地球信息科学学报 ›› 2022, Vol. 24 ›› Issue (2): 378-390.doi: 10.12082/dqxxkx.2022.210257

• 遥感科学与应用技术 • 上一篇    下一篇

基于DCT-PLS算法的MODIS LST缺值填补方法研究

刘恒孜1,2(), 吕宁1,3,*(), 姜侯1, 姚凌1,3   

  1. 1.中国科学院地理科学与资源研究所 资源与环境信息系统国家重点实验室,北京 100101
    2.中国科学院大学,北京 100049
    3.南方海洋科学与工程广东省实验室(广州),广州 511458
  • 收稿日期:2021-05-10 修回日期:2021-07-05 出版日期:2022-02-25 发布日期:2022-04-25
  • 通讯作者: *吕 宁(1981—),男,河南洛阳人,副研究员,硕士生导师,主要从事大气遥感应用研究。E-mail: lvn@lreis.ac.cn
  • 作者简介:刘恒孜(1997— ),男,河南许昌人,硕士生,主要从事遥感地学分析与计算研究。E-mail: liuhengzi18@mails.ucas.ac.cn
  • 基金资助:
    国家自然科学基金项目(41971312);南方海洋科学与工程广东省实验室(广州)人才团队引进重大专项(GML2019ZD0301)

Research on Gaps Filling of MODIS LST based on DCT-PLS

LIU Hengzi1,2(), Lü Ning1,3,*(), JIANG Hou1, YAO Ling1,3   

  1. 1. State Key Laboratory of Resources and Environmental Information Systems, Institute of Geographic Sciences and Natural Resources Research, Chinese Academy of Sciences, Beijing 100101, China
    2. University of Chinese Academy of Sciences, Beijing 100049, China
    3. Southern Marine Science and Engineering Guangdong Laboratory, Guangzhou 511458, China
  • Received:2021-05-10 Revised:2021-07-05 Online:2022-02-25 Published:2022-04-25
  • Supported by:
    National Natural Science Foundation of China(41971312);Key Special Project for Introduced Talents Team of Southern Marine Science and Engineering Guangdong Laboratory (Guangzhou)(GML2019ZD0301)

摘要:

受云层、传感器误差等因素影响,中分辨率成像光谱仪(MODIS)获取的地表温度产品(LST)在时间和空间上存在大量缺失数据,严重影响了基于时间序列数据的分析与应用。本文引进一种基于离散余弦变换与惩罚最小二乘的多维数据快速平滑方法(DCT-PLS),利用数据集自身的时间和空间信息填补缺值。本文在粤港澳大湾区开展实证研究,将DCT-PLS算法用于填补该地区2001年1月—2017年12月的月值MODIS LST数据缺值,并引入人工模拟缺值对算法进行误差分析与精度验证。算法误差分析结果表明,填补误差主要来源于三维算法对数据集中有偏LST时间信息的使用,并因此产生显著高估或低估的填补结果。基于此,本文提出了利用MODIS LST数据集自身时间序列信息自动计算获取有效辅助LST信息的优化策略,从而实现填补算法计算效率和精度的提升:平均计算时间从12.0 s提高至1.7 s,平均R从0.94提高至0.97,平均RMSE从1.94 K提高至0.74 K(相较于三维算法)。在大湾区的填补结果表明(日间结果:R=0.98、RMSE=0.79 K;夜间结果:R=0.99、RMSE=0.56 K),优化后的DCT-PLS算法可以快速鲁棒地填补MODIS LST月值数据产品中的缺值,并且具有稳定性强、不依赖外部数据集的计算特性,能够适应长时间序列MODIS LST缺值填补。

关键词: MODIS, 缺值填补, 地表温度, 长时序, 离散余弦变换, 惩罚最小二乘, 算法优化, 粤港澳大湾区

Abstract:

Affected by cloud layer, sensor error, and other factors, there are large area of data gaps in time and space of Land Surface Temperature (LST) products acquired by Moderate-resolution Imaging Spectroradiometer (MODIS), which seriously affects the analysis and application of LST data set. In this paper, a fully automatic smoothing algorithm for filling MODIS LST data sets is introduced. Based on a Penalized Least Squares (PLS) method, the algorithm allows fast smoothing of data in one and more dimensions by means of the Discrete Cosine Transform (DCT). By optimizing smoothing parameters, the DCT-PLS algorithm is a fast multidimensional data smoothing method, which uses the spatial and temporal information of LST data set to fill data gaps. This study carries out an empirical study in Guangdong-Hong Kong-Macao Greater Bay Area (GBA). The DCT-PLS algorithm is used to fill data gaps of monthly MODIS LST data set from January 2001 to December 2017 in this area. To analyze its performance, the error analysis and accuracy verification of the algorithm are performed by introducing artificial simulation data gaps. The results of error analysis show that errors in gap filling are mainly due to the use of biased LST time information in data set, which causes significantly overestimation or underestimation in filling results. In response to this problem, an optimized DCT-PLS algorithm is proposed to automatically obtain effective auxiliary LST layers from the time series of MODIS LST and provide unbiased temporal information. Meanwhile, the three-dimensional algorithm is changed to two-dimensional space to reduce computing consumption by transforming auxiliary temporal information into spatial information. The results of accuracy verification on the optimized DCT-PLS algorithm show that the computational efficiency and accuracy are optimized simultaneously. The average computing time decreases from 12.0s to 1.7s, average R increases from 0.94 to 0.97, and average RMSE decreases from 1.94 K to 0.74 K (compared with the three-dimensional algorithm). The optimized DCT-PLS algorithm is applied in GBA. The accuracy of gap filling of monthly MODIS LST data set from January 2001 to December 2017 is R=0.98 and RMSE=0.79 K in daytime, and R=0.99 and RMSE=0.56 K in nighttime. The results of accuracy verification in GBA show that the accuracy of gap filling is not affected by heterogeneous surface. The optimized DCT-PLS algorithm transforms the MODIS LST data in the time domain and frequency domain to two-dimensional space, and retains high-frequency information to fill data gaps. It is a fast and robust method for gap filling, which is completely independent of external auxiliary data sets, enabling gap filling on long-time sequence MODIS LST data set.

Key words: MODIS, gap filling, Land Surface Temperature (LST), long time sequence, discrete cosine transform, penalized least-squares, algorithm optimization, Guangdong-Hong Kong-Macao Greater Bay Area (GBA)