A Comparative Study On The Reconstruction Methods Of Long Time Series Remotely Sensed Land Surface Temperature In Chongqing

Land surface temperature(LST)is a key element that effect the energy exchanges between land surface and air.Study on the spatial and temporal of LST variation is of great significance in climate change,urban heat island,drought monitor,ecological environment,and etc.Remote sensing is an effective tool to provide LST observations at regional and global scale.Remote sensors with infrared channels have been playing an important role in estimating LST.However,it is difficult to derive the LST of cloudy pixels directly from remotely sensed thermal infrared information,because under clouds LST cannot be measured due to absorption of the surface emission by the clouds.Being remotely sensed data in the thermal range,LST shows gaps in cloud-covered areas.When deriving climatic parameters from remotely sensed time series,one should ideally work with complete data sets.Typically,this requires reconstruction of the void map areas.The spatial interpolation method was used in study on LST time series reconstruction,and there are few researches for the reconstruction in rugged terrain.The most commonly used LST products are those from the Moderate resolution Imaging Spectroradiometer(MODIS)instruments on board of Terra and Aqua NASA satellites.This paper takes Chongqing as study area and focuses on the reconstruction of MODIS LST time series.Validation of the reconstructed time series can be done by ground stations.In this paper,we use four LST reconstruction method: Harmonic Analysis of Time Series(HANTS),the Savitzky-Golay filtering method(S-G),Asymmetric Gaussians(A-G)and Double Logistic Function Fitting(D-L)to reconstruct MODIS LST data during 2010-2017.Moreover,we analysis the MODIS LST data quality and compares the qualitative and quantitative result based on the four reconstruction methods.This paper draws some conclusions as followed:(1)Quality evaluation of MODIS LST data in Chongqing during 2010-2017.According to the time series data analysis,there is the lowest cloud cover in 2013 andthe highest in 2012.From 2010 to 2017,the overall trend of MODIS LST data low-quality frequency was low-high-low.In 2013,the low-quality frequency was the lowest,and the LST data quality was the best.In 2012,the low-quality frequency was the highest,and the data quality was the worst.According to the spatial analysis,cloud cover in Chongqing distribution: Southeast > East and West > Northeast.The spatial distribution of MODIS LST data quality from 2010 to 2017 was significantly different,and the high-quality data distribution: Northeast > Southeast > East and West.Northeast has the best data quality,mainly located in Chengkou,Wuxi and Wushan,the next is in the southeast and middle of Chongqing,the worst data is in the western region of Chongqing,mainly distributed in Jiangjin,Qijiang,Rongchang,Dazu,and central districts of Chongqing.Low quality frequency of MODIS LST data is basically greater than 50%,which should be the focus of time series reconstruction.(2)The four reconstruction methods are applied in Chongqing.(1)HANTS has the best reconstruction result,which can effectively eliminate large noise areas and assign values to null values.At the same time,its reconstruction curve is smooth,and the correlation between before and after MODIS LST data reconstruction is the strongest and the most stable.(2)S-G reconstruction result is inferior to HANTS,according to time analysis,its correlation is slightly lower than HANTS,the spatial correlation is at a moderate level,and S-G is sensitive to identify high or mutation value,before and after reconstruction images' details are relatively consistent;(3)The reconstruction results of A-G and D-L are relatively similar,and the result are slightly worse in the four LST reconstruction results.After the reconstruction of D-L,there are still a small number of null values.The correlation of A-G is higher than D-L,and the A-G reconstruction result is slightly better than D-L.(3)In the same year,the spatial distribution difference of LST(MOD11A2 and MYD11A2 day/night)was not significant.Based on quality comparison,it is found that the quality of MOD11A2 is better than MYD11A2,and the quality of night data is obviously better than day data,especially the low-quality frequency of night MOD11A2 is far less than the other three groups,and the fidelity after reconstruction is higher Therefore,in the utilization of LST,MOD11A2 night data can be more used for reference.Although this study uses a variety of reconstruction methods,there are still some shortages.In the process of reconstruction,the study just uses a single temporal reconstruction method,and the shortcomings of the reconstruction methods are notimproved.In addition,the parameters of the four LST data reconstruction methods are not objective,and conducting the qualitative and quantitative evaluation,other meteorological factors are not effectively combined to provide more results.Therefore,in the later research on LST reconstruction,multiple reconstruction methods will be adopted,the parameters are set by trial and error for different surface types and areas,and more quantitative indicators will be used in the evaluation of reconstruction methods,which will improve the scientific and fidelity of research.