The Construction Of Medium Spatial Resolution Dense Time-series Remote Sensing Data

Long time series remote sensing data is applied widely,which plays an important role in the terrestrial ecosystem dynamic monitoring and climate process simulation.At present,the main time series remote sensing data products are AVHRR,VEGETATION and MODIS,etc.,with high temporal resolution,but low spatial resolution,which cannot meet the needs of higher spatial resolution remote sensing applications.However,for medium spatial resolution sensors,such as TM and ETM+,the temporal resolution are low,it can't meet the need of real-time monitoring.Therefore,the construction of medium spatial resolution remote sensing data with high temporal resolution is of great significance for realizing rapid change monitoring and improving monitoring frequency.This paper uses the 2013-2016 all available Landsat8 OLI images and Landsat7 ETM+images as supplementary data,and bases on time series surface reflectance prediction model,to research on the construction of dense time series remote sensing data with medium spatial resolution.The main research contents and results are summarized as follows:(1)Clouds and cloud shadows detection in medium resolution time series images with cloud cover less than 90%.For the images with small cloud cover,the band 9 of Landsat8 can be used to detect.While Fmask algorithm is used for images with more cloud cover,but this algorithm can't detect thin cloud and shadow,it will detect the construction land and water mistakenly.Finally,the model which proposed by Zhe Zhu is used to detect the rest of the clouds and shadow that Fmask algorithm cannot detect.(2)Making comparison and evaluating medium resolution images which are fitted by different surface reflectance prediction models.Adaptive model and basic model,that based on Landsat8 time-series data only,can only fit basic vegetation season change and the fitting results of the two models are similar.Hence,taking Landsat7 data as supplementary data and making radiometric normalization for Landsat7 data can improve the fitting precision.With increasing amount of data,the basic model is still only fitting the seasonal change,while the adaptive model can better fit the details and intra-annual multimodal variations.The fitting precision of vegetation area is the highest,RMSE of most bands is about 0.01.However,in water area and construction land,the error that applying adaptive model will be increased and reduce the accuracy.(3)The influence of radiometric normalization on fitting medium resolution images.The fitting precision will be highest when it uses Landsat8 data and radiation normalized Landsat7 data,and applys the adaptive model.According to class distribution of the study area,the fitting accuracy of the image can be improved by choosing the appropriate radiation normalization method.The main class of the study area is vegetation,which is more suitable for normalization of vegetation area.(4)Constructing the 2013-2016 dense time-series with 16 days interval.The construction results are evaluated from four aspects:the comparison between the predicted images and the real images,analysis of temporal variation of the time series data,the temporal and spatial trend of the typical features,and the effect of data sets construction on change monitoring.The results show that the predicted images is very close to the real images;The constructed data sets can reflect the seasonal and annual variation characteristics of high vegetation coverage areas in Fujian Province;The variation trend of the three seasons in the summer,autumn and winter is basically consistent with the measured spectrum;The time-series data can achieve rapid change monitoring and improve the frequency of change monitoring,which showed fine construction results.