题(12800-310430001)

Daily Estimation of Soil Moisture over Beijing-Tianjin-Hebei Region based on General Regression Neural Network Model

DENG Yawen1(), LING Ziyan2,*(), SUN Na1, LV Jinxia1,3   

  1. 1. Beijing Key Laboratory for Remote Sensing of Environment and Digital Cities, Faculty of Geographical Science, Beijing Normal University, Beijing 100875, China
    2. Beibu Gulf Key Laboratory of Environment Change and Resources Use, School of Geography and Planning, Nanning Normal University, Nanning 530001, China
    3. State Key Laboratory of Remote Sensing Science, Faculty of Geographical Science, Beijing Normal University, Beijing 100875, China
  • Received:2020-03-27 Revised:2020-06-21 Online:2021-04-25 Published:2021-06-25
  • Contact: LING Ziyan
  • Supported by:
    National Natural Science Foundation of China(41571077);National Key R&D Program of China(2016YFC0503002);Open subject of Beijing Key Laboratory for Remote Sensing of Environment and Digital Cities(12800-310430001)

摘要:

土壤湿度是地表水热交换过程和水文循环中的一个关键组成部分,获取高时空分辨率的土壤湿度数据一直是当前研究的热点。SMAP(Soil Moisture Passive and Active)主被动微波土壤湿度产品的精度高,但存在着空间分辨率低和时间分辨率缺失的问题,这限制了其在区域尺度上的应用,为解决这一问题得到更高时空分辨率的土壤湿度产品,本文利用广义回归神经网络模型(GRNN)模拟了MODIS地表温度、反射率、植被指数光学/热红外遥感数据以及高程、坡度、坡向、经纬度数据与SMAP土壤湿度的关系,从而将京津冀地区SMAP L2土壤湿度产品的时间分辨率由不连续(4~20 d)提升至1 d,空间分辨率由3 km提升至1 km,并扩展其在京津冀地区的空间覆盖范围。研究发现:① GRNN模型总体验证结果表明土壤湿度估算值与SMAP原始值的相关性较高(r=0.7392),均方根误差(RMSE)为0.0757 cm3/cm3;② 不同季节典型日期的GRNN模型估算结果精度相差较大,春季处的相关性相比其他季节最低,精度相对较高(r=0.6152,RMSE=0.0653cm3/cm3),秋季和夏季土壤湿度估算精度较为接近(r=0.6957,r=0.7053,RMSE=0.0754cm3/cm3,RMSE=0.0694cm3/cm3),冬季的估算精度最高(r=0.8214,RMSE=0.0367cm3/cm3);③ 2016年京津冀夏秋季节的土壤湿度较其他季节要显著提高,空间分布上坝上高原区域较低,而沿海地区的土壤湿度明显较高。本研究对京津冀地区的生态水文、气候预测以及干旱监测等应用领域具有重要价值。

关键词: 遥感, SMAP, 土壤湿度, 反演算法, 降尺度, 广义回归神经网络模型(GRNN), 京津冀, 机器学习

Abstract:

Surface Soil Moisture (SM) plays an important role in the land-atmosphere interaction and hydrological cycle. Low spatiotemporal resolution (i.e., 25~40 km and 2~3 days) microwave-based SM products such as the Soil Moisture and Ocean Salinity (SMOS) and the Advanced Microwave Scanning Radiometer for EOS (AMSR-E) limit their application in regional scale studies. The Soil Moisture Active Passive (SMAP) and Copernicus Sentinel 1A/B microwave active-passive surface soil moisture product (L2_SM_SP) has a higher spatial resolution (3 km), but its temporal resolution is coarse from 4 to 20 days due to the narrow overlapped swath width. In this study, we developed a machine learning algorithm using the General Regression Neural Network (GRNN) to improve the spatiotemporal resolution of the L2_SM_SP product based on multi-source remote sensing data. Land Surface Temperature (LST), Multi-band Reflectance, Normalized Difference Vegetation Index (NDVI), Enhanced Vegetation Index (EVI), Elevation, Slope, Longitude (Lon), and Latitude (Lat) were selected as input variables to simulate the L2_SM_SP soil moisture in GRNN model. Results show that: (1) GRNN-estimated soil moisture and the original estimates of L2_SM_SP were strongly correlated (r=0.7392, RMSE=0.0757 cm3/cm3); (2) the correlation between GRNN estimates and original L2_SM_SP product at typical dates of different seasons varied a lot. The correlation in spring was the lowest (rSpr=0.6152, RMSESpr=0.0653 cm3/cm3). While the correlation in winter was the strongest (rWin=0.8214, and RMSEWin=0.0367 cm3/cm3). The correlation in summer and autumn was close to each other (rSum=0.6957, rAut=0.7053, RMSESum=0.0754 cm3/cm3, and RMSEAut=0.0694 cm3/cm3); and (3) in 2016, the soil moisture in summer and autumn of the study area was significantly higher than that that in other seasons. In terms of spatial distribution, the soil moisture in the Bashang plateau area was low, while the soil moisture along coastal areas was obviously higher. In this study, we successfully improved the spatiotemporal resolution of L2_SM_SP product over Beijing-Tianjin-Hebei region from 3 km, and 4~20 days to 1 km, and 1 day. Its spatial coverage was also extended. The improved soil moisture product is of great significance for future eco-hydrological assessment, climate prediction, and drought monitoring in Beijing-Tianjin-Hebei region.

Key words: remote sensing, Soil Moisture Active Passive(SMAP), soil moisture, retrieval algorithms, downscaling, GRNN, Beijing-Tianjin-Hebei Region, machine learning