Reconstruction Of MODIS Time Series And Its Applications

Satellite remote sensing is the only viable option for large scale and long term monitoring of land surface changes.However,the capability of satellite time series for change analysis is hindered by missing data and noise.In this study,an algorithm for reconstructing time series was developed,and was validated with MODIS Leaf Area Index(LAI)product.Then,the spatio-temporal patterns of changes in Songnen grasslands were characterized using MODIS time series.The major contributions and findings are as follows:(1)An Enhanced Ecosystem-Dependent Temporal Interpolation(EEDTI)algorithm was developed to address the uncertainties in filling gaps of LAI time series for heterogeneous landscapes.EEDTI significantly increased the interpolation accuracy by improving the quantification of phenological links between pixels,the selection of reference time series and the constraints of interpolation.EEDTI showed an advantage in filling large data gaps in heterogeneous grasslands.The R~2 was about 0.9 and the Normalized Root Mean Square Error(NRMSE)about 0.2 even with 60-70%missing data.(2)A Spatio-Temporal Constraint Interpolation(STCI)and the Stationary Wavelet Transform(SWT)were coupled to reconstruct LAI time series.The STCI integrates local temporal variations,spatial variations and climatology curves to improve the use of spatial-temporal information for gap filling.SWT was introduced in the iterative interpolation progress to reduce error accumulation and smooth the time series.The coupled algorithm is expected to be more suitble for different land surface scenarios than the EEDTI.(3)Spatio-temporal dynamics of land surface phenology in Songnen grasslands during 2000-2015 were characterized.The spatial patterns of the timing of green up onset,timing of dormancy onset,and growing season length were dominated by grasslands salinization.In recent years,the timing of growing season maximum was significantly advanced due to increasing precipitation.On the other hand,knowledge on the responses of autumn phenology to climate variability in Songnen grasslands was improved by dividing the growing season into growing and declining periods.More spring and summer precipitation generally advanced the timing of growing season maximum,and subsequently led to earlier onset of dormancy.(4)Trends in Songnen grasslands during 2000-2017 were characterized by the Enhanced Vegetation Index(EVI)and the Normalized Difference Water Imdex(NDWI)time series.About 56%of Songnen grasslands showed significantly increase in EVI under increasing precipitation and reduced human activities.Diverse vegetation recovery modes were found based on the different responses of EVI and NDWI to land surface changes.Only about 2%of Songnen grasslands showed significantly negative EVI trend.Areas with not significantly EVI trend were mainly distributed in Zhenlai county,Jilin province.Under human management with spatio-temporal uncertainties,EVI trends still had spatial dependence on local topography.The low-lying terrain generally had a not significant trend or significantly negative trend.This may be related to the natural salt deposition process and the negative effects of human management.