| As the continuously innovation of satellite and sensor technology,more and more remotely sensed observations have been archived.As the important input,the remotely sensed time series data has been widly used in regional and global environmental change research.With the deep development of the research,there is an increasing demand of the spatio-temporal continuity and intergrity of remotely sensed time series data.However,the remotely sensed time series data are inevitably influenced by the observation conditations and the instrument failure during the data acquisition process causing a lot of information missing,which severely hiders the further applicaton of these data.Under the support of variational framework,the time series reconstruction algorithms are proposed according to the characteristics of remote sensing time series data in order to reconstruct high quality remote sensing time series data.This paper will propose new temporal filter and temporal reconstruction methods to overcome shortages of the existing methods focusing on improving the spatio-temporal continuity and intergrity of the remotely sensed time series data.The traditional time series reconstruction methods are generally based on the ideas of filtering or curve fitting,which do not consider enough the a priori characteristics of NDVI time series data and are difficult to effectively solve the difficult problem of missing time continuity.To this end,this paper proposes a new NDVI time series data reconstruction method based on the idea of one-dimensional variational filtering.The method innovatively introduces the information of adjacent years and combines the temporal neighborhood information to constrain the local smoothness prior and interannual similarity prior of NDVI time series data by regularization.Meanwhile,the method uses L1 parametric inscription of two regularization terms to better characterize the statistical distribution of NDVI time series data.Simulated and real experiments are conducted using Moderate Resolution Imaging Spectroradiometer(MODIS)NDVI data in the Yangtze River Economic Zone region of China and compared with five classical time-series filtering reconstruction methods.The results show that the proposed method can achieve satisfactory performance in terms of both quantitative metrics and spatio-temporal visual effects with acceptable time cost.In particular,the methods show clear advantages in recovering missing values of temporal continuum and preventing over-smoothing at vegetation growth inflection points.In tropical subtropical cloudy and rainy regions,there are a large number of missing in NDVI time-series data,and existing reconstruction methods are difficult to solve this challenging challenge.In this paper,an adaptive spatio-temporal tensor complementation algorithm is proposed to reconstruct long time series NDVI data in cloudy-rainy regions by using multidimensional spatio-temporal information.A highly correlated three-dimensional tensor is constructed by considering the similarity in spatial neighborhood,temporal neighborhood,and temporal period of NDVI data,further developing an adaptive low-rank tensor complementation model to reconstruct the missing information,and finally removing the temporal noise by iterative L1 filtering.In this chapter,the method is tested using Moderate Resolution Imaging Spectroradiometer(MODIS)NDVI data with mainland Southeast Asia,which is located in the tropical subtropics,as the study area.The quantitative and qualitative results show that the present method has obvious advantages over five comparative methods in reconstructing large scale spatio-temporal continuous missing information,and thus provides an effective solution for the reconstruction of high quality NDVI time series data in tropical subtropical cloudy rainfall regions.The current research on time series remote sensing data reconstruction methods mainly focuses on vegetation index data such as NDVI,while as a basic class of remote sensing data,time series surface reflectance data reconstruction still has great challenges.To solve this problem,this chapter takes into account the general time series remote sensing data a priori characteristics,i.e.,the sparse a priori of inaccurate quality marker data,the transformation tensor low-rank a priori and the time series smoothing a priori,and constructs a new robust tensor complementary model with temporal smoothness constraint for reconstructing general time series remote sensing data including surface reflectance,NDVI,etc.The quantitative and qualitative results show that the The quantitative and qualitative results show that this method has significant advantages in reconstructing the fluctuating,missing and noisy surface reflectance and the derived NDVI data,and the method does not require high accuracy of the quality marker data,and can obtain good reconstruction results even if there are cloud image element misses. |
